|
|
--- |
|
|
language: en |
|
|
tags: |
|
|
- ai-security |
|
|
- llm-security |
|
|
- agentic-ai |
|
|
- red-teaming |
|
|
- adversarial-ml |
|
|
- governance |
|
|
- risk-management |
|
|
- safety |
|
|
- cybersecurity |
|
|
license: apache-2.0 |
|
|
--- |
|
|
|
|
|
# Cogensec |
|
|
|
|
|
Cogensec builds security and governance for agentic AI systems. |
|
|
|
|
|
We treat AI as decision-making infrastructure, not just software. |
|
|
That means securing identity, intent, memory, autonomy, and trust across models, tools, and multi-agent workflows. |
|
|
|
|
|
## What you’ll find here |
|
|
|
|
|
We publish practical artifacts for builders, researchers, and security teams: |
|
|
|
|
|
- **Security evaluation suites** for LLMs and agentic workflows |
|
|
- **Adversarial datasets** for testing misuse, jailbreaks, prompt injection, and tool abuse |
|
|
- **Reference agents** and **defensive patterns** (guardrails, policies, enforcement primitives) |
|
|
- **Research notes and reproducible experiments** focused on real-world deployment risks |
|
|
- **Governance templates** and guidance aligned to modern AI risk frameworks |
|
|
|
|
|
## Our focus areas |
|
|
|
|
|
- **Agent security**: tool misuse, agentic escalation, multi-agent coordination risks |
|
|
- **Non-human identity**: authentication, authorization, and lifecycle for agents and tools |
|
|
- **Memory governance**: retention, leakage, poisoning, and policy enforcement |
|
|
- **Intent and control**: goal integrity, autonomy boundaries, and safe orchestration |
|
|
- **Trust and provenance**: context integrity, auditability, attestation, and monitoring |
|
|
|
|
|
## How to use our work |
|
|
|
|
|
- Start with pinned repositories for the most current releases. |
|
|
- Each repo includes: |
|
|
- installation and quickstart |
|
|
- evaluation methodology |
|
|
- dataset/model cards where applicable |
|
|
- reproducibility notes and limitations |
|
|
|
|
|
## Responsible use |
|
|
|
|
|
Cogensec publishes security research to improve safety in AI systems. |
|
|
Some materials may describe adversarial behavior to support testing and defense. |
|
|
|
|
|
- Use responsibly and ethically. |
|
|
- Don’t deploy findings to harm others or evade safeguards. |
|
|
- Report issues or concerns through the contact channels below. |
|
|
|
|
|
## Contributing |
|
|
|
|
|
We welcome: |
|
|
- issue reports with reproduction steps |
|
|
- benchmark proposals and test cases |
|
|
- dataset improvements and labeling fixes |
|
|
- PRs that improve documentation and reproducibility |
|
|
|
|
|
If you want to collaborate on research or run joint evaluations, reach out. |
|
|
|
|
|
## Contact |
|
|
|
|
|
- Website: cogensec.com |
|
|
- GitHub: github.com/cogensec |
|
|
- Twitter: x.com/cogen_sec |
|
|
- LinkedIn: linkedin.com/company/cogensec |
|
|
|
|
|
## Citation |
|
|
|
|
|
If you use Cogensec artifacts in research, please cite the relevant repository. |
|
|
Where provided, use the `CITATION.cff` file. |
|
|
|
|
|
## License |
|
|
|
|
|
Unless stated otherwise, repositories here are released under the **Apache-2.0** license. |
|
|
Datasets may have their own terms, always check the dataset card. |
|
|
|
