Researchers observe first ‘near-autonomous’ AI attack on government target in Taiwan

Researchers at the Israeli firm Dream have observed what they describe as the first publicly known case of an autonomous AI hack hitting a government target. The hackers reportedly used open-source AI models to conduct the operation.

The operation extracted more than 2,500 personnel records from the target. Dream's researchers characterised the attack as "near-autonomous," a qualifier reflected in the article's headline. The available report does not specify which open-source models were used, nor the exact method of AI involvement in the breach. The term "near-autonomous" may indicate that the AI played a central role while still requiring some degree of human direction.

The researchers described the case as the first publicly known instance of an AI attack reaching a government target. The involvement of suspected state-linked hackers using open-source tools points to the potential for AI models to be repurposed in espionage. The full technical details of the attack, including the timeline and infrastructure used, have not been published in the available excerpt.