AI model Claude Opus produced a functional Chrome exploit for just $2,283, highlighting how accessible AI can be for weaponising vulnerabilities. The article cites a breakdown showing Claude Opus 4.6 (high) costing $2,014 and Claude Opus 4.6 (high-thinking) at $267, with a small amount for Claude Sonnet / GPT-5.4 bringing the total to about $2,283 across 2,330M tokens and 1,765 requests.
It notes that Anthropic held back Mythos, but earlier, widely available models could already generate real attack code, making the risk tangible rather than theoretical. A quoted expert, according to Hacktron, Mohan Pedhapati, says he asked Claude Opus to build a full V8 exploit chain targeting Chrome 138, nine major versions behind upstream, and that the process involved about 20 hours of guidance.
The piece discusses patch gaps in Electron apps that bundle their own Chromium versions, and warns that as AI tools speed up exploit development, patching often lags behind, increasing real-world attack opportunities. It concludes that even as opinions on Mythos vary, AI-driven bug weaponisation is already a live concern.