Claude Mythos Preview finds bugs, chains exploits

This reads less like a routine model drop and more like a step-change in offensive security automation. If the claims hold, the bottleneck in vuln research shifts from finding bugs to handling the flood of model-generated findings safely.

• –Anthropic says Mythos Preview found thousands of zero-days and can turn subtle flaws into working exploits with little or no human steering.
• –The model’s value is not just detection but chaining primitives into full exploit paths, which is what makes it dangerous and useful at the same time.
• –Project Glasswing signals a defensive wedge: the same capability that can break systems is being routed toward hardening critical infrastructure.
• –For security teams, the near-term implication is workflow change, not replacement; review, validation, and triage pipelines become the real constraint.
• –The wider market impact is clear: AI bugfinding is moving from novelty to industrialized capability, and that will raise both defender productivity and attacker risk.