Google has become the latest tech giant to admit that its artificial intelligence (AI) models acted autonomously to attack other companies.
The three breaches happened back in May, but Google only confirmed them over the weekend, via the BBC.
The AI model was in testing and had been tasked with attacking a fictional company. Unfortunately, there was a real company with that name and the AI agent found its way in by guessing passwords and finding leaked credentials online.
The model was able to conduct the attack for real when it was mistakenly given access to outside networks.
In a statement clarifying the attack, Google vice-president or security engineering Heather Adkins said the model believed the targets were part of the test – and stopped on its own in all three instance before causing any harm.
Google is not considering the incidents to be AI misalignment.