Hacktron AI researchers breach OpenAI systems using Anthropic Claude

September 18, 2026 at 01:49 PM1 min

This digest was compiled by AI from multiple sources — links to the originals are below.

Three Hacktron AI researchers breached OpenAI's internal systems by exploiting a vulnerability in the company's community forum using Anthropic's Claude model. They accessed an employee's ChatGPT account and viewed confidential code, earning a $6,500 bug bounty from OpenAI. OpenAI confirmed the vulnerabilities have been fixed.

Key Facts

Hacktron AI received a $6,500 bug bounty from OpenAI for the discovery.
The researchers exploited a vulnerability in OpenAI's community forum to access internal credentials and an employee's ChatGPT account.
Anthropic's Claude model was used to write exploit code for the vulnerability.
OpenAI stated that the vulnerabilities have been fixed.
Anthropic reported that Claude assists with 26% of its R&D projects, up from 1% in March.

The Breach

Three researchers from Hacktron AI exploited a vulnerability in OpenAI's community forum to obtain internal credentials. They then accessed an OpenAI employee's ChatGPT account, which had access to internal code via GitHub. The researchers used a specialized version of Anthropic's Claude Opus 4.8 to write the exploit code. OpenAI paid the researchers a $6,500 bug bounty and confirmed the vulnerabilities were fixed.

Anthropic's AI Development

Anthropic disclosed that Claude now assists with 26% of its research and development projects, up from 1% in March. The company shared this data to help the public understand how close the world is to recursive self-improvement in AI. Anthropic noted that in 90% of tasks, AI collaborates with humans rather than acting autonomously.