Technology

Google detects first fully automated AI cyberattack

Google says threat actors have crossed a major threshold by using autonomous AI systems to identify software flaws, create exploits and launch attacks without a person approving every step.

ai cyberattacks: Google detects first fully automated AI cyberattack

AI cyberattacks have reached a new stage after Google identified a multi-agent framework that could find software vulnerabilities, build working exploits and launch them without a human approving each step.

Google Threat Intelligence Group and Mandiant disclosed the finding in a report published Monday, November 16, 2026, describing it as the first time they had identified threat actors using large language model frameworks to fully automate vulnerability discovery and exploitation.

The finding does not mean that every cyberattack is now autonomous, but it shows that parts of the attack process can operate at machine speed once criminals gain access to the right tools and infrastructure.

How the AI cyberattacks worked

Google said investigators found an exposed staging server hosting a framework called exploit_agent, which connected several frontier and locally run AI models.

The system was designed to continuously compare software before and after security patches, identify likely weaknesses, draft attack code and execute the resulting exploits automatically.

Investigators also found scripts for internal reconnaissance, brute-force attempts, webshell installation and the removal of webshells placed by other attackers.

Telemetry suggested the operator was particularly interested in French websites linked to media, entertainment and education, although Google did not identify specific victims publicly.

The framework’s activity was different from a chatbot simply helping a hacker write code.

Instead, the AI agents were connected to tools and operational workflows that allowed them to analyse targets, make tactical decisions and carry out multiple stages of an intrusion with limited or no direct intervention.

Google said it has also observed threat actors using open-source frameworks such as CAIRN, CyberStrikeAI, HexStrike AI and PentAGI to automate reconnaissance, scanning and exploitation tasks during live intrusions.

Why the development matters

Software defenders have traditionally relied on the time between a vulnerability being disclosed and an attacker developing a reliable exploit.

Autonomous AI tools can compress that window by comparing patched and unpatched code, testing possible attack paths and adjusting their approach based on real-time results.

Google said this shift is especially significant for known vulnerabilities, often called n-days, because attackers can focus on a limited set of public technical materials rather than searching broadly for an unknown flaw.

The company reported that overall vulnerability disclosures and exploits nearly doubled between January and August 2026, while zero-day exploitation rose from a 2025 average of eight incidents per month to 11 per month in 2026 before reaching 22 in August.

Those figures do not establish that AI caused every increase, but Google said its observations show AI is changing both the speed and the risk profile of vulnerability exploitation.

Autonomous attacks are fast but still noisy

Google cautioned that current autonomous attack frameworks have weaknesses of their own.

Unrestricted automation can leave behind unusual traffic, identifiable tool names and distinctive code traces that make the activity easier for defenders to detect.

That trade-off means attackers may gain speed and scale while sacrificing stealth, particularly when a framework performs many actions in rapid succession.

Google expects that balance to change as the underlying models and attack tools become more sophisticated.

More capable systems could reduce the amount of human involvement while also making automated activity harder to distinguish from normal administrative or development work.

What organizations should do next

The report reinforces the importance of applying security patches quickly, especially on internet-facing systems.

Organizations should also monitor for unusual vulnerability scanning, unexpected administrative activity, new service accounts and unexplained use of cloud computing or AI services.

Strong identity controls are increasingly important because attackers can use stolen credentials to access cloud resources, AI models and software development environments without deploying traditional malware.

Google separately reported that some attackers were hijacking cloud accounts and AI quotas, with fraudulent usage charges exceeding US$100,000 in a single day in certain incidents.

In the most serious cases described by Google, unauthorized resource use reached US$1 million before remediation.

For Canadian businesses, public agencies and schools, the practical lesson is not that an autonomous system will necessarily target them next.

It is that the speed of exploitation may increasingly be determined by automated systems rather than by the working hours, staffing levels or technical skill of an individual attacker.

That makes continuous monitoring, rapid patch verification, multifactor authentication, least-privilege access and tested incident-response plans more important than relying on periodic security reviews.

Google’s response

Google said its security teams have deployed behavioural and volumetric detection rules to identify activity associated with some agentic offensive frameworks.

Mandiant has also released hardening guidance intended to help organizations defend against the speed and scale of AI-enabled attack tools.

Google is promoting its own AI Threat Defense architecture, which combines threat intelligence, automated remediation and AI-supported monitoring across enterprise environments.

The company’s report also stressed that AI remains useful to defenders, including for discovering vulnerabilities before criminals exploit them, prioritizing risks and helping repair insecure code.

The latest evidence points to an important transition rather than a single dramatic takeover by machines.

Human operators still appear to select targets, obtain access and configure many attack systems, but autonomous tools can increasingly handle the repetitive and technically demanding work that follows.

That shift gives defenders less time to respond and raises the value of security systems that can detect, investigate and contain suspicious activity at a comparable speed.

Frequently Asked Questions

What did Google discover about AI cyberattacks?

Google identified a multi-agent framework that could compare software versions, find vulnerabilities, create exploits and execute them without human approval at each stage.

What is the exploit_agent framework?

Exploit_agent was the name Google gave to a framework found on an exposed staging server that used multiple AI models to automate vulnerability research and exploitation.

Were Canadian organizations targeted?

Google’s public report did not identify Canadian victims, and its telemetry pointed to interest in French websites connected to media, entertainment and education.

Are all cyberattacks now fully autonomous?

No. Google reported a limited but significant shift toward autonomous workflows, while many operations still involve human decisions, stolen credentials and manually configured infrastructure.

How can organizations reduce the risk?

Organizations should patch internet-facing systems quickly, use multifactor authentication, limit administrative privileges, monitor cloud and AI usage, and maintain tested incident-response plans.

Fact-Checked: Key facts were checked against Google Cloud’s November 16, 2026 GTIG report and Google’s accompanying first-party explanation of autonomous attack activity.

Disclaimer: The report describes observed threat activity and emerging capabilities, not proof that every future cyberattack will be fully autonomous.

Discover more from CTC News

Subscribe now to keep reading and get access to the full archive.

Continue reading

4 New CRA Benefit Payments Coming In November 2026

New Canada Workers Benefit Increase Coming In 2027

New Service Canada Benefit Payments Coming In November 2026

8 New Ontario Laws And Rules Coming In November 2026