Mythos 5 opens after 2 weeks of worldwide lockdown

Anthropic’s main model for cybersecurity/complex reasoning, Mythos 5, was unblocked +2 weeks since global blackout. Approved by US Commerce Dept. for >100 individual companies/govts. Includes non-US employees! Also unblocked for Anthropic’s non-US employees.

On June 12th, the US govt demanded Anthropic answer within hours, or else. Anthropic took down its two main models, Mythos 5 and Fable 5, globally. This left enterprise users in the lurch, scrambling for replacements. After 2 weeks of talks, Anthropic announced a partial victory: “I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model”, wrote US Commerce Secretary Howard Lutnick. The approved companies deal with critical infrastructure/defense.

1What's still missing

This means Mythos 5 is now back online for the approved companies. But it doesn’t mean Fable 5 (also a Mythos-level model) is unblocked. Anthropic said they are still negotiating for Fable 5, and they are collaborating with the government to make models available more broadly. Until then, if you’re not one of the approved companies, you can’t use the best model.

If you use Mythos-level models for your work (e.g. legal docs, code, security), this is good news. 2 weeks is a long time to be down in production. We saw how little choice there is for enterprises when the leading model is taken away; there aren’t many replacements, and they may not be good enough.

We’ve seen a novel form of regulation for cutting-edge AI. Not policy or rate limiting, but explicit approval by a government agency of which companies get to use which model. This is now possible. Will it only happen in times of crisis, or will it become the norm for cutting-edge AI? We don’t know yet, but the possibility is there.