E2273 - Anthropic’s Mythos is a cyber-weapon, so you can’t have it

5-minute masterclass

The corrected narrative is that Anthropic's Mythos sits inside Project Glasswing. That framing matters because it moves the discussion from product launch theater to strategic containment. If the model meaningfully amplifies offensive cyber capability, release strategy becomes a security decision.

Rob May's contribution is the more durable business lesson. He argues that inference-time compute is the new moat. That shifts the battleground away from pure training spectacle and toward the systems layer: routing, specialization, orchestration, and cost-efficient execution at inference time.

The Formula

• Dangerous capability changes go-to-market into access control
• Inference-time compute can become a stronger moat than raw training scale in applied AI

Featured Entities