The State of AI Safety
Key insights from the latest AI Safety Index | Edition #305
TL;DR
- AI capabilities are advancing faster than governance mechanisms.
- Companies like Anthropic, OpenAI, and Google DeepMind lead in AI safety, while Meta improves and xAI deteriorates.
- European AI safety regulation is strong, but Mistral AI scored poorly on safety.
- Three companies (xAI, DeepSeek, Mistral) received failing grades for AI safety.
- The industry's shift towards military AI use is flagged as an emerging harm risk.
- Major companies have weakened or voided their safety pause pledges.
- Existential safety is the weakest domain industry-wide, with no company scoring above a C-.
- Public safety rhetoric from companies often contrasts with their commercial conduct and legislative stances.
- Published safety frameworks from companies lack substantial enforcement mechanisms.