Matthew Berman: 'I Can't Believe This Happened' — Anthropic's Mythos 'Too Dangerous to Release' vs. 20K Undetected Fraud Accounts
Matthew Berman (YouTube)·medium signal
Matthew Berman takes a critical swing at Anthropic's Mythos messaging: the model is framed as too dangerous to release for cyber, yet reportedly failed to detect ~20,000 fraudulent Chinese-linked accounts that racked up millions of interactions. The contradiction — powerful offensive capability used to justify lockdown while basic fraud detection slips — was also flagged by Steven Sinofsky as unaligned messaging, feeding a broader skepticism about safety-theater framing.