Anthropic scrambles after Trump administration freezes its top AI models

The Trump administration abruptly ordered export controls on Anthropic’s latest AI models, Fable and Mythos, just days after their public release. The US Department of Commerce barred foreign nationals from accessing the models and directed Anthropic to suspend them for all users within 90 minutes, without prior detailed explanation. The decision followed a report by Amazon researchers identifying a potential ‘jailbreak’ in Fable, allowing bypass of security guardrails to expose software vulnerabilities. Amazon, which has invested $13 billion in Anthropic, raised broader concerns about frontier-model risks with US officials but declined to disclose specifics. Anthropic disputes the targeting, claiming similar vulnerabilities exist in competing systems, including those from OpenAI. The company’s CEO, Dario Amodei, has previously warned about AI risks, but the sudden freeze has raised questions about government oversight of advanced AI. Security experts note that fixing such issues is complex and time-consuming, with no guarantees against future vulnerabilities. The intervention marks an early test of how the Trump administration plans to regulate AI, with industry leaders warning that export controls could create uncertainty. Anthropic has dispatched a technical team to Washington to address the fallout, while officials and the company work to resolve the dispute. The freeze does not apply to OpenAI’s models, despite comparable capabilities, highlighting inconsistencies in enforcement. Amazon confirmed it advises governments on security risks but avoids sharing details of such discussions. The Fable model had undergone internal approvals before its release, adding to concerns about the sudden regulatory action. Experts, including former OpenAI board member Helen Toner, argue that jailbreaks are inherent to advanced AI and cannot be fully eliminated, suggesting broader challenges in governance.