What Google's Report Reveals
In a report released on November 16, 2026, Google's Threat Intelligence Group (GTIG) and its cybersecurity firm Mandiant detailed a significant shift in the cybercrime landscape. For the first time, they observed a fully automated system, dubbed 'exploit_agent',
designed to turn software security patches into weapons. This framework was discovered on an exposed server and was built to compare software code before and after a security update is released. By analysing the changes, the AI can deduce the nature of the vulnerability that was fixed and then automatically generate code to exploit it on systems that have not yet been updated. This drastically shortens the time between a patch being issued and an attack beginning, a critical window that security teams rely on to protect their systems.
AI as a Hacker's Assistant
The core of this new threat lies in its speed and automation. Previously, a skilled hacker would need to manually reverse-engineer a patch to understand the flaw—a process that could take days. Now, AI-powered tools can perform this analysis relentlessly and at machine speed, attempting to create and launch an attack while the hacker sleeps. According to Google, these AI systems are not necessarily inventing novel types of attacks but are drastically accelerating existing techniques. The 'exploit_agent' framework was found alongside tools for network reconnaissance, password-cracking, and deploying malware. Evidence suggested its operator was targeting French websites in the media, entertainment, and education sectors. Beyond this specific tool, Google noted that other open-source AI frameworks like ARTEX and CAIRN are being used in live intrusions against sectors including US aerospace.
The New Cyber Arms Race
This development signals a new 'arms race' in cybersecurity. While attackers leverage AI for offense, defenders are also using it to build smarter, faster defenses. However, the advantage often goes to the attacker, who only needs to find one flaw, whereas defenders must protect a vast and complex digital environment. The rise of autonomous frameworks means security teams are no longer just competing against human adversaries but against automated systems that can scale their operations significantly. This shift pressures organizations to accelerate their own patch cycles and response times to keep pace. Experts warn that while the attack methods are familiar, the velocity is a game-changer. The window for response is shrinking from days to hours, or even minutes.
Beyond Exploits: Other Malicious Uses
AI's role in cybercrime isn't limited to finding vulnerabilities. The Google report also highlights a trend called 'LLMjacking', where criminals hijack a company's cloud accounts to steal expensive AI computing resources. These stolen resources are then used for their own purposes or resold on underground markets. In some cases, fraudulent charges have exceeded $100,000 in a single day, with the worst incidents accumulating up to $1 million in costs for the victim company. Furthermore, state-sponsored groups are using large language models (LLMs) to enhance their operations, from creating more convincing phishing emails to generating synthetic media for influence campaigns. For example, a North Korea-linked group used AI-generated profile pictures for fake recruiter personas to target employees in the cryptocurrency sector.
What This Means for Security
The primary takeaway from Google's findings is that the speed of cyber conflict is increasing dramatically. Organizations can no longer rely on purely manual defense strategies. The need for automated security tools that can detect and respond to threats in real-time is more critical than ever. While these new AI attack tools currently seem to favour speed over stealth—leaving behind forensic trails that can aid detection—experts anticipate they will become more sophisticated and harder to catch over time. For businesses and other organizations, the fundamentals of cybersecurity remain crucial: patch systems quickly, use strong authentication methods, and maintain visibility over network activity. However, these practices must now be executed with a new sense of urgency to counter the machine-speed threat.
















