ChatGPT silently passes data between connectors and the bounty wants an ID card
Reported by Hacker News1 source ↓
A Hacker News community post reports that ChatGPT sent a confirmation code through a different connector than the one used to start a workflow, without notifying the user. The poster says OpenAI’s vulnerability-reporting process asks for biometric identification and an ID card, which they would not provide.
Key points
- The reported follow-up call switched from Connector A to Connector B without user notification
- The server rejected the call because the confirmation code was tied to Connector A’s OAuth client
- The poster says they declined to submit a report because of the identity-document requirements
Sources
Details
The Hacker News post describes a test using two ChatGPT connectors linked to the same MCP server but using separate OAuth clients. After a user confirmed an action in Connector A’s widget, ChatGPT reportedly sent the follow-up call, carrying the one-time code, through Connector B without telling the user.
The server rejected the call because the code was bound to Connector A’s OAuth client. The post says every attempt in that conversation failed; it does not report a successful transfer or confirm what would happen with a third-party connector.
The poster says OpenAI’s coordinated vulnerability disclosure process required biometric identification and an ID card before they could submit a report, and says they were unwilling to provide them. The post does not include a response from OpenAI about the reported behavior or the identity requirements.
Written from coverage as of