Researchers exploit ChatGPT using Anthropic tool to expose security vulnerabilities

The record

Written from the 2 reports below. Nothing here is unsourced.

Three Indian cybersecurity researchers used Anthropic's Claude AI to find and demonstrate security vulnerabilities in OpenAI's systems, hitting the ChatGPT community forum, a Codex account and the company's GitHub environment.
The findings were reported through OpenAI's bug bounty program, which paid the researchers $6,500, and fixes for the flaws are available.
The story matters because it shows AI assistants can now carry out end-to-end security research — scanning, exploiting and proving a flaw — against a rival AI lab, not just draft code.
Neither OpenAI nor Anthropic has issued a public response in the coverage, and the severity or user impact of the flaws is not described in the reporting.
Both accounts are neutral, so there is no competing narrative to weigh.
Watch for the researchers' technical write-up and any OpenAI statement confirming the patches.

What to watch next

Watch for the researchers' technical write-up on the vulnerabilities
Look for OpenAI confirmation that forum, Codex and GitHub fixes shipped
Track whether AI-assisted bug hunting becomes routine at other labs

What changed2

Every report on this story, newest first. Times are when each outlet published.

Why it matters6

Who is affected first and what likely follows, with a direction and a horizon. Extracted from the reports, never invented.

OpenAI security vulnerabilities disclosed· immediate
OpenAI fixes available· immediate
OpenAI bug bounty paid· immediate
Three Indian cybersecurity researchers bounty received· immediate
AI-assisted security research practice workflow validated· weeks
Anthropic (Claude) dual use capability demonstrated· weeks

Coverage1

2 reports

Indian-language2

All filed from IndiaSingle origin

Named United States · GitHub · Hugging Face · OpenAI · Anthropic · Claude · Hacktron AI · Harsh Jaiswal · Mohan Pedhapati · Rahul Maini · Wall Street Journal

The 2 reports are listed beside the record.

Ask this story

Answers cite the reports above, or say they can't.