ControlAI’s platform for AI safety and compliance is now online- go to app.controlai.org to test it!
A quick presentation of how you can leverage the platform to understand your EU AI Act obligations can be found below:
And a longer overview of all features of the platform is here:
ControlAI was born from a simple observation- that AI Safety and Compliance is too difficult to operationalise. The ecosystem is fragmented and lacking a unified vision of how to achieve safer, more robust models.
We’ve all heard discussions between e/acc members and Pause AI activists. Both sides agree we want to avoid a Terminator-like future but disagree on what needs to be done to prevent it. It is easy to sensationalise the topic because there is actually very little agreement on the basics: what does robustness means? what model is trustworthy? how much alignment and testing do we need? what kind of tests are necessary to achieve that?
ControlAI is on a mission to help you get a reliable answer to these questions and streamline your compliance journey. We want to help you understand what risks are involved in deploying LLMs before they occur and give you an easy-to-understand overview of what needs to be done to mitigate them. Too many developers have been busy deploying a particular fuzzing or guardrail tool to find out that it neither mitigates their business problems nor absolves them from legal responsibility.
ControlAI is here, uniting the fragmented ecosystem to make AI compliance simple, intuitive and easy.
How does it work?
The platform enables you to:
1. Create an inventory of AI assets and applicable threats
2. Understand what is required from you to satisfy variety of regulations such as the ISO Norms, EU AI Act and GDPR
3. Track your progress towards achieving compliance
4. Deploy verified defenses to protect your stack and monitor incidents
5. Showcase your AI safety solutions for other companies to use
Now let’s analyze each of these functionalities in detail:
Create an inventory of AI assets and applicable threats
Understanding what models are deployed within your organization is not an easy task. So is monitoring these assets for potential risks. With each week bringing new jailbreaks, even seasoned academics might have a problem identifying all potential vulnerabilities and incidents. Our platform empowers you to stay up to date with the existing risks and apply this knowledge to models deployed within your organization. ControlAI provides a single source of truth for AI inventory, enabling you to track existing assets and potential risks with ease.
Understand what is required from you to satisfy variety of regulations such as the ISO Norms, EU AI Act and GDPR
Compliance norms are often complicated, and adhering to them requires tracking dozens of requirements hidden within hundreds of pages of legal acts, often locked behind paywalls. ControlAI helps you understand what is expected from you to fulfil a specific act by streamlining the process. You no longer have to read through the detailed documentation! Just fill in a short set of questionnaires and generate requirements specific for your use case quickly.
Track your progress towards achieving compliance and security
Once you understand what needs to be done to improve your compliance, you can track the progress towards satisfying the norms through our platform. Same goes for cybersecurity. You can also generate reports to track it.
Deploy verified defenses to protect your stack and monitor incidents
Identifying risks is one thing, another is mitigating them. We offer you a directory of defenses, some made by ourselves, some made by our partners or open-source communities to enable you to patch the vulnerabilities you’ve identified.
Showcase your AI safety solutions for other companies to use
We understand that protecting LLMs is a daunting task that can only be fulfilled by a coordinated effort of the entire AI safety community. That is why we offer a chance to other AI auditing and defense tool makers to showcase their products on our platform. Those would appear in the “Verified Tools” directory. Similarly, if you are a policymaker aiming to speed the adoption of your requirements, let us know and we will add them to the platform.
Hope that was helpful! If the current version of the platform does not answer your needs, let us know too! Your feedback is critical for the further development of the platform.
ControlAI is the work of a solo founder, indie hacking my way to success. If you want to work with me in any capacity, reach out at any of the below and let’s continue this journey together!
Links:
Meet with us: https://calendly.com/lbartoszcze
Website: controlai.org
LinkedIn: https://www.linkedin.com/company/controlai
Twitter: https://twitter.com/controlyourai