Anthropic launches Project Glasswing: Ai for computer security

This is where the issue becomes interesting. Anthropic points out that Mythos was not trained specifically for cybersecurity. His skills come from two general muscles: agentic coding and reasoning. In practice, he is not a dog trained only to search for explosives: he is more like an investigator capable of reading files, connecting clues and then moving around the crime scene.

This is why Project Glasswing is not just an industrial announcement. It is also a political statement about the future of AI. Anthropic openly says it has discussed with US government officials the defensive and offensive capabilities of the model. And it links the issue of critical infrastructure protection to the much larger issue of technological competition between democracies and geopolitical rivals.

The message is simple: these capabilities will spread anyway. The question is not whether they will arrive. The question is who will use them first and with what rules.

The partners' statements all point in the same direction. Microsoft speaks of cybersecurity no longer being limited by human capability alone. Cisco says that AI has crossed a threshold that changes the level of urgency. The Linux Foundation puts its finger on the sore spot: until now, security has often been a luxury for those with large dedicated teams, while open source maintainers, who hold up much of the world's software, have been left to fend for themselves.

And this is perhaps the most concrete point of the project. Today, a huge share of modern systems rest on open source components. They are the invisible bricks of digital. If those bricks have cracks, the skyscraper wobbles. Project Glasswing wants to put an automatic magnifying glass in every maintainer's hand. Or, to use the Linux Foundation's formula, a reliable 'sidekick'.

It remains to be seen how much of this promise will be translated into practice. The most powerful models are also dual-purpose tools. They can find flaws to close doors, or to open them. Anthropic tries to play it safe and chooses a selective path: Mythos Preview will not be made generally available, but only to verified partners and organisations.

After all, the name of the project already says a lot. The glasswing is a butterfly with transparent wings. A beautiful image, but above all a useful metaphor: making the hidden weaknesses of software visible before the attackers do.

In the cybersecurity of 2026, the problem is no longer just building walls. It's about finding the cracks before the others do. And maybe doing it with a machine that never sleeps.