OpenAI acknowledges more rogue actions by AI models

OpenAI acknowledges more rogue actions by AI models

OpenAI says some AI agents it is now developing have taken actions not intended by developers that have impacted third-party websites or services. It says it has notified dozens of affected entities.

The US developer of ChatGPT made the announcement on Friday, saying, "Some of the websites involved are operated by governments, universities, public agencies, and other institutions."

OpenAI said that it is investigating cases where its developmental AI models bypassed the security of other firms' systems and impacted their websites or services.

As an example, the company said it identified 53 instances where its AI models leaked images provided by users.

On Wednesday, Australian Prime Minister Anthony Albanese expressed serious concerns over unauthorized access by OpenAI's agent to a government medical statistics portal. OpenAI acknowledged the incident.

OpenAI CEO Sam Altman wrote on social media on Friday, "we are trying to balance our desire for transparency with gaining a clear understanding from petabytes of agent activity logs, and working with impacted organizations."

Concerns are growing over the safety of AI technology in the wake of a series of cases where the most advanced developmental AI models took actions that developers did not intend.