CrowdStrike and Thales have released their global threat reports of the year, which both, separately, describe AI as accelerating adversary operations while also expanding the enterprise attack surface and complicating data security and governance.
The CrowdStrike report, tracking more than 280 named adversaries, found the average eCrime breakout time, the window between initial access and lateral movement, fell to 29 minutes in 2025, a 65% acceleration from the prior year. The fastest observed breakout took 27 seconds. In one documented intrusion, data exfiltration began within four minutes of initial access.
The Thales report, based on S&P Global 451 Research data spanning 20 countries, describes the same dynamic from the defense side: 70% of organizations cite the speed of AI-driven change as their top security risk, and 53% acknowledge their existing security budgets are still funding their AI security.
Identity as the common fault line
Both reports converge on identity as the most exposed fault line. CrowdStrike found that AI-enabled adversary activity and credential theft rose 89% year over year, with nation-state and eCrime actors using AI to automate reconnaissance, accelerate credential dumping and erase forensic evidence.
Thales found credential theft remains the leading attack technique against cloud management infrastructure, cited by 67% of organizations that experienced cloud attacks. Even after that, the same report found only 47% of sensitive data in the cloud remains encrypted.
Two threat models, one attack surface
CrowdStrike’s focus is on external adversaries weaponizing AI: Russia-nexus FANCY BEAR deployed LLM-enabled malware to automate document collection; eCrime actor PUNK SPIDER used AI-generated scripts to accelerate credential dumping; North Korea-linked FAMOUS CHOLLIMA scaled insider operations through AI-generated personas.
The report also documented adversaries injecting malicious prompts into GenAI tools at more than 90 organizations and exploiting AI development platforms to deploy ransomware, treating enterprise AI infrastructure itself as an attack surface.
The risk already inside the perimeter
Thales, however, shifts the lens inward. Its concern is the AI the enterprise has already deployed: automated systems granted broad access to sensitive data, often with fewer controls than those applied to human users.
Only 34% of organizations surveyed know where all their data resides, and just 39% can fully classify it. Against that visibility deficit, nearly 59% of organizations have already experienced deepfake-driven attacks, the report found, and 48% report reputational damage from AI-generated misinformation.
However, the biggest concern remains that only 30% have allocated dedicated AI security budgets, a figure Thales set against an attack surface that expanded substantially in 2025.