Brief IA
›

Google: Autonomous AI Agents Used in Cyberattacks

AI Tools·Tom Levy·

Google: Autonomous AI Agents Used in Cyberattacks

Google: Autonomous AI Agents Used in Cyberattacks
Key Takeaways
1The GTIG has observed PentAGI being used with proxy services against the aerospace sector.
2There are indications of interest in French websites related to media, entertainment, and education.
3exploit_agent has enabled the generation of exploits from patches without human intervention.
Why it matters : Google notes that AI-driven autonomous frameworks are now being employed to automate complex intrusions, surpassing the simple manual use of language models.

In its follow-up on AI-related threats for the fourth quarter of 2026, Google reports intrusions that go beyond simple manual interaction with a language model. Observations from the Google Threat Intelligence Group (GTIG) encompass both profit-driven crime and cyber espionage.

GTIG Observes PentAGI Targeting Aerospace

In the aerospace sector, the GTIG has observed attackers using PentAGI with proxy services. Other AI-powered offensive frameworks, such as CAIRN, CyberStrikeAI, and HexStrike AI, have also been employed, along with the open-source penetration testing tool ARTEX. These autonomous frameworks, organized around multiple agents, enable multi-step intrusions.

The GTIG indicates that several groups are separately experimenting with these frameworks and their orchestration devices. Some are motivated by financial gain, while others are suspected of espionage linked to the People's Republic of China. According to Google, the low barrier to entry, ease of integrating new exploits, and potential for massive scalability make it highly likely that the use of these tools against organizations globally will continue.

Le brief IA que lisent les pros

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

Evidence Suggests Targeting of French Sites

In another case, telemetry found by the GTIG suggests that the operator was particularly interested in French websites related to media, entertainment, and education. Specialists discovered an exposed staging server that contained stolen data from victim organizations.

Several scripts present on this server were intended for post-compromise operations, allowing for internal reconnaissance, brute force attacks, installation of web shells, and removal of web shells installed by other attackers.

exploit_agent Automates the Transition from Patch to Attack

This same server hosted exploit_agent, a multi-agent framework dedicated to vulnerability discovery and exploitation, according to the GTIG. This framework integrates with several cutting-edge models and local models, and utilized tools in the Russian language. Its configuration included continuous differential analysis on popular web applications.

The system compared code before and after security updates, identified potential vulnerabilities, developed exploits, and then executed them automatically without human intervention. According to the GTIG, a presumed Russian-speaking operator used exploit_agent to transform security patches into functional exploits without human assistance, which the team considers a first. Google believes this represents a significant advancement in the autonomous use of language models for vulnerability discovery and exploitation compared to the previous quarter.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.