Some few days ago I wrote about Claude Mythos and whether it would only be affordable for the rich. I ended that post saying the future is already here, it is just not evenly distributed. I did not expect Anthropic to prove that point quite this dramatically, quite this fast.
This week they announced Project Glasswing. And the access question just got a lot more layered.
What Is Project Glasswing?
Project Glasswing is Anthropic’s new cybersecurity initiative, announced on April 7, 2026. It brings together twelve of the biggest names in tech and finance, including Amazon Web Services, Apple, Google, Microsoft, Cisco, CrowdStrike, JPMorganChase, NVIDIA, and Palo Alto Networks, with one shared goal: use Claude Mythos to find and fix dangerous security flaws in the world’s most critical software before the wrong people find them first.
The model at the centre of this is Claude Mythos Preview, the same unreleased model I wrote about last week. It is still not available to the public. But it is now very much active, just in a very controlled and deliberate way.
Beyond the twelve launch partners, Anthropic has extended access to roughly 40 additional organisations that build or maintain critical software infrastructure. These organisations can use Mythos Preview to scan and secure both their own systems and open source code that much of the world relies on daily. Banking systems, medical records platforms, power grid software, logistics networks. The kind of infrastructure most of us never think about until it stops working.
Anthropic is backing the initiative with up to $100 million in model usage credits and $4 million in direct donations to open source security organisations, including $2.5 million to the Linux Foundation and $1.5 million to the Apache Software Foundation. This is not a pilot programme or a press release with no substance behind it. The scale of the commitment suggests they are treating this as genuinely urgent.
A Quick Note on What Mythos Actually Is
One thing worth clarifying before we go further. Claude Mythos is not a cybersecurity-only model. It is a general-purpose frontier AI with strong coding, reasoning, and agentic capabilities across the board. The cybersecurity angle is not what it was built for. It is simply what happened when you build something that can read, reason about, and write code at a level that surpasses most humans. The security implications are a byproduct of how capable it is generally.
Beyond that, the full picture of what Mythos can do is still emerging. Benchmark data, real-world performance comparisons, pricing details for broader access. Most of it has not been confirmed publicly yet. Mythos is currently pretty much as mysterious as a myth. There is a lot more to unpack, and I will be covering it as the details surface.
Here Is Where It Gets Uncomfortable
Before Anthropic made any public announcement, they were already in private discussions with US government officials. Not to show off. To warn them.
Their concern was specific. Claude Mythos is so capable at finding and exploiting software vulnerabilities that releasing it without proper safeguards in place could make large-scale cyberattacks significantly more likely this year. Anthropic said so directly in their announcement. The model’s capabilities, in the wrong hands, could be weaponised. And given how fast AI is moving, those wrong hands are not far behind.
That is a remarkable thing for a company to say about its own product. It is also, if you sit with it for a moment, a little unsettling. We are used to tech companies hyping their releases. We are less used to them briefing governments before they tell the public, and framing their own creation as something that needs to be handled carefully before it touches the open internet.
What Has Mythos Actually Done?
In the weeks before the public announcement, Anthropic quietly put Mythos to work. The results were striking.
The model identified thousands of previously unknown security flaws, what security researchers call zero-day vulnerabilities, across every major operating system and every major web browser. Some of these flaws had been sitting undetected for decades. One vulnerability in OpenBSD, an operating system known for being one of the most security-hardened in the world, had gone unnoticed for 27 years. Another in FFmpeg, a widely used video processing library, had been missed by automated testing tools that had scanned the same line of code five million times without catching the problem.
Mythos also found and chained together several vulnerabilities in the Linux kernel, the software that runs most of the world’s servers, in a way that would allow an attacker to escalate from ordinary user access to complete control of the machine.
It found all of this autonomously. Without human steering.
On the CyberGym benchmark, which tests AI models on cybersecurity vulnerability reproduction, Mythos Preview scored 83.1%. The next best Anthropic model, Claude Opus 4.6, scored 66.6%. That gap is not marginal. It represents a meaningful leap in what AI can do when pointed at security research.
All the vulnerabilities Mythos has identified so far have been reported to the relevant software maintainers and patched. But the point is not just that it fixed things. The point is that it found things no one and nothing else could.
So Who Actually Has Access?
Twelve major launch partners are using Mythos Preview as part of their defensive security work. Forty additional organisations focused on critical software infrastructure have also been granted access. After the current research preview period, participants will be able to access the model at $25 per million input tokens and $125 per million output tokens through the Claude API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry.
For everyone outside that group, including you and me, Mythos remains out of reach. And based on everything we now know, that is unlikely to change quickly.
What This Means for the Rest of Us
In my post last week I focused on pricing as the barrier. Mythos would be expensive, and expensive tools compound advantages for those who can already afford them. That concern still stands.
But Project Glasswing adds a second layer. Even if Anthropic wanted to release Mythos broadly tomorrow, they are saying they should not. The model is capable enough that putting it in the open, before the world’s most important systems are patched and protected, carries real risk. The access restriction is not just commercial this time. It is precautionary.
That is a different kind of gap. Not just who can afford the best tools, but who gets to decide when those tools are safe enough to share at all. And right now, that decision sits with Anthropic, a handful of governments, and twelve of the largest technology companies on the planet.
For everyday people trying to use AI to learn faster, work smarter, or build something, the practical reality has not changed much since last week. Mythos is not coming to a $20 plan anytime soon. The models available to you right now, Claude Sonnet, the free tier, the tools already in your browser, are genuinely powerful. Most people are still not using them anywhere near their full potential. That gap, between average use and intentional use of today’s tools, is worth more of your attention than the gap between your current plan and a model you cannot access anyway.
Frequently Asked Questions
What is Project Glasswing and why does it matter?
Project Glasswing is Anthropic’s initiative to use its most advanced AI model, Claude Mythos Preview, to find and fix critical security vulnerabilities in the world’s most important software. It matters because it signals that AI has reached a level of capability in cybersecurity that requires careful, coordinated handling rather than open release. The implications for how powerful AI tools get distributed, and to whom, are significant.
Will Claude Mythos ever be available to the public?
Anthropic has said they do not plan to make Claude Mythos Preview generally available in its current form. However, they have indicated they want to safely deploy Mythos-class models at scale once new safeguards are in place. No timeline has been given. Based on how previous models have evolved, a version of these capabilities will likely reach consumer plans eventually, but probably not before meaningful safety infrastructure is built around it.
Is the software I use every day actually at risk?
The short answer is yes, in the sense that most software contains vulnerabilities, many of them old and undiscovered. What Project Glasswing is trying to do is find and patch those vulnerabilities before bad actors with access to similarly powerful AI can exploit them. The fact that Mythos found critical flaws in every major operating system and browser is not a reason to panic. It is a reason to take software security seriously, and to appreciate that this kind of work is happening at all.
The Bigger Picture
I ended last week’s post with the line: the future is already here, it is just not evenly distributed. Project Glasswing is a vivid illustration of what that actually looks like in practice.
The most capable AI model ever built is currently working behind closed doors, with governments briefed, major corporations looped in, and the general public kept at arm’s length for reasons that are, honestly, legitimate. The people steering this process are not villains. The caution appears genuine.
But it is still worth paying attention to. Because the decisions being made right now, about who gets access, when, and under what conditions, will shape how this technology eventually reaches the rest of us.
Stay curious. Stay informed. And in the meantime, keep getting better at the tools already in front of you. By the time Mythos reaches a consumer plan, the people who mastered today’s tools will already be miles ahead.
Read the original post this article follows up on: Is Anthropic’s Most Powerful AI Only for the Rich?
Leave a Reply