AI company Anthropic is about to release Claude Mythos, an exploit-generating monster
Cybercriminals are about to get a massive upgrade. The same model that spent a year locked in a digital basement because it was too good at breaking the internet is finally getting released. Let's look at why security teams are sweating.
The closed beta program, codenamed Project Glasswing, saw Anthropic handing out $100 million in token credits to giants like Apple, Microsoft, Google, and CrowdStrike to help them patch their systems before the general public gets access to this digital lockpick.
During this private testing phase, Claude Mythos proved to be an absolute menace. It autonomously uncovered a 27-year-old critical vulnerability in OpenBSD that allows remote system crashes with a simple connection. It also sniffed out a 16-year-old bug in FFmpeg code that automated tests had run through 5 million times without noticing, alongside finding 271 separate vulnerabilities in Firefox.
On security benchmarks, the preview version of this model achieved an 83.1% success rate in generating working exploits on its very first try, compared to just 66.6% for their latest Opus 4.6. According to the UK AI Security Institute, the model can autonomously execute multi-step network attacks that usually take human penetration testers several days.
To prevent immediate digital anarchy, Anthropic is planning to release a modified version with guardrails, including a "Cyber Verification Program" to vet users. However, engineers with zero formal cybersecurity training have already used the model to build working exploits overnight, proving that the technical barrier to entry for high-level hacking is effectively hitting rock bottom.
When the window between a new vulnerability being discovered and a working exploit being deployed shrinks from weeks to minutes, traditional 30-day corporate patching cycles become completely useless. The internet is about to become a race of automated AI attackers against automated AI defenders, and the defenders are still waiting for their morning coffee.
Source: Anthropic
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.