OWASP Compliance

Security posture assessment based on OWASP Top 10 for LLM Applications (2023)

Total Categories

Passing

Partial

Failing

⚠️ Attention Required

1 Critical severity item(s) need immediate action
2 High severity item(s) should be addressed soon

Live Dependency Security

STUB

Source: pnpm audit --json (re-run server-side by /v1/security/audit-summary)

Loading latest dep audit…

LLM01: Prompt Injection

CriticalPartial

Prompt injection vulnerabilities when malicious users craft inputs that manipulate the LLM to execute unintended actions, bypass controls, or access restricted information.

Checklist (4 items)

⚠️Input sanitization for user prompts

✅Output filtering for code execution

✅Context window isolation

❌Jailbreak detection

LLM02: Insecure Output Handling

HighPass

LLM output not validated or sanitized before being used to control downstream components or systems.

Checklist (4 items)

✅Output type validation

✅SQL injection prevention

✅XSS prevention in web responses

➖Code execution sandboxing

LLM03: Training Data Poisoning

CriticalPass

Malicious actors manipulate training data or fine-tuning processes to introduce vulnerabilities, backdoors, or biased behavior.

Checklist (4 items)

✅Source data provenance tracking

✅Data integrity verification

✅Unauthorized modification detection

✅Training pipeline audit logging

LLM04: Model Denial of Service

HighPartial

Resource exhaustion attacks targeting the LLM or its infrastructure, causing degraded performance or unavailability.

Checklist (4 items)

✅Rate limiting per user

✅Context window limits

⚠️GPU resource quotas

✅Request throttling

LLM05: Supply Chain Vulnerabilities

HighPass

Compromises in the LLM supply chain, including models, datasets, libraries, or infrastructure.

Checklist (4 items)

✅Model source verification

✅Dependency vulnerability scanning

✅Signed model artifacts

✅Vendor SBOM requirements

LLM06: Sensitive Information Disclosure

MediumPartial

LLMs may inadvertently reveal sensitive information, including training data or user-provided confidential data.

Checklist (4 items)

⚠️PII detection and redaction

✅Training data confidentiality

✅Prompt data encryption at rest

❌Audit log review

LLM07: Insecure Plugin Design

HighPass

Vulnerabilities in plugin architectures or tool calling mechanisms that allow unauthorized actions or data access.

Checklist (4 items)

✅Plugin permission boundaries

✅Tool input validation

✅Authorization checks per tool

➖Plugin sandboxing

LLM08: Excessive Agency

HighPartial

LLMs may take actions beyond their intended scope, causing unintended consequences.

Checklist (4 items)

⚠️Action allowlist enforcement

✅Human-in-the-loop for sensitive operations

✅Goal constraint validation

✅Tool use logging

LLM09: Overreliance

LowFail

Users may place excessive trust in LLM outputs without verification, leading to errors or security issues.

Checklist (4 items)

✅Citation provenance tracking

❌Confidence scoring

❌Human verification prompts

❌Hallucination detection

LLM10: Model Theft

MediumPass

Unauthorized access to or exfiltration of proprietary models, fine-tuned weights, or training data.

Checklist (4 items)

✅Model artifact encryption

✅Access controls on model storage

✅Exfiltration monitoring

✅Watermarking for model identification