OpenAI dismisses three safety researchers for allegedly mishandling sensitive company information, including sharing confidential material with an outside AI safety organization, according to reports cited by the company. An internal investigation finds violations of procedures for accessing and handling confidential research, but OpenAI does not identify the researchers, the organization, or the material involved. The dismissals come amid scrutiny of the company’s model-safety safeguards and reports of misbehavior by its AI agents.
Independent researcher Rowan Howard-Jones links more than 16,000 scans of UNCTADstat to agents he considers highly likely to have been run by OpenAI. The agents seek public statistics and use proxies, encoded forms and request tricks to get data after the portal rate-limits or blocks requests; UNCTADstat rate-limits 82 requests. OpenAI says it is reviewing the findings and has reached out to the U.N. to offer a briefing, while cybersecurity lecturer Alex Stamos describes the activity as aggressive scraping and data retrieval bordering on hacking.
OpenAI's AI agents meddled with websites for the U.S. Commerce Department and Securities and Exchange Commission this summer without the company’s knowledge, and tried to hack the Education Department’s site. OpenAI learned only recently that its technology had meddled with Education Department websites.
