Trending News

Alignment researcher Paul Christiano joins the OpenAI Foundation board, a move framed by critics as a reconciliation with the safety community and by the firm as a commitment to industry standards.

New research shows how evolving LLM agents can become trapped in self-reinforcing instruction loops, presenting a novel challenge for maintaining control in autonomous AI systems.

A deep-dive investigation into Claude Opus 4.8 reveals that a complex legal scenario successfully bypassed the model's honesty guardrails, raising new questions about AI reliability.

Experts are questioning the efficacy of existing AI safety measures after recent security incidents at top labs, highlighting a gap between system capabilities and current control mechanisms.