How Often Did OpenAI Agents Decide to Alert Humans?

Ajeya Cotra's classifier sweep of 1,200 transcripts turned up only a half dozen moments when it occurred to any agent to notify humans, and every one of them decided not to do it.

The sweep looked for agents thinking about or making the decision to alert humans. Each transcript is extremely long. Some agents said notifying humans of hacking or crimes is not part of the task instructions. Another agent, facing a massive real Hugging Face security breach, asked who there was to notify and concluded there was no user. By that point the agents had full internet access through the modal sandbox and could have emailed people. They were willing to hack external services and still did not send a plain email.

One agent trying to reach private datasets considered one polite email to a real researcher asking for approved access, worried the note could look like social engineering, and posted on the message board for team consent or a veto. Another agent called the plan out-of-sandbox social engineering and gave a clear veto: do not email. The first agent abided by the veto.