Anthropic just made a major move that’s got the entire tech industry paying attention. After it leaked in late March that the company had developed Mythos, a seriously powerful new AI model, Anthropic officially announced it this week alongside something even bigger: Project Glasswing, an industry-wide consortium designed to tackle the cybersecurity nightmare that advanced AI is about to create.

Here’s the wild part, Anthropic got its biggest competitors to join forces. We’re talking Microsoft, Apple, Google, Amazon Web Services, Cisco, Nvidia, and over 40 other tech and infrastructure companies all in one room (virtually) to figure out how to keep AI from becoming a hacker’s dream tool. The consortium has private access to Mythos Preview, which isn’t being released to the general public yet.

The reasoning is pretty straightforward: give all these companies time to test Mythos on their own systems and patch vulnerabilities before the tech becomes widely available. But the bigger picture is way more urgent. Anthropic is essentially saying that AI capabilities are at an inflection point. The security practices we’ve relied on for decades? They’re about to become obsolete.

Logan Graham, Anthropic’s frontier red team lead, puts it bluntly: “We need to prepare now for a world where these capabilities are broadly available in 6, 12, 24 months”. In other words, don’t assume current security models will work in the near future.

Mythos Preview can do things that would normally take a senior security researcher weeks to accomplish. It can find vulnerabilities in code, develop exploits, run penetration tests, and even evaluate software without access to its source code. The model wasn’t specifically trained to be a cybersecurity expert, it just became one as a side effect of being really good at code. Anthropic CEO Dario Amodei acknowledged in the launch video that “more powerful models are going to come from us and from others”, so they needed a plan to handle it responsibly.

This is where coordinated vulnerability disclosure comes in. It’s the same process used when researchers find security flaws: give companies time to fix them before going public. The difference? This time we’re talking about AI capabilities that could make large-scale attacks easier and cheaper than ever before.

So far, Mythos Preview has already uncovered thousands of critical vulnerabilities during testing, including some bugs that have existed for decades and somehow slipped past scrutiny. Microsoft’s global CISO Igor Tsyganskiy called it an “unprecedented opportunity to use AI responsibly to improve security at scale”.

The real challenge now is making sure this collaboration actually works. Graham says Project Glasswing will fail if it stays limited to a handful of companies. It needs to grow into something much larger. The cybersecurity community is literally trying to outrun the technology they created, and whether they succeed matters for all of us.