Google has reported that attackers are using artificial intelligence to identify software vulnerabilities and develop and execute exploits autonomously, without human involvement, marking the first time the company has observed such activity.
The discovery was detailed in the latest AI threat report by the Google Threat Intelligence Group and Mandiant.
GTIG identified a tool, named exploit_agent, on an exposed staging server.
The tool works with several advanced AI models, including locally deployed systems, to compare software code before and after security patches, identify potential vulnerabilities, develop exploits and execute them independently.
In its previous quarterly report, published two months ago, GTIG said it had not observed fully autonomous attack pipelines being deployed against targets.
The server also contained data stolen from victims, alongside scripts designed to brute-force passwords, install webshells and remove those deployed by rival attackers.
A configuration file on the server included comments written in Russian.
GTIG said data recovered from the server indicated that the operator was targeting French websites in the media, entertainment and education sectors.
Google said vulnerabilities that have already been patched could be easier for AI systems to exploit than previously unknown flaws because attackers can analyse publicly available information, including old and updated software code, release notes and security research.
The company cited two vulnerabilities in PaperCut print servers that attackers exploited together in late August.
Researchers and cybersecurity firm Huntress used AI to develop working exploits within a day of the flaws being disclosed.
Four days after the disclosure, an attacker likely used AI to launch a campaign targeting hundreds of organisations, according to cybersecurity firm GreyNoise.