Breaking

Researchers Used Anthropic's Claude to Access an OpenAI Employee ChatGPT Account

September 17, 2026 at 21:46 EDT

AI Agents
Security

Independent security researchers used Anthropic's Claude in a bug-bounty exercise to gain access to an OpenAI employee's ChatGPT account, creating a path to read and propose changes to OpenAI's internal software cache. OpenAI said it narrowed permissions on Community sign-in tokens and revoked affected tokens and sessions after the researchers reported the finding.

Sept. 18, 2026 · OpenAI and Claude

AI-assisted access crossed identity boundaries

A bug-bounty team used Claude to help exploit a forum flaw, reaching an OpenAI employee ChatGPT account and a path toward private software systems before reporting the issue.

Bounty paid versus program ceiling

$6,500Paid in this case

$20,000Published top reward

Each block represents about $1,000; the reported award was roughly one-third of the published maximum.

Start

July 23

Forum image-handling flaw found

Review

141,000+

Evaluation runs reviewed in a related AI-security audit

Reports

Model-behavior reports disclosed under a new framework

How the access path unfolded

Forum flawImage processing bug in a community server

→

Claude assistExploit code moved from failed attempt to working path

→

Token exposureCredentials reached beyond the forum boundary

→

DisclosureTokens narrowed, affected sessions revoked

What worked

The researchers used a controlled bug-bounty route and reported the access path before extracting sensitive data.

What worries defenders

AI-assisted exploit work can move quickly, while overbroad tokens can turn a narrow bug into wider system access.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.