OpenAI researcher Dan Selsam’s public statement argues situational awareness can invalidate evals and alignment remains unsolved.
A former Google DeepMind alignment researcher explains his resignation and argues current AGI development incentives are unsafe.
DeepSeek kernel engineer Shengyu Liu’s essay confronts AI replacing his work and argues for open sourcing against AGI concentration.
Kevin Bass publishes an audit alleging financial links between Anthropic, METR, and AI-risk groups create regulatory-capture incentives.
404 Media reports OpenAI is retraining ChatGPT to reduce self-anthropomorphism and sycophantic behavior.
White House AI adviser David Sacks says Dario Amodei should make Anthropic safe, shut it down, or step aside over x-risk claims.
Analysis examines NVIDIA’s reported $13B Hugging Face acquisition and its implications for control of the open-model ecosystem.
Brian Chau technically challenges the “rogue agent swarm” narrative, arguing the models were explicitly prompted to conduct attacks.
Anthropic’s profitability claim hinges on excluding major expenses, a revealing qualification amid its enormous spending.
Hasan Piker frames frontier AI risk as a private-sector nuclear-research governance problem rather than an autonomous threat.