CAP-PC ENS flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

CVSS Scope Changed

When a vulnerability lets an attacker affect resources beyond the vulnerable component's security scope, the impact formula changes and the Privileges Required multiplier is reduced, raising the base score.

Prepared Statement

A parameterized query where SQL text is compiled separately from bound values, so user input cannot alter query structure. This is the structural defense against SQL injection.

DOM-based XSS

A cross-site scripting flaw where untrusted data flows from a client-side source such as location.hash into a dangerous sink like innerHTML or eval, without the server reflecting it.

SameSite Lax

A cookie attribute value that sends the cookie on top-level GET navigations to the site but blocks it on cross-site POSTs and subresource requests, mitigating many CSRF scenarios.

IDOR

Insecure Direct Object Reference: an access control flaw where changing an object identifier in a request returns another user's record because the server checks authentication but not authorization.

SSRF Metadata Target

The cloud instance metadata endpoint at 169.254.169.254. Server-side request forgery that reaches it can retrieve instance role credentials, so egress filters must block link-local addresses.

JWT alg none

A JWT header value that declares no signature. A verifier that trusts the header algorithm and accepts none is bypassable, so the expected algorithm must be pinned server-side.

Taint Analysis

A static analysis technique that tracks data from untrusted sources to sensitive sinks and reports a finding when a path exists with no effective sanitizer in between.

Semgrep pattern-not

A rule operator that excludes matches containing a given pattern. It is commonly used to suppress findings already guarded by a sanitizer or validation call.

CodeQL RemoteFlowSource

A CodeQL class representing data entering the program from remote input, such as HTTP parameters. Taint tracking queries use it as the starting point for source-to-sink paths.

Deserialization Sink

A dangerous function such as Java ObjectInputStream.readObject or Python pickle.loads that reconstructs objects from untrusted bytes, enabling remote code execution when fed attacker data.

Hardcoded Secret Triage

Confirming a committed credential is live by checking format validity, then searching git history with git log -S, because removing it from HEAD does not remove it from past commits.

Reachability Analysis

A technique that checks whether a vulnerable function in a dependency is actually invoked by the application, allowing low-priority triage of CVEs in unused code paths.

SafeLoader

The yaml.load parameter in PyYAML that restricts construction to basic types. Without it, untrusted YAML can instantiate arbitrary Python objects and execute code.

IMDSv2 Token Flow

AWS instance metadata v2 requires a PUT request to /latest/api/token with a TTL header, then a GET carrying that token, which blocks simple SSRF from fetching credentials.

S3 Principal Wildcard

A bucket policy statement with Principal "*" that grants actions such as s3:GetObject or s3:ListBucket to anyone. It is the primary indicator of public bucket exposure.

PassRole Escalation

An IAM misconfiguration where a principal can pass a highly privileged role to a compute service such as Lambda or EC2, gaining that role's permissions indirectly.

CloudTrail Log Validation

A setting that adds digest files so tampering with delivered log files is detectable. Disabling it is a common defense evasion step during cloud incident response.

Azure IMDS Token

A request to 169.254.169.254/metadata/identity/oauth2/token with the Metadata:true header returns an access token for the managed identity, scoped to the audience parameter.

KMS Key Policy

The resource policy on a KMS key. It must delegate to IAM for IAM policies to grant access; a restrictive key policy overrides otherwise permissive IAM grants.

Privileged Container

A container run with the privileged flag, granting all Linux capabilities and host device access. It is a high-severity finding because it enables container escape.

ATT&CK Execution Tactic

The adversary tactic covering techniques that run attacker-controlled code, such as command interpreters, scheduled tasks, and service creation, mapped by technique identifier.

Kerberoasting

Requesting a Kerberos service ticket for an account with a registered SPN and cracking it offline. Detection focuses on volume RC4 ticket requests for service accounts.

Malleable C2 Profile

A configuration that shapes beacon network traffic, including URIs, user agent, jitter, and sleep. It changes indicators without recompiling the implant.

Indirect Prompt Injection

An attack where malicious instructions are embedded in content the model retrieves, such as a document or tool output, causing the model to follow attacker directions during normal use.

Tool Allowlist

A restriction limiting which functions an LLM agent may invoke and with what scopes. It reduces the blast radius of prompt injection by denying unreviewed state-changing actions.

Model Extraction

An attack that queries a model API repeatedly to train a substitute model approximating the target, enabling offline adversarial research and IP theft. Rate limits and query monitoring are countermeasures.

Adversarial Example

An input modified by a small, often imperceptible perturbation that causes a model to misclassify. Perturbations built on a surrogate model frequently transfer to the target.

Data Poisoning

Manipulating training data so the resulting model misbehaves on chosen inputs. Provenance checks, deduplication, and isolated evaluation sets limit the impact.

Beacon Jitter

A randomization percentage applied to the sleep interval between implant check-ins, which breaks regular timing patterns that network detection systems use to identify command-and-control traffic.

Instruction Hierarchy

A model training approach that ranks system prompts above developer and user content, so lower-priority input cannot override higher-priority instructions during prompt injection attempts.

Membership Inference

A privacy attack that determines whether a specific record was part of a model's training set by analyzing confidence patterns in its outputs. Differential privacy and output limiting reduce it.