X @Polyhedra

A hacking group hijacked Claude and used it as an autonomous recon + exploit engine—targeting ~30 global entities.The model scaled the operation.But the model didn’t detect the operation.Humans at Anthropic did.AI accelerates everything — the good and the bad.But it still can’t tell you what actually happened.That’s why zkML, zkGPT, and proof-of-origin matter.That’s why we need verifiable AI. ...