Security glossary
Plain-language definitions with the distinctions that matter in practice. Each term has its own page with the guides that cover it.
271 terms
A
ABAC
Attribute-based access control (NIST SP 800-162). Each request is decided by a policy over attributes of the subject, the resource, the action and the environment, such as department, data classification or time. It expresses fine-grained rules that roles cannot, but depends on trustworthy attribute sources and on policies tested like code.
Abuse case
A scenario written from an attacker’s or misuser’s point of view that describes how a feature could be abused, such as replaying a coupon or enumerating accounts, so the team can derive security requirements and tests from it. Also called a misuse case.
Access token
A credential that an OAuth client presents to a resource server to call an API (RFC 6749). It is meant for the API named as its audience, not for telling the client who the user is, and should be short-lived; most access tokens are bearer tokens unless they are sender-constrained with DPoP or mTLS.
ACME
Automatic Certificate Management Environment (RFC 8555). A protocol with which a client proves control of a domain, usually through an HTTP-01 or DNS-01 challenge, and obtains and renews certificates without manual steps, as with Let’s Encrypt. Automation matters more as the maximum lifetime of publicly trusted TLS certificates falls toward 47 days in 2029.
AddressSanitizer
A compiler instrumentation in Clang and GCC that detects out-of-bounds accesses, use-after-free, double free and other memory errors at run time, with about a twofold slowdown. Run tests and fuzzers with it enabled; it is a testing tool, not a production defense.
Admission controller
A Kubernetes component that intercepts API server requests after authentication and authorization and can reject or modify objects before they are stored. Teams use it to enforce Pod Security Standards, ValidatingAdmissionPolicy rules written in CEL, and image signature or provenance checks.
AEAD
Authenticated encryption with associated data. It encrypts the plaintext for confidentiality and authenticates both the ciphertext and associated data sent in the clear, so decryption fails if either was changed (RFC 5116).
AES-GCM
AES in Galois/Counter Mode (NIST SP 800-38D). It is the most widely used AEAD, and every message under one key needs a unique 96-bit nonce: a repeated nonce exposes the XOR of the two plaintexts and lets an attacker forge tags, and with random nonces one key should encrypt at most 2^32 messages.
Agent goal hijack
An attack that redirects an AI agent’s objectives, task selection or plan through injected instructions, deceptive tool output, forged messages from other agents or poisoned data, so the agent pursues the attacker’s goal with its own permissions across many steps (OWASP ASI01:2026). Record the user’s goal outside the model and authorize each action against it.
AI agent
An application in which a language model plans and takes actions through tools, such as calling APIs, running code or browsing, over several steps with little human review. Its risk comes from the authority its tools carry, which the OWASP Top 10 for Agentic Applications (2026) catalogs.
AI RMF
NIST AI Risk Management Framework (NIST AI 100-1, January 2023). A voluntary framework for managing the risks of AI systems through four functions: Govern, Map, Measure and Manage. NIST AI 600-1 (July 2024) is its profile for generative AI.
Allowlist
A list of explicitly permitted values, patterns, destinations or types; anything not on it is rejected. It fails closed when attackers find a new variant, which a denylist of known-bad inputs does not.
API
Application programming interface. A defined boundary through which another component requests data or behavior; authorization must be enforced at that boundary.
Argon2
A memory-hard password hashing function and the winner of the Password Hashing Competition (RFC 9106). Use the Argon2id variant; OWASP’s minimum is 19 MiB of memory, 2 iterations and 1 degree of parallelism, with a unique salt for each password.
Argument injection
Untrusted input is passed to a program as an argument that the program reads as an option, such as a value beginning --output= or --upload-pack=, even when no shell is involved (CWE-88). Put user values after the -- end-of-options marker (--end-of-options in Git) and reject values that start with a dash.
ASPM
Application security posture management. A tool category that collects findings from SAST, SCA, secrets, IaC and DAST scanners, removes duplicates and ranks them by owner and risk; it detects nothing its scanners miss.
ASVS
OWASP Application Security Verification Standard. Version 5.0.0 (May 2025) lists 345 testable requirements in 17 chapters across three levels; cite each with its version, as in v5.0.0-1.2.5.
Attack surface
Every point where an attacker can try to send data into a system, get data out of it or act on it: endpoints, inputs, files, interfaces, dependencies, and the people and services with access. Reducing it removes paths instead of defending them.
Attack tree
A diagram with an attacker’s goal at the root and the ways to reach it as branches joined by AND and OR nodes, described by Bruce Schneier in 1999. It helps compare attack paths by cost or likelihood and shows which control cuts the most branches.
Attestation
Signed evidence about a subject that a verifier can check. In the software supply chain it is a signed statement bound to an artifact’s digest, such as SLSA provenance or a test result, usually in the in-toto format; device attestation shows that a key lives in genuine hardware, and a supplier’s attestation form is a signed self-declaration.
Audit log
An append-only record of security-relevant actions: who did what, to which resource, when, from where and with what result. It must be protected from changes by the people and services it records, kept long enough to investigate incidents, and kept free of secrets.
Authentication
Establishing a subject’s identity or the authenticity of a claim. It does not by itself establish permission to use a particular resource.
Authorization
Deciding whether a specific subject may perform a particular action on a resource in the current context.
B
bcrypt
A password hashing function based on the Blowfish cipher, with a work factor that sets its cost (Provos and Mazières, 1999). Most implementations use only the first 72 bytes of the input; OWASP recommends it only where Argon2id and scrypt are unavailable, with a work factor of at least 10.
Bearer token
A token that grants access to whoever presents it, without proof of possession of a key (RFC 6750). A stolen bearer token works until it expires, so keep it short-lived and out of URLs and logs, and use sender-constrained tokens (DPoP or mTLS) for high-value APIs.
BFLA
Broken function level authorization, OWASP API5:2023. An API lets a caller invoke an operation that their role should not reach, such as an administrative endpoint or another HTTP method on the same resource, because the check is missing or enforced only in the user interface.
BOLA
Broken object level authorization, OWASP API1:2023. An API accepts an object identifier without checking that the caller may perform the requested action on that object.
BSIMM
Building Security In Maturity Model. A descriptive study of the software security activities that real organizations perform, grouped into 12 practices in four domains and updated about once a year; BSIMM16 (2026) draws on 111 organizations. It describes what others do rather than prescribing what you should do.
Buffer overflow
Writing past the end, or before the start, of a buffer, which corrupts adjacent memory (CWE-120; out-of-bounds writes are CWE-787, with CWE-121 and CWE-122 for stack and heap buffers). In C and C++ it can overwrite return addresses, function pointers or allocator metadata; bounds-checked APIs, sanitizers in testing and memory-safe languages prevent it.
Bug bounty
A program that pays external researchers for valid vulnerability reports within a published scope and rules. It adds rewards to a vulnerability disclosure policy rather than replacing it, and it only helps when the team can triage and fix what arrives.
Business logic flaw
A weakness in an application’s rules rather than in how it parses input, such as skipping a workflow step, reusing a one-time coupon or ordering a negative quantity, so valid-looking requests reach an outcome the business never intended (CWE-840).
C
Cache poisoning
Getting a shared cache to store a harmful entry under a key that others later read (CAPEC-141): a web cache that serves a malicious response to other users, usually because an unkeyed input such as a header changed it, or a CI cache that later builds restore. Key every input that affects an entry, and never let untrusted jobs write to a cache that trusted ones read.
Canonicalization
Converting input to one standard form before checking it, for example decoding percent-encoding, normalizing Unicode and resolving ../ segments in a path. Validating before canonicalizing lets an encoded payload pass the check and take effect later (CWE-180).
CAPEC
Common Attack Pattern Enumeration and Classification. MITRE’s catalog of attack patterns, such as CAPEC-66 SQL Injection, which describe how attackers exploit weaknesses; it complements CWE, which catalogs the weaknesses themselves.
CBC
Cipher block chaining (NIST SP 800-38A). A block cipher mode in which each plaintext block is XORed with the previous ciphertext block before encryption, starting from an unpredictable IV. It provides no integrity, so it needs a MAC over the IV and ciphertext checked before decryption, or padding oracles follow; prefer an AEAD.
Certificate
An X.509 certificate: a data structure signed by a certificate authority that binds a public key to a subject name and other attributes for a validity period (RFC 5280). A verifier checks the chain to a trusted root, the name, the dates, the key usage and the revocation status before trusting the key.
Certificate authority
An organization or service trusted to issue, sign and revoke certificates after validating the requester’s control of a name or identity. Clients trust it through a root certificate in their trust store, so any trusted CA can issue for any name unless name constraints, CAA records or Certificate Transparency monitoring limit or expose it.
Certificate Transparency
A system of public, append-only logs of issued certificates (RFC 6962), so domain owners can detect certificates issued for their names without their knowledge. Chrome and Safari require signed certificate timestamps from these logs for publicly trusted TLS certificates.
ChaCha20-Poly1305
An AEAD that combines the ChaCha20 stream cipher with the Poly1305 authenticator (RFC 8439). It is fast in software on hardware without AES instructions and, like AES-GCM, must never reuse a 96-bit nonce under one key; XChaCha20-Poly1305 extends the nonce to 192 bits so that random nonces are safe.
CIEM
Cloud infrastructure entitlement management. Tools that read cloud IAM policies, role trust and activity logs to find unused or excessive permissions and privilege escalation paths, so each identity can be cut back to what it uses.
Clickjacking
A page loads the target site in a hidden or disguised frame and tricks the user into clicking it (CWE-1021). Prevent framing with the CSP frame-ancestors directive or X-Frame-Options.
CNA
CVE Numbering Authority. An organization authorized by the CVE Program to assign CVE IDs and publish CVE records within an agreed scope, such as a vendor for its own products or GitHub for open source projects that request one.
CNAPP
Cloud-native application protection platform. A product that combines cloud posture management (CSPM), workload protection (CWPP) and often entitlement, data and code findings, and correlates them into attack paths; it inherits each component’s blind spots.
Code injection
Untrusted input becomes code that the application evaluates or compiles, as with eval, exec, dynamic imports or expression languages (CWE-94). Never evaluate input; map it to a fixed set of operations or parse it as data.
Code signing
Signing software so that users and systems can verify who published it and that it has not changed since. Its value depends on protecting the signing key, often in an HSM or by using short-lived keyless certificates, and on verifiers requiring the expected signer rather than any valid signature.
Command injection
OS command injection (CWE-78). Untrusted input becomes part of an operating-system command and lets an attacker run commands. Pass arguments as a list without a shell, and validate any value that the program could read as an option.
Compensating control
A control used instead of one that a standard or policy requires but that cannot be met, which must address the same risk to at least the same degree. It needs evidence that it works and a date to revisit it.
Computer use
An AI agent capability in which the model operates a desktop or browser through screenshots, clicks and keystrokes. Everything on the screen is input it may follow, so run it in an isolated virtual machine, container or browser profile with a domain allowlist, keep credentials out of its reach and have a person confirm purchases, messages and other irreversible actions.
Confused deputy
A privileged component is tricked into using its own authority for a less privileged caller (CWE-441). An MCP server that forwards a client’s token, or an agent acting on a helper’s unverified claim, are examples.
CORS
Cross-Origin Resource Sharing. The Fetch Standard protocol in which a server’s Access-Control-Allow-* response headers let pages from other origins read its responses, relaxing the same-origin policy. It never stops a request from being sent, and reflecting any Origin with credentials allowed lets other sites read authenticated data (CWE-942).
CPE
Common Platform Enumeration. NIST’s structured naming scheme for products (cpe:2.3:part:vendor:product:version and further fields), which the NVD uses to state what a CVE affects. It maps poorly to open source packages, where purl identifies components more precisely.
CRA
Cyber Resilience Act, Regulation (EU) 2024/2847. EU law setting cybersecurity requirements for products with digital elements. Manufacturer reporting of exploited vulnerabilities and severe incidents has applied since 11 September 2026; most other obligations apply from 11 December 2027.
Credential stuffing
Automated login attempts with username and password pairs leaked from other sites, which succeed wherever people reused passwords (MITRE ATT&CK T1110.004). Phishing-resistant MFA, checks against breached passwords and rate limits keyed to the account and the source defend against it.
CRL
Certificate revocation list (RFC 5280). A list, signed and published by a certificate authority, of the serial numbers of certificates it revoked before they expired. Publicly trusted CAs must publish CRLs, and browsers distribute compact revocation sets built from them.
CRLF injection
Carriage-return and line-feed characters in untrusted input end a line where the application did not expect one, letting an attacker add HTTP headers or split a response (CWE-113) or forge log entries (CWE-117); CWE-93 is the general class. Reject or encode CR and LF in values written to line-based formats.
CSAF
Common Security Advisory Framework, an OASIS standard. A machine-readable format for security advisories, with a VEX profile that states whether products are affected by a vulnerability.
CSIRT
Computer security incident response team. A team that receives, analyzes and coordinates the response to security incidents for an organization, a sector or a country. Under the EU Cyber Resilience Act, manufacturers notify the national CSIRT designated as coordinator, and ENISA, of actively exploited vulnerabilities.
CSP
Content Security Policy, not to be confused with CSP for a cloud service provider. A response header telling the browser which sources of scripts, styles, images and other content a page may load or run. It limits the damage of an injection flaw but does not fix it (W3C Content Security Policy Level 3).
CSPM
Cloud security posture management. Tools that read cloud control-plane APIs and configuration to find public storage, open administrative ports, disabled logging and other benchmark failures; they do not see what runs inside a workload.
CSPRNG
Cryptographically secure pseudorandom number generator. Its output cannot feasibly be predicted, even by someone who has seen earlier output, so it is the source for keys, nonces, salts, session IDs and reset tokens. Use the operating system’s generator through APIs such as Python’s secrets module or crypto.getRandomValues, never Math.random or a general-purpose random module.
CSRF
Cross-site request forgery (CWE-352). Another site makes the victim’s browser send an authenticated request the user did not intend. Defenses depend on how credentials are sent, the request type and the application context.
CVD
Coordinated vulnerability disclosure. The process in which the finder of a vulnerability, the affected vendors and any coordinators share information privately, prepare fixes and publish advisories on an agreed schedule, often within 90 days. ISO/IEC 29147 and ISO/IEC 30111 describe the vendor’s side.
CVE
Common Vulnerabilities and Exposures. An identifier and record for a publicly disclosed vulnerability, assigned by a CVE Numbering Authority; an identifier alone does not establish your exposure.
CVSS
Common Vulnerability Scoring System, maintained by FIRST; version 4.0 was published in November 2023. It scores technical severity from 0 to 10, not the likelihood of exploitation or the risk to a particular business.
CWE
Common Weakness Enumeration. MITRE’s catalog of software and hardware weakness types, such as CWE-79 for cross-site scripting, distinct from records of particular vulnerabilities.
CWPP
Cloud workload protection platform. Tools that protect running virtual machines, containers and functions through an agent, an eBPF sensor or disk snapshots: vulnerable packages, malware and runtime attacks.
CycloneDX
A bill-of-materials standard maintained by OWASP and published by Ecma International as ECMA-424, covering SBOMs and inventories of services, cryptography and machine-learning models, with built-in support for VEX.
D
Dangling DNS
A DNS record, usually a CNAME, an A record or an NS delegation, that still points to a cloud resource, address or zone that has been deleted or released. Whoever can claim that resource next controls the name, which is how subdomain takeovers happen.
DAST
Dynamic application security testing. Tools test a running application from outside by sending crafted requests and observing the responses, as ZAP does. They need no source code and find runtime and configuration flaws, but only on paths they can reach and authenticate to.
Data binding
A framework feature that copies request parameters onto an object by property name, including nested property paths. Unless the bindable fields are restricted, a client can set properties the developer never meant to expose, which is mass assignment and, in Spring4Shell, reached the class loader.
Data flow
The movement of information among actors, processes, stores, and external systems. Modeling it helps identify where security assumptions change.
Data minimization
Collecting, keeping and exposing only the personal data needed for a stated purpose (GDPR Article 5(1)(c)). Data that is never collected cannot leak, so minimization also reduces the impact of a breach and the work of retention.
Data poisoning
Manipulating training, fine-tuning or retrieval data to degrade a model or plant behavior that chosen inputs trigger (OWASP LLM05:2026). In 2025 research about 250 poisoned documents backdoored models from 600 million to 13 billion parameters alike.
Defense in depth
Independent controls at different boundaries that reduce reliance on a single mechanism. Multiple copies of the same failure-prone control may not provide independence.
DEK
Data encryption key. The key that encrypts the data itself in envelope encryption, typically generated for each object, file or tenant and stored beside the data only in wrapped, that is encrypted, form.
Denial of service
Making a service unavailable to legitimate users by exhausting a resource such as CPU, memory, connections, disk or a paid quota, or by crashing it (CWE-400). In applications it often comes from unbounded input sizes, expensive queries, ReDoS or decompression bombs rather than from traffic volume.
Dependency
A component required to build or operate software. Direct dependencies are explicitly selected; transitive dependencies are introduced through other components.
Dependency confusion
An attacker publishes a package to a public registry under the name of an organization’s internal package, usually with a higher version, and an installer that consults both registries fetches the public one. Use scoped or reserved names, a single registry proxy and hash-pinned lockfiles.
DevSecOps
Building security practices and automated checks into DevOps delivery, so security work happens in the same pipelines and teams that build and run the software rather than as a gate at the end. NIST SP 800-204D covers software supply chain security in DevSecOps CI/CD pipelines.
Differential privacy
A mathematical guarantee that the result of an analysis, or a model trained on data, changes very little whether or not any one person’s data is included, bounded by a privacy-loss parameter epsilon; a smaller epsilon means stronger privacy and noisier results. NIST SP 800-226 explains how to evaluate such claims.
Digital signature
A public-key mechanism used to verify the origin and integrity of a signed message under defined assumptions. It does not conceal the message.
DOM-based XSS
Cross-site scripting in which client-side script takes attacker-controlled data from a source such as location.hash or a postMessage event and writes it to a dangerous sink such as innerHTML or eval, so the payload may never reach the server. Use safe sinks such as textContent and enforce them with Trusted Types.
DORA
Digital Operational Resilience Act, Regulation (EU) 2022/2554. It sets ICT risk management, incident reporting, resilience testing and ICT third-party risk rules for EU financial entities, including contract terms and a register of information for their ICT providers, and applies from 17 January 2025.
DPoP
Demonstrating Proof of Possession (RFC 9449). An OAuth client signs a short-lived proof JWT for each request with its own private key, and the authorization server binds the tokens it issues to that key, so a stolen access or refresh token is useless without the key.
DSPM
Data security posture management. Tools that sample cloud storage and databases to find where sensitive data lives, how it is classified and which identities and networks can reach it.
DSSE
Dead Simple Signing Envelope. A signature format that signs an arbitrary payload together with its payload type, so a signature over one kind of document cannot be passed off as another; in-toto attestations travel in DSSE envelopes.
E
ECB
Electronic codebook (NIST SP 800-38A). A block cipher mode that encrypts each block independently with the same key, so identical plaintext blocks give identical ciphertext blocks and patterns in the data show through. Do not use it to encrypt data; use an AEAD.
ECDHE
Elliptic-curve Diffie-Hellman ephemeral. A key exchange in which each side generates a fresh elliptic-curve key pair for every handshake, as with X25519 in TLS 1.3 (RFC 9846), so the session keys have forward secrecy; a signature with the server’s certificate key authenticates the exchange.
ECDSA
Elliptic Curve Digital Signature Algorithm (FIPS 186-5). Each signature uses a secret per-signature nonce, and a repeated or biased nonce reveals the private key; deterministic nonces (RFC 6979) or EdDSA avoid that risk.
EdDSA
Edwards-curve Digital Signature Algorithm (RFC 8032), usually used as Ed25519. Its nonces are derived deterministically from the key and the message, which removes ECDSA’s nonce-reuse failure, and FIPS 186-5 approves it.
Embedding
A vector of numbers that a model computes to represent the meaning of text, images or other data, so that similar items lie close together. RAG systems store embeddings in a vector database to find relevant chunks; embeddings can be inverted to recover much of the source text, so protect them like the data (OWASP LLM09:2026).
Encryption
A keyed transformation intended to protect confidentiality. Integrity, authentication, access control, and replay resistance require appropriate additional design.
Envelope encryption
Encrypting data with a data encryption key (DEK) and then encrypting, or wrapping, that DEK with a key encryption key (KEK) held in a KMS or HSM. Only the wrapped DEK is stored with the data, so the KEK never leaves its service, and rotating it means rewrapping DEKs rather than re-encrypting all the data.
EPSS
Exploit Prediction Scoring System, from FIRST. It estimates the probability that a published CVE will be exploited in the wild in the next 30 days; combine it with exposure, impact and evidence of known exploitation.
EU AI Act
Regulation (EU) 2024/1689, the EU’s Artificial Intelligence Act. It classifies AI systems by risk, has prohibited some practices since 2 February 2025 and has applied obligations for general-purpose AI models since 2 August 2025; high-risk systems must meet requirements including accuracy, robustness and cybersecurity (Article 15).
Excessive agency
An LLM application or agent can take damaging actions because its tools have more functionality, permission or autonomy than the task needs (OWASP LLM03:2026).
Expression language injection
Untrusted input becomes part of an expression in a language such as Spring Expression Language (SpEL), OGNL or Jakarta Expression Language, which can call methods and create objects, so the attacker runs code (CWE-917). Never evaluate request data as an expression; map it to fixed operations.
F
Fail closed
When an authorization or validation decision cannot safely be completed, the protected operation is refused instead of implicitly allowed.
FedRAMP
Federal Risk and Authorization Management Program. The US government program for assessing and authorizing the cloud services that federal agencies use, based on NIST SP 800-53 controls. Its FedRAMP 20x process is replacing document-heavy reviews with automated, evidence-based validation.
FIDO2
The FIDO Alliance’s specifications for passwordless, phishing-resistant authentication: W3C WebAuthn for the browser and platform API, and CTAP for talking to external authenticators such as security keys. Passkeys are FIDO2 credentials.
Fine-tuning
Further training a pre-trained model on a smaller dataset to adapt it to a task, domain or style. It can weaken safety training and, with poisoned data, plant backdoors, so fine-tuning data needs the same provenance and review as code.
Forward secrecy
A property of a key exchange in which a later compromise of long-term keys does not expose past session keys, because each session uses ephemeral keys that are then discarded. TLS 1.3 provides it through (EC)DHE key exchange, though resumption with a pre-shared key alone does not.
Fuzzing
Automated testing that feeds a program large numbers of generated or mutated inputs and watches for crashes, sanitizer reports, hangs or broken invariants. Coverage-guided fuzzers such as libFuzzer and AFL++ keep the inputs that reach new code, which finds parser bugs that example-based tests miss.
G
GDPR
General Data Protection Regulation, Regulation (EU) 2016/679. EU law on processing personal data: Article 25 requires data protection by design and by default, Article 32 requires security appropriate to the risk, and Article 33 requires notifying the supervisory authority of a personal data breach within 72 hours where feasible.
Guardrail
A control placed around a model that checks or constrains its inputs and outputs, such as a prompt-injection classifier, a content filter or a schema validator. Guardrails reduce risk probabilistically and can be bypassed, so authorization for consequential actions must still be enforced outside the model.
H
Hash function
A function mapping inputs of any length to fixed-size digests. Cryptographic hash functions are designed to resist preimage, second-preimage and collision attacks; they are not encryption and, used alone, do not authenticate a message.
HKDF
HMAC-based Extract-and-Expand Key Derivation Function (RFC 5869). It extracts a uniformly random key from input keying material, optionally with a salt, then expands it into as many keys as needed, with an info string binding each key to its purpose. It suits shared secrets and master keys, not passwords.
HMAC
Hash-based message authentication code (RFC 2104). A tag computed from a message, a shared secret key and a cryptographic hash. Anyone holding the key can create valid tags, so a valid tag shows that some key holder produced the message, not which one.
HSM
Hardware security module. A tamper-resistant device or service that generates, stores and uses cryptographic keys so that private keys cannot be exported in plaintext; its FIPS 140-3 validation level states how much physical and logical protection it provides.
HSTS
HTTP Strict Transport Security (RFC 6797). A response header with which a site tells browsers to connect to it only over HTTPS for max-age seconds, optionally including subdomains, which stops downgrades to plain HTTP after the first visit; the browsers’ preload list covers the first visit too.
HttpOnly
A cookie attribute that stops scripts in the page from reading the cookie through document.cookie (RFC 6265); a sensitive cookie without it is CWE-1004. It limits theft of session cookies through XSS but not their misuse: injected script can still send requests that carry the cookie.
I
IaC
Infrastructure as code. Machine-readable definitions used to provision and configure infrastructure, which require review of both source and resulting deployment plans.
IAM
Identity and access management. The policies, processes and systems that create, change and remove identities and decide what they can access; in a cloud platform, the IAM service evaluates every API call against identity and resource policies.
IAST
Interactive application security testing. An agent inside the running application observes data flow while tests or real traffic exercise it, combining the code-level detail of SAST with the runtime evidence of DAST; it covers only the code that is exercised.
ID token
A JWT that an OpenID Connect provider issues to the client to say who signed in, when and how (OpenID Connect Core 1.0). The client must validate its signature, issuer, audience (its own client ID), expiry and nonce, and must not send it to APIs as an access token.
Idempotency key
A unique value that a client sends with a request that must not take effect twice, such as a payment, so the server performs the operation once and returns the stored result when the request is retried with the same key. The server must scope keys to the caller and record them atomically with the result.
Identity provider
A service that authenticates users and issues assertions or tokens about them to the applications that rely on it, such as an OpenID provider or a SAML identity provider. A compromised or misconfigured identity provider affects every application that trusts it.
IDOR
Insecure direct object reference. Common shorthand for missing authorization around a user-controlled object identifier. Random identifiers do not replace access checks.
IMDS
Instance metadata service. A link-local HTTP endpoint, 169.254.169.254 on the major clouds, through which a virtual machine reads its configuration and temporary credentials. SSRF to it is a common route to cloud credential theft; AWS IMDSv2 requires a session token obtained with a PUT request and limits how many network hops a response can travel.
in-toto
A framework for securing software supply chains with signed metadata about each step (who performed it, on which inputs and producing which outputs), so a verifier can check that the steps of a defined layout were followed. Its attestation format underlies SLSA provenance.
Incident response
The organized handling of a security incident: preparing, detecting and analyzing it, containing, eradicating and recovering, and learning lessons afterward. NIST SP 800-61 Rev. 3 (April 2025) organizes it around the CSF 2.0 functions.
Indirect prompt injection
Prompt injection delivered through content the model reads rather than through the user’s own input, such as a web page, an email, a document, a tool result or a retrieved chunk. The user may never see the instructions, so applications must treat all retrieved content as untrusted.
Information disclosure
Exposure of sensitive information to someone not authorized to see it (CWE-200), such as stack traces, configuration, source code, version details or personal data. MITRE prefers mapping a finding to its specific cause, such as error messages (CWE-209) or directory listings (CWE-548).
Ingress controller
The Kubernetes component that reads Ingress (or Gateway API) objects and configures a proxy to route outside traffic to services in the cluster. It turns user-supplied objects into proxy configuration and often holds TLS secrets, so input validation, network exposure and its permissions decide how much one bug can reach.
Input validation
Checking at a trust boundary that input has the expected type, length, range, format or allowed values before using it, preferably with allowlists and after canonicalization. It narrows what reaches the code but does not replace output encoding or parameter binding (CWE-20).
Insecure deserialization
Rebuilding objects from attacker-controlled bytes with a format that can create arbitrary types or run code, such as Python pickle or Java serialization (CWE-502). Use data-only formats or an allowlist of types.
Integer overflow
An arithmetic result exceeds the range of its integer type and wraps around or is truncated (CWE-190). In size calculations it leads to undersized allocations and then buffer overflows; check arithmetic before allocating or use checked operations.
Integrity
Assurance that data or behavior has not been altered contrary to the system’s rules. It is distinct from confidentiality and availability.
ISO/IEC 27001
The international standard for an information security management system (ISMS). Its 2022 edition’s Annex A lists 93 controls in four themes; certification shows that an audited management system operates for a defined scope, not that a particular product is secure.
IV
Initialization vector. A per-message input to a block cipher mode that makes identical plaintexts encrypt differently. Requirements differ by mode: CBC needs an unpredictable IV, while counter-based modes such as GCM need a unique one (a nonce), and reuse under the same key can expose plaintext.
J
Jailbreak
An input crafted to make a model ignore its safety training or system instructions, for example through role-play, obfuscation or many examples, so that it produces content or actions it should refuse. It is a form of prompt injection; model refusals are one layer of defense, not a security boundary.
JWKS
JSON Web Key Set (RFC 7517). A JSON document of public keys that a verifier fetches from a trusted issuer configuration to check token signatures; the verifier needs an explicit key-refresh policy.
JWT
JSON Web Token (RFC 7519). A compact, usually signed, container of claims. Decoding a JWT does not establish authenticity, intended audience, purpose or current permission; verify the signature and claims first.
K
KDF
Key derivation function. A function that derives one or more cryptographic keys from a secret input such as a shared secret, a master key or a password. Key-based KDFs such as HKDF expand an already strong secret; password-based KDFs such as PBKDF2, scrypt and Argon2 are deliberately slow to resist guessing.
KEK
Key encryption key. A key used only to encrypt, or wrap, other keys such as data encryption keys; it usually stays inside a KMS or HSM, and rotating it means rewrapping the keys it protects, not re-encrypting the data.
KEV
Known Exploited Vulnerabilities catalog, maintained by CISA. It lists CVEs with reliable evidence of exploitation in the wild and clear remediation guidance; US federal civilian agencies must fix them on deadlines set by BOD 26-04.
Key rotation
Replacing a cryptographic key with a new one on a schedule, its cryptoperiod, or after a suspected compromise, while keeping old keys available to decrypt or verify existing data until it is re-encrypted or expires. Rotation limits how much data one key protects; it does not undo a past compromise.
KMS
Key management service. A managed service, such as AWS KMS, Google Cloud KMS or Azure Key Vault, that creates and stores keys, often in HSMs, and performs encryption, decryption, signing and key wrapping under access policies with audit logs, so applications never handle the key material.
L
LDAP injection
Untrusted input changes the meaning of an LDAP search filter or distinguished name (CWE-90), letting an attacker bypass a login check or read other directory entries. Escape filter values as RFC 4515 requires and DN values as RFC 4514 requires, or build filters with a library.
Least privilege
Granting only the permissions needed for the intended task, for the required duration and scope. Apply it to humans, services, build jobs, and agents.
LINDDUN
A privacy threat modeling framework from KU Leuven, named for its threat categories: linking, identifying, non-repudiation, detecting, data disclosure, unawareness and unintervenability, and non-compliance. Like STRIDE, it is applied to a data flow diagram of the system.
LLM
Large language model. A neural network, usually a transformer, trained on large text corpora to predict the next token, which applications prompt to produce text, code or tool calls. It does not separate instructions from data, so anything placed in its context can steer it.
Lockfile
A file written by a package manager, such as package-lock.json, poetry.lock or Cargo.lock, that records the exact resolved version, and usually a checksum, of every direct and transitive dependency. Installing from it makes builds repeatable and detects a changed artifact.
Log injection
Untrusted input written to a log forges entries, breaks log parsers, or injects markup or lookup expressions into the tools that read the log (CWE-117). Encode CR, LF and delimiters and log structured fields; Log4Shell showed the danger of a logging library that interprets the messages it logs.
M
MAC
Message authentication code. A short tag computed from a message and a secret key, which the receiver recomputes to detect any change and to confirm that a holder of the key sent it; HMAC, CMAC and the tags of AEAD modes are MACs. Compare tags in constant time, and add a counter or timestamp, because a MAC alone does not stop replay.
Mass assignment
An API copies request fields straight into an object, so a caller can set properties they should not, such as a role or a price (CWE-915, part of OWASP API3:2023).
MASVS
OWASP Mobile Application Security Verification Standard. It sets security requirements for mobile apps in eight control groups (storage, cryptography, authentication, network, platform, code, resilience and privacy), and the companion MASTG explains how to test them.
MCP
Model Context Protocol. An open protocol connecting AI applications (clients) to servers that expose tools, resources and prompts. Revision 2026-07-28 is stateless and defines OAuth-based authorization for its HTTP transport.
Membership inference
An attack that infers whether a particular record was in a model’s training data (MITRE ATLAS AML.T0024.000). Measure it by the true-positive rate at a low false-positive rate, not by average accuracy.
Memory poisoning
Corrupting what an AI agent keeps and reuses across sessions, such as saved memories, conversation summaries, embeddings or retrieval stores, so that an attacker’s instruction or false fact steers later reasoning or tool use (OWASP ASI06:2026 Memory & Context Poisoning; MITRE ATLAS AML.T0080). Gate memory writes in code, scope them per user and record their source.
Memory safety
Freedom from memory-access errors such as out-of-bounds reads and writes, use-after-free and double free. Memory-safe languages, including Rust outside unsafe blocks, Go, Java, C#, Python and Swift, rule these classes out by construction; CISA and partner agencies urge vendors to publish roadmaps for moving to them.
MFA
Multi-factor authentication. Authentication that requires two or more different kinds of factor: something you know, something you have and something you are. SMS codes, TOTP and push approvals can be phished in real time; passkeys and other FIDO2 authenticators resist phishing (NIST SP 800-63B).
MITRE ATLAS
Adversarial Threat Landscape for Artificial-Intelligence Systems. MITRE’s knowledge base of tactics and techniques used against AI-enabled systems, modeled on ATT&CK and supported by case studies; technique IDs look like AML.T0051, LLM Prompt Injection.
MITRE ATT&CK
A knowledge base of adversary tactics and techniques observed in real intrusions, such as T1078 Valid Accounts and T1195 Supply Chain Compromise, used to plan detections, describe incidents and test defenses.
ML-DSA
Module-Lattice-Based Digital Signature Algorithm (FIPS 204). A post-quantum signature scheme derived from CRYSTALS-Dilithium, whose public keys and signatures are much larger than ECDSA’s, which matters for certificates and constrained protocols.
ML-KEM
Module-Lattice-Based Key-Encapsulation Mechanism (FIPS 203). A post-quantum scheme derived from CRYSTALS-Kyber for establishing a shared secret; TLS deployments pair ML-KEM-768 with X25519 in the hybrid X25519MLKEM768 group, which stays secure if either component holds.
Model extraction
Building a functionally similar copy of a model, or recovering its parameters, by querying it at scale and training on the answers (MITRE ATLAS AML.T0024.002). Rate limits, query monitoring and returning less detail, such as no raw probabilities, make it costlier.
Model inversion
Reconstructing sensitive features of training data, such as a face or a record’s attributes, from a model’s outputs or confidence scores (MITRE ATLAS AML.T0024.001).
mTLS
Mutual TLS. TLS in which the client also presents a certificate and proves possession of its private key, so both ends authenticate each other; it is common between services and binds OAuth tokens to a client certificate in RFC 8705. The server must still map the certificate’s identity to an authorization decision.
N
NIS2
The EU directive on a high common level of cybersecurity, Directive (EU) 2022/2555. It requires essential and important entities in listed sectors to manage cybersecurity risk and report significant incidents, starting with an early warning within 24 hours; member states had to apply it from 18 October 2024.
Nonce
A value used only once within the scope a cryptographic construction requires. Some schemes need only uniqueness and others unpredictability; reusing an AES-GCM nonce with the same key breaks both confidentiality and authenticity.
NoSQL injection
Untrusted input changes the structure of a NoSQL query, for example a JSON object such as {"$ne": null} where the code expected a string, which turns an equality test into an operator (CWE-943). Check types and build queries from fixed operators.
NULL pointer dereference
A program uses a pointer that is NULL as if it pointed to valid memory (CWE-476). It usually crashes the process, which is a denial of service, and in some kernels and embedded systems it can be exploitable.
NVD
National Vulnerability Database, run by NIST. It enriches CVE records with CVSS scores, CWE mappings and CPE applicability data; since 2024 a large backlog has left many records without that enrichment, so use other sources as well.
O
OAuth
OAuth 2.0, the authorization framework (RFC 6749) with which a user lets a client application obtain limited, scoped access tokens for an API without sharing credentials. It delegates authorization and does not by itself authenticate the user to the client, which OpenID Connect adds; RFC 9700 is the current security best practice.
OCSP
Online Certificate Status Protocol (RFC 6960). A client asks a CA’s responder whether a certificate has been revoked, or the server staples a recent signed response to its handshake. It is now optional for publicly trusted TLS certificates, and Let’s Encrypt turned its responders off in 2025.
OIDC
OpenID Connect. An identity layer on OAuth 2.0: the provider issues an ID token telling the client who signed in, when and how. An access token is for calling APIs and does not prove identity to the client (OpenID Connect Core 1.0).
Open redirect
An application redirects to a URL taken from the request without validating it (CWE-601), so attackers can borrow its domain for phishing or to capture OAuth codes and tokens.
Origin
A browser security boundary defined by a URL’s scheme, host and port (RFC 6454). Different paths on one origin do not create separate origins.
OSPS Baseline
The OpenSSF Open Source Project Security Baseline: a minimum set of security controls for an open-source project in three maturity levels, covering access control, build and release, documentation, governance, legal, quality, security assessment and vulnerability management. The OpenSSF Security Baseline SIG maintains it; the current version is v2026.08.28.
OSV
Open Source Vulnerabilities. A JSON schema, and a database at osv.dev, that records vulnerabilities by package ecosystem and affected version ranges, aggregating sources such as GitHub Security Advisories, PyPI and the Go vulnerability database, so scanners can match lockfile versions precisely.
Out-of-bounds read
Reading memory before the start or past the end of a buffer (CWE-125), which can disclose secrets such as private keys or reveal memory layout that helps another exploit; Heartbleed (CVE-2014-0160) was one.
Output encoding
Converting data into a safe form for the exact context it is written into, such as HTML entity encoding for element text or JavaScript string escaping inside a script, so the interpreter treats it as data. Encoding for the wrong context leaves the injection open.
P
Padding oracle
An observable difference in decryption handling that can expose information about encrypted data. Errors, timings, or other signals can act as the oracle.
Parameter binding
Providing data values separately from a fixed query statement through a database driver. Placeholders do not generally bind identifiers or keywords.
Passkey
A FIDO credential that replaces a password: a key pair bound to one site’s relying-party ID, whose private key stays in an authenticator or is synced by a credential manager and signs each WebAuthn sign-in challenge. Because the signature is tied to the site and no shared secret is sent, passkeys resist phishing and credential stuffing.
Path traversal
Input such as ../ escapes the intended directory when an application builds a file path from it (CWE-22). Resolve the full path and check that it stays inside the allowed root.
PBKDF2
Password-Based Key Derivation Function 2 (RFC 8018). It applies a pseudorandom function, usually HMAC-SHA-256, many times to a password and salt. It is not memory-hard, so it needs a high iteration count, 600,000 for PBKDF2-HMAC-SHA256 in OWASP’s guidance; use it where FIPS 140 validation is required.
PCI DSS
Payment Card Industry Data Security Standard. Requirements from the PCI Security Standards Council for entities that store, process or transmit cardholder data; version 4.0.1 (June 2024) is current, and Requirement 6 covers secure development and the vulnerability management of software.
Penetration test
An authorized, time-boxed attempt to find and exploit vulnerabilities in a defined scope, to show what an attacker could achieve. Its value depends on the scope, the testers’ skill and whether the findings are fixed and retested (NIST SP 800-115).
Pepper
A secret value combined with each password before or during hashing and stored apart from the password database, for example in an HSM or a secrets manager, so a stolen database alone does not allow offline guessing. Unlike a salt it is shared across records and must stay secret, and losing it makes every hash that uses it unverifiable.
PKCE
Proof Key for Code Exchange (RFC 7636). The client sends a hash of a one-time secret with the authorization request and the secret itself when redeeming the code, so a stolen code is useless. RFC 9700 requires it for public clients and recommends it for all others.
PKI
Public key infrastructure. The roles, policies, trust anchors, certificates, and lifecycle processes used to bind public keys to identities or other assertions.
Post-quantum cryptography
Public-key algorithms designed to resist attack by large quantum computers, which would break RSA and elliptic-curve cryptography. NIST published ML-KEM (FIPS 203), ML-DSA (FIPS 204) and SLH-DSA (FIPS 205) in August 2024 and selected HQC as an additional key-encapsulation mechanism in March 2025.
Prompt injection
Text that makes a language model act against the application’s intent, typed by a user or hidden in documents, web pages, tool results or memory the model processes (OWASP LLM01:2026). Enforce authority outside the model and treat its output as untrusted.
Prototype pollution
In JavaScript, untrusted keys such as __proto__ or constructor.prototype modify a shared object prototype and change the behavior of every object that inherits from it (CWE-1321).
Provenance
Information about how, where and from what an artifact was produced, such as a SLSA provenance attestation. Its value depends on who produced it, how it is verified and whether it is bound to the exact artifact.
Pseudonymization
Processing personal data so that it can no longer be attributed to a specific person without additional information that is kept separately under technical and organizational controls (GDPR Article 4(5)). For anyone who can reattribute it, pseudonymized data is still personal data.
PSIRT
Product security incident response team. The team in a vendor that receives vulnerability reports about its products, coordinates fixes, publishes advisories and often acts as the CVE Numbering Authority; FIRST’s PSIRT Services Framework describes its services.
purl
Package URL. A structured identifier for a package’s ecosystem, namespace, name, version and qualifiers, such as pkg:npm/lodash@4.17.21. It identifies a component, not whether a deployment is affected by a vulnerability.
R
Race condition
Behavior that depends on the timing of concurrent operations. A time-of-check to time-of-use race (CWE-367) separates a check from the action it guards; make the check and the action one atomic operation.
RAG
Retrieval-augmented generation. An application retrieves documents, often by vector similarity, and adds them to a model’s context. Retrieval must enforce the user’s access rights before content reaches the model.
RASP
Runtime application self-protection. Instrumentation inside a running application that detects and blocks attacks using the application’s own context, such as an injected query reaching the database driver (NIST SP 800-53 control SI-7(17)). It is a runtime mitigation, not a fix for the flaw.
Rate limiting
Bounding how many requests a client, account, key or address may make in a period, to slow guessing and scraping and to protect capacity and cost (OWASP API4:2023). Key the limit to the resource at risk, such as the target account for login attempts, not only to the source IP address.
RBAC
Role-based access control. Permissions are granted to roles, and users or services receive permissions by being assigned roles. It is simple to review but cannot by itself express per-object rules such as ownership, which need ABAC or ReBAC checks as well.
ReBAC
Relationship-based access control. Access follows relationships between subjects and objects, such as owner of a document or member of a group that can view a folder, stored as a graph and checked by query; Google’s Zanzibar paper (2019) popularized it, and OpenFGA and SpiceDB implement it.
ReDoS
Regular expression denial of service (CWE-1333). A pattern with nested or overlapping quantifiers takes exponential or polynomial time on crafted input in a backtracking regex engine. Use a linear-time engine such as RE2 or Rust’s regex, bound the input length, or rewrite the pattern.
Reentrancy
A smart contract makes an external call that lets the callee call back in before the first call has finished updating state (OWASP Smart Contract Top 10 SC05, CWE-1265). Update state before external calls and guard every function that shares it.
Refresh token
A long-lived credential that an OAuth client exchanges at the token endpoint for new access tokens without involving the user (RFC 6749). Because it outlives sessions, store it securely; for public clients RFC 9700 requires sender-constrained refresh tokens or rotation that detects reuse.
Replay attack
Reuse of a previously valid message or credential to cause an unintended repeated operation. Message authenticity alone does not establish freshness.
Reproducible build
A build that produces bit-for-bit identical artifacts from the same source, build environment and instructions, so independent parties can rebuild a release and confirm that the published binary matches its source.
Request smuggling
HTTP request smuggling (CWE-444). A front end and a back end disagree about where one request ends, so an attacker hides a second request inside the first.
Risk acceptance
A documented decision by an accountable owner to tolerate a known risk without further treatment, for a stated period and with conditions that trigger review. When it waives a policy requirement it is often called a risk exception; it records a choice and does not reduce the risk.
RSA
A public-key algorithm for encryption and digital signatures whose security rests on the difficulty of factoring the product of two large primes (RFC 8017). Use keys of at least 2048 bits, OAEP padding for encryption and PSS or PKCS #1 v1.5 for signatures; a large quantum computer would break it.
S
S2C2F
Secure Supply Chain Consumption Framework, an OpenSSF framework for consuming open-source software safely. It groups its requirements into eight practices (ingest, scan, inventory, update, audit, enforce, rebuild, and fix and upstream) and four maturity levels.
Safetensors
A file format for model weights that stores only tensors and a JSON header, so loading a file cannot run code, unlike pickle-based PyTorch checkpoints, which can execute arbitrary code when loaded (CWE-502).
Salt
A random, non-secret value stored with each password hash and fed into the hashing function, so identical passwords produce different hashes and precomputed tables are useless. Generate a unique salt for every password with a CSPRNG; NIST SP 800-63B requires at least 32 bits.
Same-origin policy
The browser rule that script from one origin can read responses and data from another origin only when that origin allows it, for example through CORS or postMessage. It restricts reading, not sending: cross-origin form posts and image loads still go out, often with the user’s cookies, which is why CSRF defenses exist.
SameSite
A cookie attribute (Strict, Lax or None) that controls whether the browser sends the cookie with cross-site requests; a Lax cookie goes only with top-level navigations that use safe methods such as GET, and Chromium-based browsers treat cookies without the attribute as Lax. It reduces CSRF but does not replace origin checks, because same-site is broader than same-origin (CWE-1275).
SAML
Security Assertion Markup Language (OASIS SAML 2.0). A federation protocol using signed assertions; the service provider must validate the exact signed assertion it consumes and its destination, audience, subject and time conditions.
SAMM
OWASP Software Assurance Maturity Model. Version 2 defines 15 security practices in five business functions; each practice is scored from 0 to 3 from evidence about two streams of activities.
Sandbox
A restricted execution environment that lets code reach only the resources it is authorized for. Browsers, mobile operating systems and Electron sandbox their own processes; for model-generated or other untrusted code, supply the boundary yourself with a hardened container, gVisor, a microVM or WebAssembly, not with restrictions inside the interpreter.
SARIF
Static Analysis Results Interchange Format, an OASIS standard (version 2.1.0). A JSON format for static analysis findings, with their locations, rules and fingerprints, which code-scanning platforms such as GitHub ingest from any tool.
SAS
Shared access signature, an Azure Storage token appended to a resource URL that grants whatever resources, permissions and time window it encodes to anyone who holds it. Account and service SAS are signed with the storage account key and a user delegation SAS with a Microsoft Entra key; a service SAS tied to a stored access policy can be revoked by deleting the policy, but an account SAS only by rotating the key.
SAST
Static application security testing. Tools analyze source code, bytecode or binaries without running them, tracing data from sources to sinks to find weaknesses such as injection. They see every code path but report false positives and miss flaws that depend on runtime configuration.
SBOM
Software bill of materials. A machine-readable inventory of the components in a software release and their relationships, usually in SPDX or CycloneDX. CISA’s 2026 minimum elements define baseline fields; completeness depends on how it was produced.
SCA
Software composition analysis. Methods and tools for identifying component dependencies and relevant vulnerability or license information.
SCIM
System for Cross-domain Identity Management (RFC 7643 and RFC 7644). A protocol for provisioning users and groups; disabling a directory user does not by itself prove that application sessions and queued work have been revoked.
scrypt
A memory-hard password-based key derivation function (RFC 7914) whose cost parameter N, block size r and parallelization p set how much CPU time and memory each guess costs. OWASP recommends it when Argon2id is unavailable.
Secret
A value whose confidentiality is necessary to enforce a security property, such as a private key, password, signing key, or access token.
Secret scanning
Searching code, commits, build logs and other artifacts for credentials such as API keys, tokens and private keys, ideally blocking them before they are pushed and covering the full history. A detected secret must be revoked and rotated; deleting it from the repository is not enough.
Security champion
A developer or engineer on a product team who takes on part-time security responsibilities, such as threat modeling, triage and review, and links the team with the security group. A program needs protected time, training and outcome measures to work (OWASP Security Champions Guide).
Security misconfiguration
Insecure settings anywhere in the stack, such as debug modes left on, unnecessary features and endpoints, default accounts, verbose errors or missing hardening. OWASP ranks it A02 in the Top 10:2025.
Service account
A non-human identity used by a workload or automation, such as a Kubernetes ServiceAccount or a cloud service account. Bind each one to one workload, prefer short-lived federated credentials to exported keys, and grant only the permissions that workload needs.
Session
A continuing interaction bound to an authenticated or anonymous context. Session credentials and lifecycle transitions need explicit protection.
Session fixation
An attacker makes the victim authenticate with a session identifier the attacker already knows, then uses that session (CWE-384). Issue a new session identifier at login and at every change of privilege, and reject identifiers the server did not issue.
Session hijacking
Taking over an authenticated session by stealing or predicting its identifier, through XSS, network interception, malware, logs or a leaked URL, and presenting it as the victim (CAPEC-593). Protect session cookies with Secure, HttpOnly and SameSite, rotate identifiers and require reauthentication for sensitive actions.
SIEM
Security information and event management. A system that collects, normalizes and correlates logs and events from many sources to detect, alert on and investigate security incidents; its detections are only as good as the logs it receives.
Sigstore
An OpenSSF project for signing and verifying software: Fulcio issues short-lived signing certificates bound to an OpenID Connect identity, Rekor records signatures in a public transparency log, and cosign signs and verifies artifacts. A verifier must require the expected identity and issuer, not merely a valid signature.
SLSA
Supply-chain Levels for Software Artifacts. A framework of tracks and levels describing increasingly strong guarantees about how an artifact was built and, since version 1.2, about its source. A level claim needs provenance that a verifier checks.
SOC 2
System and Organization Controls 2, an AICPA attestation report. An independent CPA firm gives an opinion on a service organization’s controls for security and, optionally, availability, processing integrity, confidentiality and privacy, the Trust Services Criteria. A Type 2 report covers how the controls operated over a period; a Type 1 report covers their design at a point in time.
SPDX
Software Package Data Exchange. A Linux Foundation standard for SBOMs and license information, published as ISO/IEC 5962:2021 for version 2.2.1; SPDX 3.0 adds profiles for security, AI models and datasets, among others.
SPIFFE
Secure Production Identity Framework For Everyone, a CNCF standard for workload identity. Each workload receives a SPIFFE ID (spiffe://trust-domain/path) in a short-lived X.509 or JWT SVID from a Workload API, such as SPIRE’s, so services can authenticate each other without long-lived secrets.
SQL injection
Untrusted input changes the structure of a database query (CWE-89). Bind values as parameters and allowlist any identifiers or keywords built into the query.
SSDF
NIST Secure Software Development Framework, SP 800-218. Version 1.1 (February 2022) is final and version 1.2 has been a draft since December 2025. Its practices fall into four groups: prepare, protect, produce and respond.
SSO
Single sign-on. One sign-in at an identity provider gives access to several applications through a federation protocol such as OpenID Connect or SAML. It concentrates risk in the identity provider, and logout and deprovisioning depend on each application honoring session lifetimes and revocation.
SSRF
Server-side request forgery (CWE-918). An attacker makes the server send a request to a destination of the attacker’s choosing, often an internal service or a cloud metadata endpoint the attacker cannot reach directly.
SSTI
Server-side template injection. User input becomes part of a template’s source instead of being passed to it as data, so the template engine evaluates it, which in engines such as Jinja2, Twig or FreeMarker often leads to remote code execution (CWE-1336). Render fixed templates and pass user input only as variables.
SSVC
Stakeholder-Specific Vulnerability Categorization. A decision-tree method from Carnegie Mellon’s SEI that turns facts such as exploitation status, automatability, technical impact and mission impact into an action; CISA’s version outputs Track, Track*, Attend or Act.
Staged rollout
Releasing a change to progressively larger groups of systems, starting with a small canary, with a bake time and health checks between steps so a bad change can be halted and rolled back before it reaches everyone. It applies to configuration, rules and other content as much as to code.
STRIDE
Microsoft’s threat classification: spoofing, tampering, repudiation, information disclosure, denial of service and elevation of privilege, each the violation of one property (authentication, integrity, non-repudiation, confidentiality, availability and authorization). Teams apply it to each element of a data flow diagram to ask what can go wrong.
Subdomain takeover
Someone else gains control of content served on one of your subdomains because a DNS record still points to a cloud or SaaS resource you released, and the provider lets them claim it. The subdomain then receives cookies and trust meant for your domain.
Subresource Integrity
A W3C mechanism in which an integrity attribute on a script or stylesheet element carries a cryptographic hash of the expected file, and the browser refuses to use a fetched file that does not match. It protects against a modified copy on a CDN, not against a compromised first-party server.
Supply chain
The people, systems, source, dependencies, build processes, and delivery paths involved in producing and distributing software.
System prompt
Instructions that an application places before user input in a model’s context to set its role, rules and tools. Users can often extract or infer it, so it must not hold secrets or serve as an access control (OWASP LLM08:2026 Hidden Context Exposure).
T
Threat model
A maintained description of a system, its assets and trust boundaries, plausible failure or abuse scenarios, and the controls and decisions addressing them. The Threat Modeling Manifesto frames it as four questions: what are we working on, what can go wrong, what are we going to do about it, and did we do a good enough job?
Timing attack
A side-channel attack that infers secrets from how long operations take, such as a comparison that returns at the first wrong byte of a MAC, token or password hash (CWE-208). Compare secrets with a constant-time function such as hmac.compare_digest, crypto.timingSafeEqual or MessageDigest.isEqual.
TLS
Transport Layer Security. The protocol that authenticates the server, and optionally the client, with certificates and encrypts and integrity-protects data in transit. TLS 1.3 is current and is specified by RFC 9846 (July 2026), which replaced RFC 8446; TLS 1.2 is still in use, under requirements RFC 9846 adds; TLS 1.0 and 1.1 are deprecated (RFC 8996).
Tokenization
Replacing a sensitive value, such as a card number, with a surrogate token that cannot be reversed without access to a protected vault or key. Systems that hold only tokens can reduce the scope of PCI DSS, but the tokenization service and every system that can detokenize become critical.
Tool calling
A model’s structured request, usually JSON that matches a declared schema, asking the application to run a named function with arguments. The application executes it, so it must validate the arguments and authorize the action as the user rather than trusting the model’s choice.
Tool poisoning
Hidden instructions in a tool’s name, description or schema that steer an AI agent reading them; a server can also change a definition after users approved it. Pin a hash of each reviewed definition and review any change.
TOTP
Time-based one-time password (RFC 6238). An authenticator app computes a short code with HMAC from a shared secret and the current 30-second time step. It proves possession of the secret but can be phished in real time, so servers must rate-limit attempts and reject a code that was already used.
Trust boundary
A point where data or execution crosses between contexts with different privileges, ownership, or security assumptions.
Trusted publishing
Publishing packages to a registry with short-lived credentials that the registry issues in exchange for a CI workload’s OpenID Connect identity, such as a specific GitHub Actions workflow, instead of long-lived API tokens. PyPI, npm, RubyGems and crates.io support it.
Trusted Types
A browser API, enforced through Content Security Policy with require-trusted-types-for 'script', under which dangerous DOM sinks such as innerHTML accept only typed values created by named, reviewed policies, so DOM XSS is blocked at the sink.
TUF
The Update Framework, a CNCF graduated specification for secure software update systems. Separate roles (root, targets, snapshot and timestamp) sign expiring metadata with threshold keys, so a compromised key or mirror cannot serve malicious, outdated or rolled-back updates.
TURN
Traversal Using Relays around NAT. A protocol and server that relays media and data for WebRTC peers that cannot reach each other directly; a TURN server is an outbound proxy, so it must authenticate users and refuse to relay to internal addresses.
Typosquatting
Publishing a malicious package, or registering a domain, under a name that is a plausible misspelling or variant of a popular one, such as reqeusts for requests, to catch mistyped installs and copied commands.
U
Use-after-free
A program keeps using a pointer after the memory it points to has been freed (CWE-416). If an attacker can get that memory reallocated with data they control, the stale pointer reads or calls through it, which often leads to code execution.
V
VDP
Vulnerability disclosure policy. A public statement of how to report vulnerabilities to an organization, which systems are in scope, what testing is authorized under safe harbor and what reporters can expect in return; CISA BOD 20-01 required one from US federal civilian agencies, and a security.txt file (RFC 9116) points to it.
Vector database
A database that stores embeddings and returns the vectors nearest to a query vector, used for retrieval in RAG. It must enforce the same per-user and per-tenant access rules as the source documents inside the query, not by filtering results afterward (OWASP LLM09:2026).
VEX
Vulnerability Exploitability eXchange. A machine-readable statement of whether a product is affected by a vulnerability (not affected, affected, fixed or under investigation), with a justification. Formats include CSAF, OpenVEX and CycloneDX; the issuer, evidence and freshness still matter.
W
WAF
Web application firewall. A proxy or module that inspects HTTP traffic and blocks requests that match attack signatures or anomaly rules, such as those of the OWASP Core Rule Set. It can buy time against known exploit patterns but can be bypassed, and it does not fix the vulnerability.
WebAuthn
Web Authentication, the W3C API with which browsers and platforms create and use public-key credentials scoped to a relying party; Level 3 became a W3C Recommendation in August 2026. The server sends a random challenge, the authenticator signs it with a private key bound to the site, and the server verifies the signature, the origin and the user-verification flags.
Webhook
An HTTP request that one service sends to a URL another service registered, to report an event. The receiver should verify a signature over the raw body and a timestamp, usually an HMAC, and reject stale or repeated deliveries; the sender should treat registered URLs as an SSRF risk.
WebRTC
Web Real-Time Communication. Browser and native APIs and IETF protocols for audio, video and data between peers, using ICE, STUN and TURN for connectivity and DTLS-SRTP for encryption. ASVS 5.0 chapter V17 covers its TURN, media and signaling servers.
Workload identity
A verifiable identity assigned to a running service or job. It needs bounded credentials, audience and authorization rules just as a human identity does.
X
XSS
Cross-site scripting (CWE-79). Untrusted content is interpreted as script or markup in a victim’s browser. The correct defense depends on the output context: HTML text, an attribute, a URL or a script.
XXE
XML external entity injection (CWE-611). An XML parser resolves entities that point to files or URLs, exposing local files or making server-side requests. Disable DTDs or external entity resolution.
Z
Zero trust
A security model that removes implicit trust based on network location: each request to a resource is authenticated and authorized per session from the identity, the device’s state and other context, as close to the resource as possible (NIST SP 800-207).
No matching terms. Try another spelling or search every page.