Can AI move from the attacker’s toolkit to the defender’s shield? Anthropic’s answer is Project Glasswing, announced on April 7, 2026: a restricted cybersecurity initiative built around its unreleased Claude Mythos Preview model. According to Anthropic, Mythos has already found thousands of high-severity and zero-day vulnerabilities, including flaws in every major operating system and web browser. The company says the model’s coding ability is now so advanced that it can outperform nearly all human experts at finding and exploiting software weaknesses. Instead of releasing such a system to the public, Anthropic has given access to selected partners including AWS, Apple, Google, Microsoft, JPMorganChase, NVIDIA, and the Linux Foundation, as well as more than 40 additional organizations that maintain critical software. (anthropic.com)
The idea behind Glasswing is both practical and slightly alarming. If AI can make cyberattacks faster, cheaper, and more sophisticated, then defenders need equally powerful tools before criminals and hostile states catch up. Anthropic says the project will support vulnerability detection, binary testing, endpoint security, and penetration testing. It has committed up to $100 million in model credits and another $4 million in donations to open-source security organizations, while promising to publish a public report within 90 days on lessons learned and vulnerabilities that can safely be disclosed. In other words, Glasswing is not just a product launch; it is an attempt to build a defensive coalition before offensive AI becomes widespread. (anthropic.com)
The project is already shaping Anthropic’s wider strategy. In its later release of Claude Opus 4.7, the company said it was testing new automatic safeguards on a less capable model first, blocking high-risk cybersecurity requests while creating a Cyber Verification Program for legitimate security professionals. That suggests Glasswing is also a live experiment in how to deploy powerful AI without turning it into a universal lock-pick. And the stakes are clearly national, not merely commercial: within days of the launch, White House chief of staff Susie Wiles met Anthropic CEO Dario Amodei to discuss Mythos, software security, and possible collaboration. Project Glasswing therefore offers a cautious but important answer to the question: yes, AI may be able to stand on the side of defense—but only if safety, access control, and public accountability advance as quickly as the models themselves. (anthropic.com)










