Ensuring advanced AI is trustworthy, secure and beneficial for everyone.

FAR.AI is organizing the first European edition of the Technical Innovations for AI Policy (TIAP) workshop, connecting EU policymakers with leading independent technical experts six months after the EU AI Office gains enforcement powers under the AI Act.

Training against white-box deception detectors in a realistic coding environment reveals two obfuscation strategies models can develop: modifying internal representations to evade the detector, or producing deceptive text with justifications that bypass it. However, sufficiently strong KL regularization combined with a detector penalty can suppress both, validating deception detectors as viable training signals against reward hacking.

We tested the effectiveness of "defense-in-depth" AI safety strategies, where multiple layers of filters are used to prevent AI models from generating harmful content. Our a new attack method, STACK, bypasses defenses layer-by-layer and achieved a 71% success rate on catastrophic risk scenarios where conventional attacks achieved 0% success against these multi-layered defenses.

GPT-4o was as effective at increasing conspiracy beliefs ("bunking") as decreasing them ("debunking"), and OpenAI's guardrails did little to prevent this.