Last updated: March 16, 2026
proofagent is open-source software released under the MIT License. You are free to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the software.
The software is provided "as is", without warranty of any kind, express or implied. In no event shall the authors be liable for any claim, damages, or other liability.
Full license text: github.com/camgitt/proofagent/blob/main/LICENSE
proofagent is a defensive security and quality assurance tool. It is designed to help developers test, evaluate, and improve the safety and reliability of AI agents and language models. Intended uses include:
You may not use proofagent or its built-in datasets to:
The built-in safety prompt packs contain adversarial inputs by design. These exist solely for defensive testing. The prompts are questions, not instructions — they test whether models refuse harmful requests, not how to extract harmful outputs.
proofagent sends prompts to third-party AI providers (OpenAI, Anthropic, Google, etc.) using your API keys. You are solely responsible for ensuring that your use of proofagent complies with the terms of service and acceptable use policies of your chosen AI provider(s).
Some providers require prior authorization for automated safety testing, red-teaming, or adversarial evaluation. Before running built-in safety packs or adversarial prompts against any provider's API, review their current acceptable use policy. proofagent does not obtain authorization from providers on your behalf.
The AI Safety Leaderboard presents evaluation results based on automated testing. Models marked "verified" were tested live using the proofagent framework. Models marked "pending" show estimated results awaiting API access. All results reflect a specific point in time and should not be treated as definitive safety ratings.
Model behavior can vary based on system prompts, API versions, temperature settings, and other parameters. A model's score on this leaderboard does not constitute a comprehensive safety evaluation.
We do not endorse or recommend any specific AI model or provider. The leaderboard is provided for informational and educational purposes only.
proofagent is a testing tool, not a certification authority. Passing all proofagent tests does not guarantee that a model or agent is safe, unbiased, or fit for any particular purpose. AI safety is a complex, evolving field — automated testing is one layer of a comprehensive safety strategy.
The software, website, leaderboard, cost calculator, and all related content are provided "as is" and "as available" without warranties of any kind, whether express, implied, statutory, or otherwise, including but not limited to implied warranties of merchantability, fitness for a particular purpose, and non-infringement.
To the maximum extent permitted by applicable law, the authors and maintainers of proofagent shall not be liable for any indirect, incidental, special, consequential, or punitive damages, or any loss of profits or revenue, whether incurred directly or indirectly, or any loss of data, use, goodwill, or other intangible losses, resulting from:
You agree to defend, indemnify, and hold harmless the authors and maintainers of proofagent from and against any claims, damages, obligations, losses, liabilities, costs, or expenses arising from your use of the software or violation of these terms.
Nothing. proofagent has zero telemetry, zero analytics, and zero tracking in the Python package. We do not collect, transmit, or store any data from your evaluations, test results, API keys, or prompts.
The proofagent.dev website is a static site hosted on Vercel. We do not use cookies, analytics scripts, or tracking pixels. Vercel may collect standard web server logs (IP address, user agent, timestamps) as part of their hosting infrastructure — see Vercel's privacy policy.
When you use proofagent to test models, your prompts are sent directly from your machine to the AI provider's API (OpenAI, Anthropic, Google, etc.). proofagent does not proxy, log, or intercept these requests. Your data is governed by the respective provider's privacy policy and terms of service.
If you discover a security vulnerability in proofagent, please report it responsibly by emailing security@proofagent.dev or opening a private security advisory on GitHub. Do not open a public issue for security vulnerabilities.
We may update these terms from time to time. Changes will be posted on this page with an updated "Last updated" date. Your continued use of proofagent after changes constitutes acceptance of the revised terms.
These terms shall be governed by and construed in accordance with applicable law, without regard to conflict of law principles.
For questions about these terms, open an issue on GitHub.