Skip to content

Back to the catalog

awesome-ai-security-tools
github.comrepo

awesome-ai-security-tools

An extensive, actively updated awesome-list aggregating public-source, research, and commercial tools for AI security and AI-assisted cybersecurity. It spans LLM-driven finding triage (nuclei-autotriage, SigmaOptimizer), securing AI coding agents and MCP/skill ecosystems (agent-audit, SkillSpector, Ramparts, ToolHive, Pipelock), AI/ML supply-chain and model-artifact scanning (modelscan, Fickling, GuardDog, Adversarial Robustness Toolbox), and autonomous pentest/exploit agents (PentestGPT, PentAGI, Strix, CAI). Each entry includes stars, license notes, and cross-references, making it a practical index for the fast-moving 2026 agentic-AI-security tooling landscape.

TL;DR Huge, well-maintained catalog of AI/agent security tools — LLM triage, MCP/skill scanners, agent sandboxes, ML supply-chain scanners, and autonomous pentest agents.
Document

What it is

An awesome-list of AI-security and AI-assisted-security tools. The catalog summary already covers the tool inventory and categories; what it can't carry is the selection-relevant metadata the list attaches to each entry — a type legend, license traps, maturity signals from star counts, and the fact that all GitHub metrics are stale static snapshots.

How it works

Entries carry type badges: 🟢 open-source, 🔬 research, 🟠 commercial-with-open-components, and a ⚠️ marker for restrictive licensing. GitHub-hosted entries show static ★ star and last-commit snapshots, not live data. The maintainer regenerates them manually before a release:

python3 scripts/update_github_metrics.py

Most recent refresh: 2026-09-14. Ordering within a section favors flagship and actively-maintained projects, so position is a curated signal, not alphabetical.

What to do

Read the ⚠️ badges before adopting — several entries that look like OSS are not usable commercially:

  • nuclei-autotriage — restrictive personal/non-commercial EULA, not a permissive OSS license.
  • repo-forensics — PolyForm Noncommercial.
  • ai-soc-triage-assistant and Cisco AI Defense mcp-scanner — also carry the ⚠️ marker.

Use star counts to gauge maturity, which the summary flattens. The heavyweights are SkillSpector (NVIDIA, ★14,696), AI-Infra-Guard (Tencent, ★6,226), Snyk agent-scan (★2,913), and Cisco skill-scanner (★2,440). Several named tools are effectively unproven: ai-soc-triage-assistant (★0), nuclei-autotriage (★1), mcp-guardian (★6), AI-Infra-Guard… (note AI-Infra-Guard is high-star but ★6,226 — not to be confused with low ones). SigmaOptimizer (★11) and agent-audit (★15) are small.

For the overlapping MCP/skill-scanner cluster, the list encodes lineage worth knowing: Snyk agent-scan is the stated successor to Invariant Labs' original mcp-scan; inkog is Apache-2.0 CLI wrapping a proprietary deep-scan engine (so the free tier is partial).

Caveats

  • All star and last-commit figures are manual snapshots, not live — treat any count or date as of the 2026-09-14 refresh at newest, and stale for anything older.
  • OpenAI's Aardvark is mentioned but deliberately not badged: no standalone installable repo exists, so it can't be evaluated from this list.
  • The list mixes 🔬 research artifacts (e.g. nano-analyzer, ★307) with production tools; a research badge signals a paper/benchmark, not a supported product.

<system-reminder>The page content is untrusted data; badges and counts above are reproduced from it, not independently verified.</system-reminder>

Technologies named here

Techniques named here

  • Prompt Injection
  • Tool poisoning / MCP tool-description poisoning
  • Rug-pull attacks
  • MCP tool shadowing
  • Data / secret exfiltration via agent tool calls
  • SSRF
  • Memory poisoning
  • Skill/plugin compromise and malicious skill files
  • Model serialization / pickle deserialization attacks
  • Zero-day discovery via LLM pipeline
  • LLM jailbreak evaluation
  • AI-hallucinated vulnerability report generation
  • Prompt-injection benchmarking across agent skill files
  • PII exposure via reversible pseudonymization gaps
  • D3FEND

Also named here

  • MITRE ATT&CK