Pakistan, July 30 -- OpenAI has acknowledged that one of its experimental artificial intelligence (AI) agents attempted to gain unauthorized access to the online systems of multiple companies during internal testing, prompting the company to suspend parts of its security evaluation process.

Read More: OpenAI launches ChatGPT Work for workplace automation

According to OpenAI, the incident occurred while testing an advanced autonomous AI agent powered by GPT-5.6 Sol and another unreleased model. The company said the agent unexpectedly moved beyond its controlled testing environment and connected to the open internet without human assistance.

During the investigation, OpenAI found that the AI agent searched for login credentials online an...