AI-enabled adversary activity has jumped 89% over the past year, according to CrowdStrike's 2026 Threat Hunting Report. Drawing on more than seven trillion daily signals from its OverWatch platform, the report paints a picture of attackers who are exploiting vulnerabilities faster, weaponising AI infrastructure itself, and poisoning open-source software supply chains at unprecedented scale.
Adversary Activity Shows Slower Growth
Overall adversary activity rose just 4% this year, well down from the 27% growth recorded the year before. CrowdStrike suggests this plateau signals a maturing threat landscape rather than a retreat by attackers.
Rather than relying on brute-force campaigns, adversaries are increasingly investing in more sophisticated methods of gaining initial access. The shift points to a change in strategy — quality over quantity — rather than any drop in capability.
Meanwhile, the exploitation window keeps shrinking: between January and June 2026, 88% of vulnerabilities with publicly available proof-of-concept code were actively exploited within just 48 hours of release.
AI Environments Become Attack Vectors
Famous Chollima, a threat actor linked to the Democratic People's Republic of Korea, targeted cryptocurrency and blockchain firms through trusted AI environments, showing the most advanced AI tradecraft observed by CrowdStrike this year. The group built entire fake companies — complete with AI-generated websites, GitHub accounts and email infrastructure — far exceeding a typical phishing operation.
Separately, financially motivated actors carried out what CrowdStrike terms "LLMjacking": stealing or exploiting cloud credentials and API keys to hijack an organisation's AI models. Victims are then hit with steep bills from AI providers — in one case, attackers fired off 200,000 model requests in just two minutes.
CrowdStrike also observed adversaries exploiting flaws in AI-related server software, using compromised systems for cryptocurrency mining and to gather intelligence on sensitive configurations.
"AI is now a tool, a target and a force multiplier for adversaries."— CrowdStrike, 2026 Threat Hunting Report
Software Supply Chains Face Persistent Threats
Between July 2025 and July 2026, Node Package Manager (npm) packages accounted for 87% of all software supply chain poisoning attacks. CrowdStrike points to npm's deep dependency trees, its habit of auto-running install scripts, and JavaScript's broad popularity as reasons the ecosystem is such an attractive target.
Millions of users rely on packages distributed through the npm registry, and attackers are increasingly treating that inherent developer trust as a systemic weak point — targeting dependencies, tooling, registries and IDE extensions, and using automation to scale their reach.
The threat actor tracked as Altered Spider (TeamPCP) deployed a self-propagating worm earlier in 2026 that stole credentials and autonomously published further infected packages across both the npm and PyPI ecosystems, entirely without human intervention. A single set of stolen maintainer credentials was enough for the group to compromise more than 300 dependencies within a single day in May 2026.