Home/Events/Security Researchers Use Claude to Hack into OpenAI

Security Researchers Use Claude to Hack into OpenAI

Resolved

Confidence

80%

Impact: 70%

Updated 1h ago

Consensus Brief

A team of three security researchers from Hacktron successfully hacked into OpenAI employee accounts using Anthropic's Claude Opus 4.8 and 5. They accessed OpenAI's GitHub repository, known as 'Monorepo', by exploiting a vulnerability in the Discourse forum software related to HEIF image processing.

Sourced from

The VergeT2

Primary: The Verge

What Changed Since Last Update

1h ago

This incident marks a significant security breach involving the use of AI tools to exploit vulnerabilities in a major tech company's systems.

Claim Ledger

4 claims tracked across sources

Confirmed Fact

Hacktron accessed OpenAI's GitHub repository, called 'Monorepo', which contains OpenAI's algorithmic secrets.

Confirmed Fact

Hacktron was able to achieve RCE on Discourse Cloud within 24 hours of Claude Opus 5's launch.

Confirmed Fact

OpenAI paid Hacktron $6,500 for finding the bug.

Confirmed Fact

Hacktron's HEIF Heist project adapted to different companies using less than $3,000 in tokens.

Role-Based Impact Analysis

Source Timeline

1 source corroborating

Security researchers used Claude to help them hack into OpenAI

The Verge·1h ago