TechResearchers Used Claude to Compromise OpenAI's Internal Systems in Under 72 Hours
Indian security researchers (Hacktron AI) leveraged Anthropic's Claude AI and a zero-day vulnerability in OpenAI's Discourse forum to access employee accounts, internal services, and OpenAI's monorepo. The operation cost under $3,000 in tokens within 72 hours as part of OpenAI's bug bounty program.
TLDR
The hack demonstrates real-world offensive capability of frontier AI models and raises urgent questions about AI-enabled cybersecurity risks, attack speed, and industry security postures. It amplifies AI safety concerns by showing models can effectively chain vulnerabilities and act as sophisticated agents.
Combined views
20
1 Source, first seen 1d ago
Researchers Used Claude to Compromise OpenAI's Internal Systems in Under 72 Hours
Indian security researchers (Hacktron AI) leveraged Anthropic's Claude AI and a zero-day vulnerability in OpenAI's Discourse forum to access employee accounts, internal services, and OpenAI's monorepo. The operation cost under $3,000 in tokens within 72 hours as part of OpenAI's bug bounty program.
TLDR
The hack demonstrates real-world offensive capability of frontier AI models and raises urgent questions about AI-enabled cybersecurity risks, attack speed, and industry security postures. It amplifies AI safety concerns by showing models can effectively chain vulnerabilities and act as sophisticated agents.