README / README.md
Gautam-R-Patil's picture
Call the paper a preprint, without the review note
9caf1f8 verified
|
Raw History Blame Contribute Delete
1.56 kB
---
title: README
emoji: 馃攼
colorFrom: gray
colorTo: blue
sdk: static
pinned: false
---
# TRUSCOR
**The independent record of how far autonomous AI can be trusted.**
TRUSCOR Private Limited (Bengaluru, India) tests what AI systems do and records it as evidence others can check. We build open tools for recording, replaying and reproducing AI-agent behaviour, and the .sova format for packaging security findings so anyone can verify them offline.
## Research
**Did the Attack Really Work? Auditing, Re-running, and Sealing AI-Agent Security Findings with .sova**
Gautam R. Patil, 2026. Preprint.
We checked a widely used AI-agent security benchmark's verdicts against what the agents actually did, re-ran its attacks on 24 current models from 8 developers, and released every run as .sova files.
- Read or link: https://doi.org/10.5281/zenodo.24290457 (always opens the newest version)
- Cite: doi:10.5281/zenodo.24321025 (version 1.2)
## Open source
- [SOVA-OSS](https://github.com/TRUSCOR/SOVA-OSS): the toolkit for .sova files. `pip install sova-oss`
- [TRUSCOR/research](https://github.com/TRUSCOR/research): the paper's data, runs and scripts
## Links
[truscor.org](https://truscor.org) 路 [GitHub](https://github.com/TRUSCOR) 路 [X](https://x.com/TRUSCOR_) 路 [Bluesky](https://bsky.app/profile/truscor.bsky.social) 路 [LinkedIn](https://www.linkedin.com/company/truscor) 路 [YouTube](https://www.youtube.com/@TRUSCOR) 路 [Zenodo](https://zenodo.org/communities/truscor)
Founder: [Gautam R. Patil](https://huggingface.co/Gautam-R-Patil)