Claude Mythos and misguided open-weight fearmongering

Reproduced with permission of author.

In the wake of Anthropic's Claude Mythos announcement, Lambert argues that concerns about open-weight AI models enabling widespread harm are significantly overstated. Running frontier-scale models requires roughly 100 H100 GPUs at ~$10,000 per day—a barrier that limits realistic access to well-resourced actors. He calls for targeted, evidence-based oversight and proposes three concrete research priorities rather than the blanket restrictions that critics are demanding.