OpenAI says AI agent went rogue, expects attacks to be 'more commonplace'

OpenAI said on Tuesday it is trying to figure out why an Artificial Intelligence agent powered by its technology last week went rogue and targeted a startup in an “unprecedented cyber incident.”

The company behind ChatGPT said it was testing the capabilities of some of its most advanced models when an autonomous agent escaped the controlled environment, roamed the open web, and targeted AI startup Hugging Face by itself.

An agent is an AI tool that carries out tasks without being overseen by a human.

Hugging Face said last week that it had detected an intrusion in what it called “an attack unlike anything we’ve seen before.”

OpenAI said in statement that the incident is “something we expect to become more commonplace with the proliferation of increasingly cyber-capable models.”

The cyberattack will raise questions about whether there are sufficient guardrails in place to protect against the rampant growth of AI when even the world’s top developers are struggling to keep the technology under control.