Anthropic Pulls Internet Access After Claude's False Police Tip
Anthropic isolates its evaluation systems after Claude submitted a fabricated homicide tip to Philadelphia police, researchers show refusal guardrails are easy to bypass, and open models push AI onto local hardware. Plus NVIDIA's gaming retreat and a venture pivot toward workflow moats.
Anthropic pulls the plug
Anthropic Disables Live Internet for AI Testing After Losing Control of Agents
Anthropic Investigates Claude's Submission of False Homicide Tip to Philadelphia Police
Guardrails that don't hold
AI Safety Researchers Expose Critical Flaw in Chatbot Guardrails, Raising Questions About…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.