Key Points
- The U.S. cleared Anthropic's Mythos 5 model for use by "certain trusted partners"
- It reverses an unprecedented order that had forced Anthropic to disable the model
- Commerce Secretary Howard Lutnick announced the easing
- 100+ cybersecurity experts and Adobe and Nvidia leaders had urged the government to lift the curbs
The standoff between Anthropic and the Trump administration over its most powerful model is cooling. The U.S. has cleared Anthropic's Mythos 5 for use by "certain trusted partners," Commerce Secretary Howard Lutnick said — restoring access to a system the government had forced the company to switch off less than two weeks earlier.
What changed
Earlier in June, the administration ordered Anthropic to disable Fable 5 and Mythos 5 under an export-control directive — an unprecedented move driven by Mythos's ability to surpass human cybersecurity experts at finding and exploiting software vulnerabilities. Anthropic had already limited the high-end model to select customers for the same reason. The reversal followed pressure from industry: more than 100 cybersecurity experts and leaders from companies including Adobe and Nvidia urged the government to lift the curbs.
The read
The whiplash is the point. In the same stretch, Washington gated OpenAI's GPT-5.6 to roughly 20 approved partners and now eased its grip on Anthropic's flagship — improvising an AI-security policy in real time, model by model. The throughline is that the U.S. no longer treats frontier-model releases as private product decisions; "trusted partners" is becoming the unit of distribution, and the labs are negotiating access the way defense contractors negotiate clearances. For Anthropic — the lab that built its brand on safety — being both the first model disabled and the first restored is a preview of the regulatory back-and-forth every frontier release now runs. Anthropic's unreleased Mythos Preview system — a distinct research model, not Mythos 5 — surfaced again in July, when it was credited with finding new cryptographic weaknesses in HAWK and AES.
Source: Bloomberg, The Washington Post
Comments