Agentic LLM Vulnerability Scanner / AI red teaming kit 🧪
1.9k
Stars
266
Forks
70
Open issues
26
Contributors
AI Analysis
Agentic Security is a specialized vulnerability scanner for LLM-based agent systems and large language models, designed to identify jailbreaks, fuzzing vulnerabilities, and multimodal attacks. It serves security researchers, AI practitioners, and development teams building and deploying AI agents who need to proactively test and harden their systems against adversarial inputs and multi-step attack sequences.
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
Open-source red-teaming toolkit for LLM and agent security testing with multimodal attack support
Agentic Security is a Python-based vulnerability scanner designed to stress-test LLMs and AI agent workflows against jailbreaks, fuzzing, and multimodal attacks. Built for security researchers, ML engineers, and responsible AI teams, it offers a web UI, CLI, and CI/CD integration. The project targets a specialized but growing segment: organizations that need proactive LLM security validation before deployment.
Created in April 2024, the project emerged during a period of heightened concern about LLM safety and jailbreak attacks. It represents the broader shift toward treating LLM security as an operational requirement rather than an afterthought.
The project gained 1,913 stars over ~26 months, with most growth in the first 12 months (typical for security-focused tooling). Recent activity (last push June 2026) shows sustained maintenance rather than accelerating adoption. Growth appears driven by enterprise security initiatives and academic research into LLM robustness, not viral adoption.
adoption not verified — no case studies, public deployments, or end-user testimonials mentioned in README. Enterprise or research adoption is plausible given the problem domain, but concrete evidence is absent.
Based on README, appears to be a Uvicorn-based HTTP API server with a web UI, CLI commands for initialization and CI integration, and pluggable dataset/module system. Likely uses LLM API templates (HTTP spec format) to send attack prompts. Supports reinforcement learning and multi-step attack orchestration, though implementation details are not visible from README.
not documented in README
Last push 2026-06-23 (10 days before evaluation date) indicates active maintenance. README shows recent additions (CI/CD config examples, TOML config format) dated January 2025. Project is being actively maintained, not stagnant, though star velocity remains low (1 star/week suggests modest but stable interest).
ADOPT IF: your organization runs LLMs in production and needs a lightweight, open-source red-teaming toolkit with multi-modal attack simulation; you have in-house security/ML expertise to configure HTTP specs and interpret results. AVOID IF: you require commercial SLA, extensive documentation for non-technical stakeholders, or pre-built integrations with proprietary LLM platforms beyond OpenAI templates. MONITOR IF: you are evaluating LLM security solutions and want to track this project as a potential cost-effective complement to commercial alternatives; the active maintenance and expanding feature set (RL attacks, CI/CD) suggest it will mature further.
Independent dimensions
Mainstream potential
4/10
Technical importance
6/10
Adoption evidence
2/10
- Adoption not verified in real-world production environments; unclear if organizations are relying on it for actual security decisions.
- Test coverage not documented; no visibility into how robustly the attack vectors themselves are validated.
- HTTP spec configuration approach requires manual labor per LLM integration; may not scale to large multi-model deployments without tooling expansion.
- Depends on Hugging Face Datasets and external LLM APIs; reliability tied to third-party services.
- Security scanner accuracy and false-positive rates not discussed in README; unclear how well attack detection correlates with actual vulnerabilities.
Likely to remain a niche but stable tool used by security-conscious ML teams and academic researchers. May see adoption growth if enterprises adopt LLM red-teaming as standard practice, but unlikely to become a mainstream, required dependency.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
Information
- Language
- Python
- License
- Apache-2.0
- Last updated
- 2w ago
- Created
- 27mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
Top contributors
Recent releases
Similar repos
JoasASantos/NeuroSploit
NeuroSploit is an autonomous, multi-model penetration testing framework that...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
msoedov/agentic_security
this
|
1.9k | — | Python | 7/10 | 2w ago |
 MLSysOps/MLE-agent
|
1.6k | — | Python | 7/10 | 12mo ago |
|
JoasASantos/NeuroSploit
|
1.2k | — | Rust | 7/10 | 10h ago |
 modelscope/ms-agent
|
4.3k | — | Python | 7/10 | 2d ago |
 huhusmang/Awesome-LLMs-for-Vulnerability-Detection
|
1.1k | — | Python | 8/10 | 12h ago |
Similar star count (1,564) but broader scope (agent evaluation). Agentic Security is narrower and security-focused.
Larger (4,327 stars) but aims at agent building, not red-teaming. Different purpose within the agent ecosystem.
Rust-based, 1,199 stars. Agentic Security is Python-centric; trade-offs around language ecosystem and performance.
Agentic Security consumes these datasets. Not a direct competitor but a dependency.
