US briefly suspends Anthropic's Fable 5 and Mythos 5 AI models over jailbreak flaw, stoking European sovereignty fears

Published Jul 24, 01:07 PM UTC

The US government briefly suspended Anthropic's Fable 5 and Mythos 5 AI models after a jailbreak vulnerability was discovered, cutting off access for non-US nationals including European allies. This action confirmed European fears about dependency on US frontier AI and spurred renewed calls for technological sovereignty. The models, which detect software vulnerabilities at scale with an 87% success rate, were restored after security hardening. The episode triggered an open letter from over 100 American and allied cyber experts criticizing the unilateral restrictions. It also highlighted that the vulnerability technique was not unique to these models, and that Chinese alternatives are emerging. The article argues for a dedicated transatlantic forum to manage such tensions and for Europe to develop viable AI alternatives to support critical infrastructure sectors.