tldrdailynews.com logo

Opus 5 reveals AI-driven breach route into OpenAI via image library bug

1 min read
Source: The New Stack
Opus 5 reveals AI-driven breach route into OpenAI via image library bug
Photo: The New Stack
TL;DR Summary

Hackers used Claude Opus 5 to chain a memory-corruption flaw in the libheif HEIC处理 library on a Discourse forum, achieving remote code execution and then leveraging over-privileged sign-on tokens and a Codex-linked OpenAI account to access OpenAI’s private monorepo—exposing a multi-step, AI-enabled attack path that OpenAI quickly mitigated by tightening tokens and paying a bounty; the incident underscores how AI agents can automate complex security exploits, finishing the sequence in under 72 hours.

Topics:business#ai-security#anthropic#claude#libheif#openai#technology
Share this article
  1. Claude couldn’t hack OpenAI. Then Anthropic shipped Opus 5.  The New Stack
  2. Exclusive | Hackers Used Anthropic’s Claude to Break Into OpenAI  WSJ
  3. Hackers breached OpenAI, adding to fever pitch of security and safety concerns  NBC News
  4. OpenAI ‘ethically hacked’ with help of Anthropic’s Claude chatbot  The Guardian
  5. Researchers used Claude to hack OpenAI  Ars Technica

Reading Insights

Total Reads

0

Unique Readers

6

Time Saved

71 min

vs 72 min read

Condensed

99%

14,214 → 75 words

Want the full story? Read the original article

Read on The New Stack