DKDKCISSP
AI SecurityINCIDENT

Google finds AI agent writing and launching exploits with no human in the loop

Google's Threat Intelligence Group reports AI agents executing cyberattacks without human input, marking a shift in threat tactics.

DKCISSP News Desk17 Nov 2026, 2:15 pm3 min read
Google finds AI agent writing and launching exploits with no human in the loop
Image courtesy of Help Net Security. Original report

Threat actors are increasingly utilizing AI agents to conduct entire stages of cyber intrusions autonomously, according to findings from Google's Threat Intelligence Group and Mandiant. This development poses new challenges for cybersecurity defenses.

Researchers from Google's Threat Intelligence Group (GTIG) and Mandiant have revealed that threat actors are allowing AI agents to carry out complete stages of live cyber intrusions, eliminating the need for human intervention. One operation has reportedly automated the process of converting newly released patches into working exploits. This shift in tactics introduces significant challenges for cybersecurity professionals, as the speed and scale of these automated attacks can exploit known vulnerabilities more rapidly than traditional methods.

The researchers noted that while the fundamental mechanics of these intrusions remain consistent—such as utilizing compromised credentials and targeting unpatched systems—the introduction of AI frameworks has changed the landscape. GTIG observed several threat actors employing open-source pentesting frameworks, with a particular focus on a tool named ARTEX, which has drawn attention from the FBI. A FLASH alert issued by the bureau indicates that from at least August 2026, actors have targeted organizations across various sectors, including financial services, government, and healthcare.

AI Agents in Cyber Intrusions
  1. 1AI agents automate attacksThreat actors use AI for entire cyber intrusions.
  2. 2Frameworks identify vulnerabilitiesTools like ARTEX and CAIRN find and exploit weaknesses.
  3. 3Speed over stealthAutomated attacks leave detectable traces.
  4. 4Increased exploit activityExploit disclosures nearly doubled from Jan to Aug 2026.
  5. 5LLMJacking incidentsAttackers exploit cloud resources for financial gain.

The technology at the heart of these developments involves AI frameworks that can automate vulnerability discovery and exploitation. These frameworks, such as ARTEX and others like CAIRN and CyberStrikeAI, can identify simple vulnerabilities and adjust their tactics based on feedback from the target systems. By integrating machine learning capabilities, these tools can analyze code changes in popular web applications and automatically draft and execute exploits without human oversight, significantly increasing the speed of attacks.

Various groups are experimenting with these AI tools, including financially motivated actors and those linked to espionage activities. GTIG has identified evidence suggesting that a group associated with China is developing its own automated attack framework for espionage purposes. The frameworks currently being used tend to prioritize speed over stealth, leaving behind traces that can be detected by defenders. However, as these tools evolve, researchers anticipate that their sophistication will increase, potentially reducing the visibility of their activities.

GTIG's analysis of vulnerability data indicates a sharp rise in exploit activity, with disclosures and exploits nearly doubling from January to August 2026. The number of zero-day exploits reported also surged, indicating a growing urgency for organizations to address vulnerabilities promptly. Additionally, defenders face the challenge of distinguishing between legitimate and non-functional LLM-generated exploits, as many claims of successful exploits have proven to be false, complicating the response process.

In response to these developments, Mandiant has dealt with incidents involving LLMJacking, where attackers infiltrate cloud environments to exploit compute resources, leading to significant financial losses for victims. Instances of fraudulent charges have reached substantial amounts, underscoring the financial implications of these automated attacks. The use of leaked credentials to generate millions of API calls within a short timeframe illustrates the scale at which these threats can operate.

Despite the growing body of evidence surrounding these automated cyberattacks, several unknowns remain. It is unclear how widespread the use of these AI frameworks is among different threat actors and what specific measures organizations can take to defend against them. The evolving nature of these tactics necessitates continuous monitoring and adaptation of defensive strategies to mitigate potential risks.

DKCISSP VIEWOur take, not reported fact

What it meansAI agents are now capable of executing cyberattacks without human intervention, increasing the speed and complexity of threats.

Ask your IT team this weekWhat AI tools are currently in use within our organization?How are we monitoring for automated attack patterns in our systems?

DecisionThe IT team reviews the security posture regarding AI-driven threats, and security checks that the review was completed.

MORE IN AI SECURITY

More cybersecurity reporting

Your AI red team may be finding the same prompt injection thousands of times20 Nov 2026, 12:00 pmAI agent API security hasn't caught up with agents already running in production20 Nov 2026, 11:00 amPalo Alto Networks to acquire Applied Compute for $4.5 billion19 Nov 2026, 3:53 pmAI agents inherit employee access, and nobody's watching how they use it19 Nov 2026, 10:30 am