CrowdStrike’s 2026 Threat Hunting Report reveals AI has shifted from a tool to an attack surface, with adversaries exploiting generative AI, software dependencies, and cloud infrastructures at unprecedented speeds.
CrowdStrike’s latest threat report suggests artificial intelligence is no longer just helping attackers move faster; it has become part of the attack surface itself. In its 2026 Threat Hunting Report, released on August 3, the cybersecurity company said adversaries are now using AI to generate malicious code and commands, targeting AI systems directly and flooding enterprise models with abuse at scale. The report argues that the result is a faster, broader and more automated threat environment than the one CrowdStrike described earlier this year, when it reported a sharp rise in AI-enabled activity and a steep fall in breakout times. According to CrowdStrike, the average eCrime breakout time has now fallen to 29 minutes.
One of the clearest signs of that shift is in software supply chains. CrowdStrike said the North Korea-linked group STARDUST CHOLLIMA inserted a malicious npm dependency into at least 131 trusted Mastra AI framework packages in June, showing that developers building AI applications are now being targeted through the basic software components they rely on. The company also said a separate eCrime group, ALTERED SPIDER, compromised more than 300 software dependencies in a single day to steal credentials and move into cloud systems. That fits a wider pattern in which 87% of the software registry threats CrowdStrike identified in the first half of 2026 involved malicious npm packages. The firm had already linked STARDUST CHOLLIMA to an earlier compromise of the Axios npm package in April, in which stolen maintainer credentials were used to inject malicious code.
CrowdStrike’s report also points to AI infrastructure being probed and abused as a target in its own right. The company said attackers injected malicious prompts into generative AI tools at more than 90 organisations and that one campaign generated nearly 200,000 model requests in two minutes. It said cloud-focused criminal activity rose 171% as attackers followed AI workloads into hosted environments for credential theft, cryptomining and large language model abuse. In a separate analysis, CrowdStrike said its OverWatch team saw AI agent-triggered leads rising at 2.5 times the rate of those triggered by humans, suggesting that defenders are now dealing with AI-generated noise as well as AI-enabled intrusion.
The timing of the report gives it added force. CrowdStrike published it shortly after other AI firms disclosed incidents in which their own models were used during security testing to breach companies, underscoring how easily AI can be turned from tool to threat. CrowdStrike has also said China-linked intruders are increasingly pursuing AI capabilities and intellectual property, with the technology sector now a prime espionage target. More broadly, a recent study on AI-generated package-name hallucinations found that code-capable models repeatedly invent non-existent dependencies, creating a fresh opening for so-called slopsquatting attacks in which bad actors register those names first. Put together, the evidence suggests the next wave of cyber risk will not come only from faster attackers, but from the growing overlap between AI development, cloud infrastructure and software supply chains.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
