Anthropic has revealed a preview of a new frontier AI model called Mythos, and it’s not being pitched as just another chatbot upgrade. Instead, the company is tying it to a new cybersecurity initiative called Project Glasswing, aimed at helping defend critical software and infrastructure. TechCrunch reports that Mythos is one of Anthropic’s most powerful models yet, while Anthropic says the model is being deployed in a tightly controlled preview for defensive security work.
What Anthropic Is Actually Launching
According to Anthropic, Claude Mythos Preview is a general-purpose unreleased frontier model with unusually strong coding and reasoning capabilities. The company says it has already been used to find thousands of high-severity vulnerabilities, including issues in major operating systems and web browsers, and says many of those flaws had survived years or even decades of human review and automated testing.
Project Glasswing brings together 12 launch partners including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Anthropic says those partners will use Mythos Preview for defensive security work, while more than 40 additional organizations that maintain critical software infrastructure will also get access to the preview. Anthropic also says it is committing up to $100 million in usage credits and $4 million in direct donations to open-source security organisations.
Why This Model Is Different
The pitch here is not just that Mythos is smarter. It’s that Anthropic believes models have now crossed a threshold where they can outperform almost everyone except the most skilled human experts at finding and exploiting software vulnerabilities. On its Project Glasswing page, Anthropic says Mythos Preview found vulnerabilities in every major operating system and every major web browser, and in some cases developed exploits with little or no human steering.
Anthropic’s examples are striking. The company says Mythos found a 27-year-old OpenBSD flaw, a 16-year-old FFmpeg bug, and chained together Linux kernel vulnerabilities to escalate from ordinary user access to full machine control. Anthropic says these vulnerabilities were reported and patched, and says other findings will be disclosed later once fixes are in place.
A Security Tool With Bigger Implications
This is where the story gets more interesting. Anthropic isn’t presenting Mythos as a public product launch. It’s presenting it as a warning that AI cyber capabilities are advancing fast enough to become a serious defensive and offensive issue at the same time. On its Glasswing page, the company argues that the same capabilities that can help defenders find and fix flaws could also be used by attackers if similar models spread without safeguards.
TechCrunch adds another layer to that tension. It notes that Mythos was referenced in a previously leaked internal document, where the model was reportedly described as larger and more intelligent than Anthropic’s Opus line, and potentially dangerous if weaponised by bad actors. The article also points out the awkward timing of Anthropic launching a security-heavy initiative after recent code exposure and repository takedown mishaps tied to Claude Code.
Why Anthropic Is Framing This Around Defense First
Anthropic’s message is clear: these capabilities are coming either way, so defenders need access before attackers gain the same edge. The company says the work of defending critical software may take years, while frontier AI capabilities could advance dramatically in just months. That urgency is the core logic behind Project Glasswing.
It also suggests Anthropic wants to shape the narrative around powerful cyber-capable models before broader access becomes possible. Rather than releasing Mythos widely, it’s using a partner-led security initiative to argue that frontier AI should be rolled out cautiously in areas where the upside for defence is high but the misuse risk is also obvious. That’s the strategic subtext of the launch, even if Anthropic is framing it in more technical language.
Why this matters for Australia
Cybersecurity isn’t an abstract overseas issue for Australian readers. Hospitals, banks, energy providers, government systems and businesses here all depend on the same kinds of software infrastructure that Anthropic says Mythos is being used to secure. If frontier AI really is becoming dramatically better at finding vulnerabilities, that has direct implications for Australian institutions too.
It also points to a bigger shift in the AI story. The focus is no longer just on chatbots, image generators or office productivity. Models are increasingly being framed as tools that could change the balance between cyber defenders and attackers, which raises the stakes far beyond consumer AI.
For readers, the takeaway is simple: Anthropic is treating Mythos less like a normal model release and more like an early warning about how fast AI-powered cybersecurity is moving.
Source TechCrunch | Anthropic Project Glasswing | Anthropic Newsroom
