Schneier on Security

Bruce Schneier offers quick observations on Anthropic's Claude Mythos Preview and Project Glasswing, an initiative to proactively find and patch vulnerabilities before the model's cyberattack capabilities can be exploited publicly. He notes the announcement is partly a PR play, but acknowledges the models do show real advances in autonomous exploit writing and vulnerability chaining. Crucially, a security firm replicated Anthropic's findings using older public models, suggesting the threat is already present. Schneier warns that the defender currently holds a slight edge since AI finds vulnerabilities more easily than it operationalizes them, but that advantage will erode. He urges preparation for a world where zero-day exploits are abundant and attackers gain offensive capabilities far beyond their skill level.