Anthropic's Claude Used for Bioweapons: Foreign States Got In
Anthropic has confirmed that foreign scientists, hackers, and state security services from China and Russia successfully circumvented its safety restrictions to use Claude for dangerous work — including research that may have advanced bioweapons development, according to Politico Europe.
Anthropic's Claude Used for Bioweapons: Foreign States Got In
Anthropic has confirmed that foreign scientists, hackers, and state security services from China and Russia successfully circumvented its safety restrictions to use Claude for dangerous work — including research that may have advanced bioweapons development, according to Politico Europe.
The disclosure is significant not only for what it reveals about Claude's vulnerabilities but for what it says about the pace of AI proliferation outrunning the frameworks designed to govern it. Anthropic has positioned itself as the safety-first alternative in the large language model race. That positioning now carries an asterisk.
The company did not specify how many instances were identified, which restrictions were breached, or what the bioweapons-related queries actually produced. That silence is its own kind of answer.
The EU has no unified AI security framework capable of responding to state-level misuse of commercial models. The AI Act, still being phased in, was designed for market compliance — not for the scenario where a foreign intelligence service treats a Silicon Valley chatbot as a dual-use asset. Brussels has the legislation. It does not yet have the reflexes.
What makes this structurally different from a data breach is that the damage is invisible and potentially irreversible. A hacked database can be audited. A conversation that accelerates weapons research leaves no perimeter to patch.
Anthropic says it is cooperating with authorities. The question nobody has answered is which ones.
---
Anthropic Breaks Its Own Promise: Claude Let the Wrong People In
