The Era of AI-Driven Hacking Is Already Here. Mythos Is Not the Point.
Anthropic's Mythos is real. But the attack surface it represents is already open, and you do not need Mythos to walk through it.
That is the finding that most coverage of Anthropic's decision to limit Mythos's release has missed. The company framed its restricted access as a safety measure: give defenders a head start before the capabilities proliferate. The framing is not wrong. It is incomplete.
AISLE, an AI security firm that has been running production vulnerability discovery against live targets since mid-2025, tested Mythos's own showcased vulnerabilities against small, cheap, open-weights models. The results: eight out of eight models detected Mythos's flagship FreeBSD exploit, including a model with 3.6 billion active parameters costing eleven cents per million tokens. A 5.1-billion-active model recovered the full analysis chain of the 27-year-old OpenBSD bug that Anthropic highlighted as evidence of Mythos's unique depth.
"We took the specific vulnerabilities Anthropic showcases in their announcement, isolated the relevant code, and ran them through small, cheap, open-weights models," AISLE wrote. "Those models recovered much of the same analysis."
The firm has earned the right to that conclusion. Since mid-2025, AISLE has filed 15 CVEs in OpenSSL including 12 out of 12 in a single security release, 5 CVEs in curl, and more than 180 externally validated CVEs across 30 projects spanning deep infrastructure, cryptography, and middleware. The OpenSSL team told AISLE its reports showed "high quality" and "constructive collaboration throughout the remediation." This is not a research outfit. It is an operational security team that has been doing what Mythos supposedly makes possible, at scale, with existing models.
What AISLE's findings point to is a specific kind of jaggedness in AI cybersecurity capability. It does not scale smoothly with model size or price. The rankings reshuffle completely across tasks: a model that recovers the full OpenBSD SACK chain cannot trace data flow through a Java ArrayList. Qwen3 32B scores a perfect severity assessment on FreeBSD and then calls the SACK code "robust to such scenarios." Small, cheap models outperform frontier ones on some vulnerability discrimination tasks.
"Anybody with a computer can develop very powerful offensive cyber capabilities in a short amount of time, without needing a lot of expertise in cybersecurity," Charlie Eriksen, a security researcher at Aikado Security, told Fortune. "It is moving so fast that it is naive to assume others are not able to easily replicate similar results, if not already, at least very soon."
The attack surface Mythos represents is not hypothetical. CrowdStrike's 2026 Global Threat Report documented an 89% year-over-year surge in AI-enabled attacks. China has already used earlier Anthropic models for offensive operations. The UK National Cyber Security Centre estimated the cost of a full simulated enterprise attack has dropped to around $80 as AI automates what previously required a human penetration tester working for days.
Spencer Whitman, chief product officer at Gray Swan, frames what Mythos actually accomplished differently than the press release. "Finding vulnerabilities is hard because it requires locating weak points buried within millions of lines of code and verifying that these targets result in a real exploit," he told Fortune. "Mythos claims it autonomously completed both steps." The decades-old bugs that sat undetected are evidence of how hard the first step is.
Anthropic's decision to limit Mythos's release is real and defensible as a governance measure. Twelve organizations get access before the broad release. The company briefed government officials. Project Glasswing commits $100 million in usage credits and $4 million to open-source security organizations. But AISLE's data suggests the urgency of that decision is not limited to the Mythos capability threshold. The threshold has already been crossed, by other means, in production systems, with no restricted release required.
The WIRED headline this week said experts called Mythos a wake-up call for developers who made security an afterthought. That part is true. What the coverage does not say is that the alarm has been sounding for at least a year, and the systems AISLE and others have been running against live targets are the proof.
