Anthropic widened access to its Mythos AI, adding roughly 150 new organisations and pushing the total user base to around 200.
The company warned that Mythos can identify software flaws so effectively it could aid attackers in stealing data or sabotaging infrastructure. Despite that warning, Anthropic opened the model to the new partners in early June, spanning 15 countries. The expansion brings the user count to a figure the firm calls "controlled" rather than public.
The move spotlights the trade‑off between responsible disclosure and practical research. Security teams gain a powerful tool for hardening systems, but the same capability could be misused if it leaks beyond vetted partners. Anthropic hopes the limited rollout reins in abuse while still advancing defensive research.
In short, the company is walking a tightrope: it brands Mythos as too dangerous for open release, yet it is quietly seeding it among a growing circle of trusted customers.