Three Researchers Used Claude Opus to Breach OpenAI Internal Repos for Under $3,000

Three Indian security researchers — Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini — disclosed a complete penetration demo against OpenAI assisted by AI: starting from an image-upload flaw, they chained two vulnerabilities into takeover of multiple OpenAI employees' ChatGPT accounts and accessed internal code repositories, in roughly 72 hours for under $3,000 in token costs.

Confirmed

The attack was carried out by the three researchers, none with elite-school or big-tech backgrounds, relying on technical skill.
Starting from an image-upload flaw, chained exploitation of two vulnerabilities escalated to employee account takeover; after takeover they reportedly could manipulate Codex in the accounts.
The model used was Claude Opus (referred to as Claude Opus 5 in several posts); cost was under $3,000 in tokens over about 72 hours (some posts say 2 days).
Disclosed by security team Hacktron AI; @marcvanderchijs notes the disclosure dates the breach to July 25, 2026.
The story spread on Reddit as "three guys with Claude hacked OpenAI" and drew wide discussion.

Unconfirmed

OpenAI's official response, fix status, and exact bounty: @marcvanderchijs says the bounty was only $6,500, which cannot be cross-verified with other posts.
m1's shared blog post contains only a title and link without attack details; full details rest on Hacktron AI's disclosure.

Why it matters

A landmark case of AI used in real penetration: a competitor's model (Claude Opus) breached a top AI company's employee accounts and internal repos at very low cost, demonstrating LLM capability in chained vulnerability discovery.
The attackers lacked elite credentials, showing such capability is rapidly commoditizing — a warning for AI companies and enterprise security broadly.

2026-09-20 ~ 2026-09-21 · 5 related posts

Primary sources

Three Hackers Breached OpenAI Internal Repos in 3 Days Using Claude for $3K — marcvanderchijs · 2026-09-21 19:02 (UTC+08:00)
3 researchers hacked OpenAI in 72 hours for under $3,000 using Claude Opus — FinanceYF5 · 2026-09-21 13:50 (UTC+08:00)
Researchers Use Claude Opus 5 to Hijack OpenAI Employee Accounts for Under $3,000 — FinanceYF5 · 2026-09-20 16:54 (UTC+08:00)

[source] Researchers Use Claude Opus 5 to Hijack OpenAI Employee Accounts for Under $3,000 — FinanceYF5 · 2026-09-20
3 Indian researchers hacked OpenAI in 2 days for under $3,000 using Opus — FinanceYF5 · 2026-09-21
[source] 3 researchers hacked OpenAI in 72 hours for under $3,000 using Claude Opus — FinanceYF5 · 2026-09-21
OpenAI 'hacked by three guys with Claude': blog post makes the rounds — lucky-puke · 2026-09-21

1 near-duplicate retellings: marcvanderchijs