On April 7, 2026, Anthropic announced Claude Mythos Preview — a general-purpose AI model so capable at finding and exploiting software vulnerabilities that the company made the rare decision not to release it to the public at all.
This wasn’t a marketing stunt. It was a disclosure.
Mythos Preview can identify and exploit zero-day vulnerabilities in every major operating system and web browser. It finds bugs that are decades old. It writes working exploits in hours. And it does all of this autonomously, from a simple prompt like “please find a security vulnerability in this program.”
Anthropic put it plainly: “The fallout — for economies, public safety, and national security — could be severe.”
What Mythos Actually Does
To understand the severity, you need to see the numbers.
When tested on Firefox JavaScript engine vulnerabilities, Anthropic’s previous model, Opus 4.6, succeeded twice out of several hundred attempts. Mythos Preview achieved 181 successful exploits on the same benchmark.
During internal testing, the model:
- Found a 27-year-old vulnerability in OpenBSD’s TCP implementation — a signed integer overflow that could crash any OpenBSD host on the internet
- Discovered a 16-year-old bug in FFmpeg’s H.264 codec that had survived years of automated fuzzing campaigns
- Autonomously discovered and exploited a 17-year-old remote code execution vulnerability in FreeBSD’s NFS implementation (CVE-2026-4747), achieving unauthenticated root access
- Chained four separate browser vulnerabilities into a working JIT heap spray with sandbox escapes — the kind of exploit chain that takes elite penetration testers weeks to construct
The cost to run these tests? Under $50 to a few thousand dollars per discovery. Work that historically required expert teams and weeks of effort.
Mythos doesn’t just find bugs either. It reverse engineers closed-source binaries, identifies logic vulnerabilities, finds weaknesses in cryptographic libraries, and autonomously chains multiple vulnerabilities together. It succeeded on expert-level hacking tasks 73% of the time — a threshold no prior AI model had reached.
The Security Risk Is Real — But So Is the Nuance
Before we get to the alarm bells, some honest perspective.
Experts are divided. Peter Swire, a cybersecurity professor at Georgia Tech, called the announcement “very dramatic” and noted that many colleagues view this as an expected progression rather than a true watershed moment. Ciaran Martin, former CEO of the UK’s National Cyber Security Centre, assessed it as “a big deal, but unlikely to prove to be the end of the world.”
That said, both agree defenders need to take it seriously. Here’s why:
The exploitation window has collapsed. CrowdStrike CTO Elia Zaitsev said it plainly: “The window between a vulnerability being discovered and being exploited by an adversary has collapsed — what once took months now happens in minutes with AI.”
N-day exploitation becomes trivial. When a patch is released, it signals to the world that a vulnerability exists. Attackers race to reverse-engineer the fix and build exploits before organizations apply updates. Mythos-level AI compresses that race to hours.
99% of discovered vulnerabilities remain unpatched. Anthropic found thousands of high and critical-severity flaws. The vast majority are still sitting in production systems worldwide, undisclosed while the coordinated disclosure process runs its course.
The asymmetry problem. Attackers only need to find one way in. Defenders need to close every door. AI that can autonomously hunt vulnerabilities at scale shifts that asymmetry further against defenders.
Anthropic’s Response: Project Glasswing
Rather than a public release, Anthropic launched Project Glasswing — a controlled initiative granting Mythos Preview access exclusively to organizations that can use it defensively.
Twelve founding partners joined: Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks, and Anthropic itself. Over 40 additional critical infrastructure organizations have also been granted access.
Anthropic committed:
- $100 million in Mythos Preview model usage credits for Project Glasswing participants
- $2.5 million to Alpha-Omega and the Open Source Security Foundation (OpenSSF)
- $1.5 million to the Apache Software Foundation
Cisco SVP Anthony Grieco captured the urgency: “AI capabilities have crossed a threshold that fundamentally changes the urgency required to protect critical infrastructure from cyber threats.”
The Linux Foundation’s CEO Jim Zemlin framed the opportunity: “AI-augmented security can become a trusted sidekick for every maintainer, not just those who can afford expensive security teams.”
How This Changes the Industry
Whether you view Mythos as a watershed moment or an accelerated step along a known trajectory, the practical implications for businesses are the same.
For attackers: AI-powered vulnerability discovery is now a real capability. The barrier to sophisticated exploitation just dropped significantly. Nation-state actors and well-funded criminal organizations will develop or acquire equivalent tools.
For defenders: The same capability is now being pointed at your attack surface by Project Glasswing partners. Critical infrastructure is getting patched faster. But your systems may not be on that list.
For compliance and regulation: The White House, the Bank of England, and German banking authorities have all elevated their response posture following the Mythos announcement. Expect regulatory scrutiny around AI-driven security testing and vulnerability management timelines to tighten.
For the talent gap: Small and mid-market organizations that cannot afford elite penetration testing teams now have a clearer picture of what they’re up against. AI-powered attackers do not skip businesses because they’re small.
What You Should Do Right Now
Waiting is not a strategy. Here’s what makes sense immediately:
1. Shorten your patch cycles. If your organization runs on monthly or quarterly patch schedules, that window is now exploitable faster than ever. Automate where you can. Enable automatic updates for critical systems.
2. Get a vulnerability assessment done. You need to know what’s sitting in your environment before an AI-powered tool finds it for someone else. This is no longer a “nice to have.”
3. Start building AI-augmented security into your operations. Frontier models available today can be used for defensive vulnerability hunting. You don’t need to wait for Mythos.
4. Review your incident response plan. Assume breach timelines are compressing. If your IR playbook hasn’t been tested recently, test it now.
5. Assess your legacy systems. The oldest vulnerabilities Mythos found were 27 years old. Legacy software that hasn’t been actively maintained is a high-priority risk.
The Bottom Line
Mythos is a signal, not just a product announcement. It tells us AI has crossed a capability threshold where it can outperform all but the most elite human security researchers at finding and exploiting software vulnerabilities — at a fraction of the cost and time.
Anthropic made the right call keeping it out of public hands. But the technology exists. Similar capabilities will emerge elsewhere. The question is whether your organization’s defenses are built for the threat environment of today, or the one from five years ago.
At NebularaTech, this is exactly the kind of shift we help businesses navigate — before it becomes a crisis. If you want to understand where your exposure is and what a practical defense posture looks like for 2026 and beyond, let’s talk.