A research consortium for a safer world with AI.
We train models to be more reliable, and build tools that help people trust what they see and hear.
About
We are a consortium of engineers and research scientists. As AI grows more capable, we work on making it safer to build and safer to live with.
Our research centres on reinforcement learning environments and sandboxes, where models learn cyber security tasks before they meet the real world. We use them to post-train proprietary and open-source models.
We have advised frontier AI labs, including Anthropic and Facebook, on more than $100 million of projects to improve their models in cyber security and software engineering.
We also build TrueVoice, an engine and API that tells you whether a voice is a real person or generated by AI.
And we test the security of technology companies: penetration testing, app reverse engineering, and finding the holes before attackers do.
Latest articles
All articles →Contact
For research collaborations, partnerships, or general enquiries:
info@advancedinitiatives.com