Threat actors are beginning to ‘weaponise’ AI to scale their operations and accelerate attacks, a new report from Crowdstrike has warned, and are increasingly targeting the autonomous AI agents reshaping enterprise operations.

According to the cyber firm’s 2025 Threat Hunting Report, which tracked more than 265 named adversaries, multiple threat actors have been observed gaining unauthenticated access, establishing persistence, harvesting credentials, and deploying malware against the tools used to build AI agents.

CrowdStrike said that in April, its researchers found cyber-attackers exploiting CVE-2025-3248, an unauthenticated code injection vulnerability in Langflow AI, a widely used tool for building AI agents and workflows.

The threat actors were reportedly able to exploit this vulnerability using specially crafted HTTP requests to achieve unauthenticated remote code execution, gaining unauthenticated access, establishing persistence, harvesting credentials, and deploying malware.

In sophisticated, multipronged attacks, cybercriminals were found dumping password and shadow files as well as targeting cloud environments, deploying cryptomining malware, multi-stage remote access tools and ransomware.

Malware, Deepfakes, and Fake Resumes

Fuelling this crimewave is genAI. Crowdstrike found that genAI-powered tradecraft is transforming traditional threats into scalable, persistent operations that can threaten entire industries, not just sole enterprises.

The most notable example is the North Korean threat group dubbed ‘Famous Chollima’, which used genAI to automate every phase of its insider attack operation, including building fake resumes and conducting deepfake interviews, to completing technical tasks under false identities. 

Other instances Crowdstrike identified of genAI being used to power large-scale cyber-attacks included the Russia-backed Ember Bear group utilising it to amplify pro-Russia narratives, and the Iranian linked adversary Charming Kitten, which deployed LLM-crafted phishing lures targeting US and EU organisations.   

Even at the lower-tier or criminal operations, cyber-attackers and hacktivists are abusing AI to generate scripts, solve technical problems, and build malware – automating tasks that once required advanced expertise. 

Crowdstrike said that malware families like Funklocker and SparkCat have leveraged genAI to write malicious code using unrestricted LLMs and create AI-powered optical character recognition, early proof points that AI-built malware is no longer theoretical.

Recommended reading

• Majority of Companies Already Past AI Agent Experimentation Phase
• Report: AI Agents Are Triggering Business Restructure
• 80% of Firms Say Their AI Agents Have Taken Rogue Actions

However, Crowdstrike said that although the technology has enhanced threat actors abilities, genAI alone is not likely to benefit threat actors operations.

At the moment, AI-generated code still requires human expertise to be effective, meaning that more sophisticated users are likely to maintain an edge over those with less technical ability, and these tools depend on attackers having system availability, offensive capabilities, and operational integration.

“The AI era has redefined how businesses operate, and how adversaries attack,” said Adam Meyers, head of counter adversary operations at CrowdStrike.

“We’re seeing threat actors use genAI to scale social engineering, accelerate operations, and lower the barrier to entry for hands-on-keyboard intrusions. At the same time, adversaries are targeting the very AI systems organisations are deploying.

“Every AI agent is a superhuman identity: autonomous, fast, and deeply integrated, making them high-value targets. Adversaries are treating these agents like infrastructure, attacking them the same way they target SaaS platforms, cloud consoles, and privileged accounts. 

“Securing the AI that powers business is where the cyber battleground is evolving.”