Chinese Hackers Impersonate Anthropic Employee to Extract AI Secrets

Proofpoint has identified a threat actor it calls TA419 — a designation with all the menace of a suburban bus route — and says it impersonated an Anthropic employee in order to extract AI secrets. Chinese hackers, the Futurism headline calls them, though Proofpoint itself prefers the coy formulation that TA419 'will likely continue targeting think tanks and policy experts working on technologies, and in geographies, of particular interest to the Chinese government.' That is the security-industry way of naming a country without spelling it out and then taking a phone call from a lawyer. The whole report reads like a document composed by people who may one day have to testify about it.

Nobody needed a rogue agent, a kill switch or a containment failure. OpenAI is currently being subpoenaed over exactly those, while the intrusion at Anthropic was, apparently, purely social: a stranger with a plausible job title and a convincing email signature walking through the front door. All that compute, all that safety literature, and the weakest link remains a person who simply wants to be helpful to a new colleague. The frontier labs have built fortresses and are being robbed at the reception desk.

And the timing, as ever, is impeccable. The report lands amid the great regulation debate, where the White House, Mark Zuckerberg and Jensen Huang are pushing back against calls to regulate, while Dario Amodei calls for some regulations — the good kind, presumably, the sort that apply chiefly to other people's labs. So the state of play: a group named after a spreadsheet cell is burgling the policy shops, the industry cannot agree whether regulation is an existential threat or a branding problem, and the most advanced technology in human history is being undone by someone claiming to be from the same company. Enjoy the debate.