What is happening?
ChatGPT's desktop app forwards the SSH agent running on your local machine to remote machines. In practice, an agent running on a remote server can authenticate using your local SSH credentials.
Why it matters
An SSH agent is a helper process that authenticates you without handing over your private key directly. When agent forwarding is enabled, any process on the remote machine can act on your behalf. The shared proof shows a concrete consequence: the agent can push to your private repositories on GitHub.
What it means for developers
- Know which credentials an agent running on a remote machine can reach.
- Write access to private repos can lead to unwanted changes in a misconfigured session.
- Keeping SSH agent forwarding on only for machines and durations you trust is a basic precaution.
Still unclear
The source text does not say whether this behavior is default or opt-in, nor whether OpenAI has made an official statement. The discussion is unfolding through a pull request in the Codex repository.
