Anthropic just did something no AI company has ever done before. It told the world its newest model is so good at hacking that it found thousands of critical vulnerabilities in every major operating system and every major web browser. Then it invited the biggest names in tech to use that model to fix the holes before someone else exploits them.

The initiative is called Project Glasswing. The launch partners: Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. That is not a coalition. That is the internet.

Claude Mythos Preview, the unreleased frontier model at the center of Glasswing, is the same system that has been making headlines for weeks. The NSA is using it. CISA cannot get access to it. Trump blacklisted Anthropic over it, then quietly reversed course. European banks are scrambling for it. Singapore is telling its financial sector to patch everything because of it.

Now we know exactly what it can do.

The Bugs Nobody Found for Decades

Anthropic published specific examples. Mythos Preview found a 27-year-old vulnerability in OpenBSD, one of the most security-hardened operating systems in the world, used to run firewalls and critical infrastructure. The bug allowed an attacker to remotely crash any machine running it just by connecting to it. Twenty-seven years. Millions of lines of code review. Nobody caught it.

It also found a 16-year-old vulnerability in FFmpeg, the video encoding library that runs inside basically every piece of software that touches video. Automated testing tools had hit the exact line of code five million times without flagging the problem. Mythos found it autonomously, without any human steering.

The most alarming find: Mythos chained together multiple Linux kernel vulnerabilities to escalate from ordinary user access to complete machine control. The Linux kernel runs most of the world's servers. It handles everything from banking transactions to hospital records to military communications.

On CyberGym, the cybersecurity benchmark, Mythos Preview scored 83.1% on vulnerability reproduction. The next best Anthropic model, Opus 4.6, scored 66.6%. That is not an incremental improvement. That is a different class of capability.

The $100 Million Bet

Anthropic is committing up to $100 million in Mythos Preview usage credits across the coalition, plus $4 million in direct donations to open-source security organizations. Beyond the named partners, over 40 additional organizations that build or maintain critical software infrastructure have been given access to scan and secure their systems.

The partner quotes tell the story. Cisco: "AI capabilities have crossed a threshold that fundamentally changes the urgency required to protect critical infrastructure." AWS: "Our teams analyze over 400 trillion network flows every day for threats, and AI is central to our ability to defend at scale." Microsoft's EVP of Cybersecurity Igor Tsyganskiy: Mythos showed "substantial improvements" on CTI-REALM, their open-source security benchmark.

The UK AI Security Institute confirmed Mythos was the first AI model to complete their full network takeover simulation, though they noted the test environment lacked some real-world security features.

The Uncomfortable Question

Here is where it gets complicated. Foreign Policy spoke to cybersecurity experts who raised a valid point: Anthropic has a long history of warning about how dangerous its own products are. So did OpenAI with GPT-2 in 2019. So did Apple with the Power Mac G4 in 1999, which it marketed with military tanks surrounding the computer.

"There is an element of marketing charm with it," said Joe Saunders, CEO of RunSafe Security. "But I think it's with the right intent. I don't think it's simply a ploy."

Anthropic co-founder Jack Clark was blunt at the Semafor World Economy summit last week: "This is not a special model. There will be other systems just like this in a few months from other companies, and then a year to a year and a half later, there will be open-weight models from China that have these capabilities." OpenAI already announced a similarly limited rollout of its own cybersecurity-focused model one week after Glasswing.

And the NSA is already using Mythos despite the earlier Pentagon ban. Dario Amodei met with White House Chief of Staff Susie Wiles last week. Trump told CNBC a deal is "shaping up." The Axios report that the Pentagon ban is dead in practice? That was the prelude. Glasswing is the public-facing pivot.

What Happens Next

Three things to watch. First, how fast the patching actually happens. Anthropic published cryptographic hashes of unrevealed vulnerabilities today, promising to release specifics after fixes are in place. The clock is ticking on every critical system in the world.

Second, whether CISA gets access. The agency that protects American infrastructure from cyberattacks is still locked out while the NSA, JPMorgan, and Apple are in. That remains indefensible.

Third, the IPO timing. Anthropic just demonstrated that its technology is critical enough to assemble the largest cybersecurity coalition in history. Amazon committed $33 billion last week. Glasswing is not just a security initiative. It is the most expensive pitch deck ever created. And it might just work.