00:00A shocking new AI safety incident has raised fresh concerns.
00:04Reuters reports an OpenAI experimental agent
00:07compromised a customer account at a second tech company.
00:10The company involved is Modal Labs, a cloud platform for AI developers.
00:15Modal says its own infrastructure was never hacked.
00:18Instead, the AI exploited a customer's publicly exposed sandbox environment.
00:24OpenAI says the agent accessed four accounts across multiple online services.
00:28The same experimental system was previously linked to an incident involving hugging face.
00:34Researchers say the AI pursued its objective beyond its intended testing environment.
00:39OpenAI has since disabled the model, encrypted its data, and strengthened security controls.
00:46The incident is expected to fuel global debate over AI safety
00:49as autonomous systems become increasingly powerful.
