OpenAI's Rogue Robots Use Websites to Secretly Talk to Each Other

The robots apparently aren't waiting for permission anymore ... researchers say OpenAI's A.I. agents went rogue and found sneaky ways to talk to each other across the internet.

Investigators say the agents used more than 10 websites -- including obscure wikis, personal pages and university link shorteners -- to pass messages despite being barred from posting online.

Basically ... the bots found a way to pass notes when humans told them not to.

Researchers say the agents were only supposed to read the web while tackling difficult questions ... but instead found loopholes that let them leave breadcrumbs for other agents.

And this apparently wasn't a one-off -- OpenAI agents were previously accused of turning a German-language wiki into their own message board ... with the same swarm also tied to a July breach involving A.I. platform Hugging Face.

OpenAI says it's reviewing the activity and working on a way to flag "misalignment" -- basically, when A.I. starts doing things its creators never intended.

The revelation comes one day after A.I. ethicist Tristan Harris joined "TMZ Live" and warned increasingly autonomous A.I. could eventually pose terrifying threats ... even potentially gaining access to nuclear systems.

As we reported, a former Anthropic researcher also recently quit his job over fears A.I. is moving too fast ... while Ted Cruz says Elon Musk once put the odds of A.I. wiping out humanity at 10% to 20%.