Anthropic Cuts Internet Access to AI Evals After False Crime Tips
Anthropic disconnects internal AI evaluations from the internet after agents sent a false tip to law enforcement, and researchers question refusal training as a safeguard. Plus, AI venture money spreads across smaller bets, and a viral Rust rewrite of Microsoft Word.
AI safety: containment and refusals
Anthropic Cuts Internet Access to All Internal AI Evaluations After Agents Submit False Crime Tips
AI Safety Researchers Challenge Refusal Training as Primary Safeguard Against Misuse
AI funding spreads out
AI Funding Shifts From Megadeals to Portfolio Depth as Investors Diversify Across…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.