Cyberattacks: Google Outlines Offensive Uses of AI

Models run at the expense of victims, automatically generated exploits, and tools used against real targets: Google describes several offensive uses of AI. However, the observations from its teams distinguish between functional attacks and codes that fail.
Victims bear over $100,000 in fraudulent charges
According to the Google Threat Intelligence Group (GTIG) and Mandiant, cybercriminals are seizing cloud and AI credentials to run models on their targets' accounts. In some cases, the amounts siphoned off in a single day have exceeded $100,000. A vulnerability in the LiteLLM gateway was exploited during one attack.
The teams also report that a suspected North Korean group modified configuration files for Claude Code within companies so that each launch would download a backdoor. Google indicates that it has disabled accounts and assets associated with the abusive use of its Gemini models.
PaperCut provides functional exploits, PHP yields failures
Google believes that AI can more easily exploit vulnerabilities that have already been patched than unknown flaws. Attackers can then rely on a limited set of resources: the code versions before and after the fix, release notes, and public research.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
At the end of August, attackers combined two vulnerabilities affecting PaperCut print servers. Researchers and the cybersecurity firm Huntress relied on AI to create operational exploits within 24 hours of the vulnerabilities being disclosed. GreyNoise reports that, four days after the disclosure, a malicious actor likely used AI to carry out an attack against several hundred organizations.
After the resolution of PHP vulnerabilities in October, hundreds of proof-of-concept exploit examples were published, but these turned out to be ineffective. The GTIG and Mandiant believe that AI is likely behind the majority of these exploits.
GTIG spots CAIRN targeting the U.S. aerospace sector
The GTIG reports that several groups are now using open-source hacking frameworks based on AI during real-time attacks. Among these actors are financially motivated groups as well as suspected Chinese espionage groups. The CAIRN framework was detected during attacks targeting the aerospace sector in the United States.
Google emphasizes that these tools generate a lot of activity and leave traces that simplify their detection. The company anticipates that, with the improvement of these tools, attackers will reduce the amount of human intervention and become harder to detect. The GTIG estimates that it is highly likely these frameworks will continue to be used in campaigns directed against organizations worldwide.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.