This is a div block with a Webflow interaction that will be triggered when the heading is in the view.
This research provides a technical assessment of the shift from automated to autonomous AI-orchestrated cyber operations (hereafter AI.Attackers ). We present primary evidence that offensive use of AI has moved from laboratory and proof-of-concept to operational deployment at scale, with documented campaigns in which AI agents execute the majority of attack lifecycle tasks with limited human intervention.
Key findings
- The Anthropic GTG-1002 campaign showed an AI agent executing roughly 80-90% of tasks across reconnaissance, exploitation, credential harvesting, lateral movement, and exfiltration against 30 organizations, with human input at only about 4-6 strategic decision points.
- Analysis of 33 attack infrastructure clusters found a mean composition of about 61 domains, 88 subdomains, 35 SSL certificates, and 11 social media accounts per cluster.
- Approximately 82% of cluster-associated domains had not triggered any security vendor detection at the time of measurement.
- The average window from domain registration to overtly malicious use is 72 days, defining a measurable interval for IoPA-based detection and MTTP.
- Malicious GitHub repositories grew from 6,498 in 2022 to about 110,000 in 2024, roughly 1,592% growth, with similar 285-340% acceleration across GitLab and Bitbucket in 2023-2024.
- Mean estimated cost per attack cluster is about $9,223, allowing a threat actor to run 10 or more active clusters for under about $100,000.
What is inside
- Executive Summary
- Terminology and Scope
- Evolution: From Automation to Autonomous Attack
- Case 1: AI-Orchestrated Cyber Espionage (Anthropic GTG-1002)
- Case 2: Industrial-Scale Attack Infrastructure (Malanta Cluster Analysis)
- Case 3: Exponential Growth in Malicious Resources (GenAI Correlation)
- Synthesis: Why Pre-Attack Prevention Is Technically Necessary
- Data Sources and Methodology
Frequently asked questions
What is an AI.Attacker?
An AI.Attacker is an autonomous or highly automated AI agent that conducts offensive cyber operations with minimal human-in-the-loop control. The report documents this shift with the Anthropic GTG-1002 campaign, in which an AI agent executed roughly 80-90% of attack lifecycle tasks.
What is an IoPA?
An IoPA, Indicator of Pre-Attack, is a validated signal that adversary infrastructure is being prepared before use. In this report, IoPAs include domain registration patterns, certificate issuance, and repository creation observed during MITRE ATT&CK TA0042, Resource Development.
How is an IoPA different from an IOC?
An IOC documents malicious activity that already occurred. An IoPA identifies validated evidence of infrastructure preparation before execution, which is why IoC-only defense is structurally late against compressed AI-driven attack timelines.
What is MTTP?
MTTP means Mean Time to Preempt. It measures the interval between observing adversary preparation and taking defensive action against it. The report argues MTTP should be elevated alongside legacy MTTD and MTTR metrics for security teams and cyber insurance underwriting.
What is the 72-day setup window?
The setup window is the average time from domain registration to overtly malicious use, measured at 72 days in Malanta cluster analysis. It defines a concrete interval in which defenders can observe IoPAs and disrupt staged infrastructure before first contact, especially since about 82% of cluster domains had triggered no vendor detections.
How does GenAI relate to the growth of malicious infrastructure?
Malicious GitHub repositories grew from 6,498 in 2022 to about 110,000 in 2024, with a sharp inflection coinciding with broad availability of GenAI and public LLMs. The same 285-340% acceleration appears across GitHub, GitLab, and Bitbucket, supporting systematic AI-assisted resource generation.
