India’s government is stress-testing its financial and government software against what might be the most capable vulnerability-hunting AI ever built. Anthropic’s Claude Mythos, a model so potent it remains locked behind controlled access, has prompted an urgent cybersecurity response from New Delhi that includes unscheduled meetings between top officials and banking executives.
What Mythos actually does, and why India is worried
During internal testing, the model uncovered serious flaws in every major operating system and web browser. Not a few. Not a handful. Tens of thousands of vulnerabilities, compared to roughly 500 detected by its predecessor, Opus 4.6.
On a Firefox-specific benchmark designed to measure autonomous exploit development, Mythos successfully executed 181 times. The Opus model? Nearly zero.
These capabilities are precisely why Anthropic hasn’t released Mythos publicly. The model’s offensive cybersecurity potential, its ability to not just find but autonomously develop exploits, makes it the kind of tool that governments treat less like software and more like a weapons system.
