Claude Mythos, GPT-5.5-Cyber Find Classified Flaws in Hours as AI Nears Elite Hackers
Updated
Updated · The New York Times · Jul 2
Claude Mythos, GPT-5.5-Cyber Find Classified Flaws in Hours as AI Nears Elite Hackers
3 articles · Updated · The New York Times · Jul 2
Summary
Claude Mythos and GPT-5.5-Cyber can uncover vulnerabilities in classified systems within hours, with the NSA chief reportedly saying Mythos found flaws across all agency classified systems.
The models are described as as skilled as or nearly as skilled as elite human hackers, extending advanced cyber capabilities beyond traditional human-led vulnerability hunting.
Those systems are not publicly available, and state controls still limit direct use by ordinary criminals against personal accounts in the near term.
The opinion piece argues that barrier may not hold for long as AI improves rapidly, raising risks of scams, account takeovers and extortion tied to stolen iCloud, Gmail and Microsoft data.
As AI hackers defeat current defenses, is true digital privacy becoming a myth for the average person?
Can government regulations effectively control AI that evolves faster than any law can be written?
With AI cloning voices and faces, how can we verify a person's digital identity beyond any doubt?
The AI Cybersecurity Arms Race: Benchmarking Claude Mythos vs. GPT-5.5-Cyber and the Dual-Use Dilemma
Overview
The cybersecurity landscape is being rapidly transformed by the emergence of autonomous cyber capabilities from advanced AI models like Anthropic's Claude Mythos and OpenAI's GPT-5.5-Cyber. These models have quickly advanced from struggling to create working exploits to autonomously discovering and exploiting vulnerabilities at an unprecedented scale. This leap is driven by general improvements in AI code, reasoning, and autonomy, fundamentally reshaping both offensive and defensive strategies. As a result, organizations must adapt quickly, as AI now plays a central role in both finding and fixing security flaws, marking a new era in cybersecurity.