Anthropic is limiting access to its powerful Mythos AI model over safety concerns, even as the company seeks to advance its public profile and repair ties with the Trump administration.

What happened: Anthropic teased Mythos last month as too dangerous to launch publicly. The company has instead granted access to a few dozen organizations for testing and preview, while CEO Dario Amodei curtailed a European trip to focus on the model's rollout.

Why it matters: Mythos is described by staff as more capable than Anthropic's earlier models, but the company worries it could be weaponized for cyberattacks. The cautious rollout reflects Anthropic's effort to balance its safety-focused image with pressure to advance autonomous AI and strengthen relationships with policymakers.

What to watch: Amodei is managing a delicate juggling act to bring Mythos to the public while maintaining safety guardrails. The model's eventual wider availability, and how it performs under real-world use, will test whether Anthropic can deliver powerful AI without compromising its stated safety standards.