On November 16, 2026, Google Threat Intelligence Group (GTIG) published the latest edition of its AI Threat Tracker, written by Aurora Blum, Kelli Vanderlee, Jess Xia and David Mainor and titled "Agentic Threat Activity Crosses a Significant Threshold." Drawing on Mandiant incident response and its own tracking, GTIG says adversaries have moved from basic prompting to autonomous multi-agent frameworks, automated vulnerability exploitation and new ways of monetizing stolen AI access. A companion post on Google's blog summarizes the finding as the arrival of autonomous attacks and presents Google AI Threat Defense as the response.

The first finding concerns offensive frameworks. Besides the widely discussed pentesting tool ARTEX, GTIG says actors are using open source multi-agent frameworks including CAIRN, CyberStrikeAI and PentAGI, and has seen experimentation with HexStrike AI. GTIG says it independently observed CAIRN and PentAGI in the wild. It reports CAIRN used against targets that included the US aerospace sector, and a threat actor using a customized platform built from PentAGI between late August and early October 2026 for reconnaissance and scanning against organizations in Asia. A PRC-nexus group was also developing its own automated attack framework, according to the report. GTIG notes that the intrusion mechanics are unchanged, such as stolen credentials and unpatched systems, but speed and scale are not. It adds that these tools currently trade stealth for velocity, leaving forensic traces that help detection.

The second finding is the one GTIG calls a first. It says it identified threat actors using LLM frameworks to automate vulnerability discovery and exploitation with no human in the loop. The example is an exposed staging server hosting a multi-agent framework dubbed exploit_agent. It compared code before and after patch announcements, drafted exploits and ran them automatically, and victim data on the server suggested a focus on French media, entertainment and education websites. GTIG's analysis of vulnerability statistics found that overall disclosures and exploits nearly doubled between January and August 2026, while zero-day exploits rose from an average of 8 per month in 2025 to 11 per month in 2026, spiking to 22 in August. GTIG also warns that LLM-generated exploits claimed to work often do not, adding verification work for defenders.

The report also covers money and influence. Mandiant responded to several LLMJacking incidents in 2026, in which criminals used compromised cloud environments to consume victims' compute and model quotas. In some cases single-day fraudulent charges exceeded $100,000, and in the most severe cases cumulative abuse reached $1 million before remediation. GTIG is tracking illicit "LLM Mixers" that resell access to stolen model credentials, with advertised prices of $0.10 to $25 per million tokens. Information operations actors, it says, now routinely use LLMs for synthetic personas and content at scale, and a suspected DPRK cluster abused organizations' AI tools in a supply chain attack path.

On defense, GTIG says Google Security Operations has deployed rulepacks to detect ARTEX use, and Mandiant released a hardening guide for LLM-enabled offensive tooling. These are Google's descriptions of its own findings and products. The telemetry behind the report is not public, so outside readers cannot verify the counts independently.

References