PEGAPOLL · NEWS

Researchers used Anthropic's Claude to hack OpenAI systems in under 72 hours

The Decoder (AI) · 2026-09-18
🤖 AI-generated content — The title and summary were produced automatically by artificial intelligence, without human editorial review.

Three Hacktron researchers say Anthropic's Claude Opus 5 helped them breach OpenAI via its community forum in under 72 hours. They chained a libheif HEIC image flaw with an OAuth-style “Sign in with OpenAI” weakness to reach employee ChatGPT/Codex accounts and OpenAI’s GitHub monorepo, proving access with a harmless pull request while claiming no sensitive data was viewed.

Continue in the app — vote & join in ➔
Source: The Decoder (AI) · via ahirlevel.hu