CrowdStrike 2026 Threat Report: Why foundational AI Models are Cybercriminals' top target

Cybersecurity provider CrowdStrike has released its 2026 Technology Threat Landscape Report, detailing a shift across global espionage networks. Frontline telemetry compiled by the company's Counter Adversary Operations team indicates that foundational artificial intelligence building blocks, algorithmic logic systems, and model configurations now comprise highly targeted enterprise intellectual property.

Because critical generative computing innovations remain concentrated within private technology corporations, the technology sector has become the most targeted industry globally. According to the report, state-sponsored actors are actively attempting to bypass traditional development timelines by exfiltrating underlying software architectures from these firms. 

China-Nexus Actors Focus on Industrial AI Theft

The data reveals that China-nexus threat groups remain a primary driver of highly focused corporate targeting campaigns, accounting for more than 58% of state-sponsored targeted intrusions against the technology sector. Monitored adversary groups—including MURKY PANDA, MUSTANG PANDA, OVERCAST PANDA, SUNRISE PANDA, and WARP PANDA—consistently prioritised technology targets over other commercial fields.

The scale of these industrial espionage operations is illustrated by a single password-spraying campaign conducted by MURKY PANDA, which compromised or impacted more than 340 entities based in the United States. Threat intelligence notes that this systematic target profiling operates effectively as cyberespionage as an industrial policy, aimed at closing the domestic artificial intelligence innovation gap by absorbing external breakthroughs.

DPRK Operational Infiltration and Automated eCrime Extraction

Concurrently, the report highlights an acceleration in non-traditional entry methods and monetizable extortion operations across alternative state-backed and criminal syndicates:

• AI-Enhanced Personas: North Korean threat group FAMOUS CHOLLIMA utilised artificial intelligence to generate fraudulent personnel profiles and operated through domestic front companies to secure remote IT employment within western technology infrastructure. This operational vector accounted for 47% of all state-sponsored interactive intrusions observed within the sector, channelling employment revenue directly to state programs.

• Financially Motivated Ransom Operations: Financially motivated eCrime operations constituted 65% of all hands-on-keyboard operations targeting tech entities. Initial access brokers increased their listings by nearly 30%, advertising active entry points into 277 technology firms, while Big Game Hunting (BGH) adversaries listed 572 separate technology organisations on public data leak sites for extortion.

• Machine-Speed Log Evasion: Advanced cybercriminals are increasingly adopting automated scripting to speed up extraction times. Adversaries now deploy custom AI-generated code to execute rapid credential dumping and erase local forensic telemetry at machine speed, drastically shortening the detection and response window available to enterprise defenders.

Poisoning Open-Source Developer Supply Chains

A significant finding in the report involves threat actors shifting focus upstream to poison open-source developer ecosystems, multiplying their impact across millions of downstream applications. Threat group STARDUST CHOLLIMA successfully compromised the Axios NPM package—a foundational component downloaded approximately 100 million times per week—exposing downstream enterprise pipelines to systemic vulnerability.

Additionally, prior to public infrastructure disruption executed by security researchers, operators behind the Glassworm botnet compromised 350 distinct GitHub repositories. The actors injected malicious code blocks directly into active JavaScript and Python environments, establishing persistent beachheads inside corporate build systems.

Adversaries are also targeting the enterprise adoption of macOS endpoints within engineering circles. A newly identified information stealer variant, designated Skrawl, was observed spreading through fraudulent OpenClaw extensions and counterfeit download networks mimicking authentic AI tools to harvest credentials from local developer environments.

"Every AI breakthrough creates a competitive advantage and new attack surface at the same time,” stated Adam Meyers, Head of Counter Adversary Operations at CrowdStrike. He noted that because artificial intelligence capabilities have become a core prize for adversaries, organisations must ensure security configurations are natively embedded from the initial stages of code development and framework adoption.

Read More:

How AMD's AI strategy is opening new growth avenues for channel partners

How Hybrid Cloud Complexity Is Driving Demand for Managed Services