Anthropic expands Project Glasswing to 150 organizations

Anthropic's "defenders-first" release strategy is a pragmatic paradigm shift in AI safety that moves past the futile goal of delaying model capability, opting instead to build an asymmetric defensive advantage by putting cutting-edge tools in trusted hands early.

• –Racing Claude Mythos Preview to 150 critical infrastructure entities creates a crowd-sourced, defensive red-teaming force at an unprecedented scale.
• –Shifting safety from "capability withholding" to "asymmetric defense" recognizes that adversary models will inevitably catch up, making proactive patching the only viable long-term strategy.
• –Targeting sectors like power, water, and healthcare highlights the high-stakes reality of securing foundational software codebases that billions rely on.