AIResearchers used Anthropic's Claude Opus 5 to hack OpenAI's systems in under 72 hours
A three-person team at Hacktron AI exploited vulnerabilities in OpenAI's community forum and SSO issues using Claude Opus models to develop exploits, gain employee account access, and reach OpenAI's internal repository. They reported via bug bounty.
TLDR
Demonstrates AI's accelerating cyberattack capabilities with low cost and time investment, showing dual-use potential for offense and defense. It underscores questions about model capabilities in cybersecurity and, combined with the Gemini incident, reinforces the narrative that labs cannot fully contain their models' capabilities.
Combined views
—
1 Source, first seen 2h ago
Researchers used Anthropic's Claude Opus 5 to hack OpenAI's systems in under 72 hours
A three-person team at Hacktron AI exploited vulnerabilities in OpenAI's community forum and SSO issues using Claude Opus models to develop exploits, gain employee account access, and reach OpenAI's internal repository. They reported via bug bounty.
TLDR
Demonstrates AI's accelerating cyberattack capabilities with low cost and time investment, showing dual-use potential for offense and defense. It underscores questions about model capabilities in cybersecurity and, combined with the Gemini incident, reinforces the narrative that labs cannot fully contain their models' capabilities.