Eastern open-sourceartificial intelligence models have reached parity with leading Western counterparts in offensive cyber capabilities while operating at roughly a fifth of the cost, according to findings from Ensign InfoSecurity.
The evaluation, detailed in Ensign's 2026 Cyber Threat Landscape Report, challenges long-held industry assumptions that top-tier cyber threat actors would predominantly rely on Western-developed foundation models.
Testing frontier models in simulated environments
Ensign evaluated more than 150 AI models before shortlisting the ten best-performing frontier systems, comprising five Western and five Eastern models. Each model assumed the role of an advanced cyber threat actor within Ensign's AI Cyber Range.
The assessment tested the models inside a simulated university Learning Management System (LMS) portal equipped with standard defensive controls and exploitable vulnerabilities. Each candidate underwent 16 test runs across eight standardized attacker objectives, spanning initial perimeter breach to establishing persistent footholds.
Across the 160 total test runs, three frontier models emerged as top performers:
OpenAI's GPT-5.6 Sol
Anthropic's Claude Opus 4.8
Z.AI's GLM-5.2
All three systems achieved High Success ratings in seven out of the eight defined offensive objectives. However, Z.AI's open-source GLM-5.2 matched the offensive performance of GPT-5.6 Sol while requiring approximately 20% of the operational expenditure, demonstrating that open-weight architectures deliver significantly higher attack capability per dollar invested.emergent+2
Detection evasion remains a defensive barrier
Despite strong performance across breach scenarios, the assessment revealed operational bottlenecks. While all evaluated models secured initial system access with ease, none achieved reliable network detection evasion.
Even the three top-performing systems—GPT-5.6 Sol, Claude Opus 4.8, and GLM-5.2—attained only partial success when attempting to bypass internal network monitoring. For cybersecurity defenders, this operational limitation highlights that internal behavioural telemetry and detection controls remain effective intervention points against autonomous AI-driven intrusions.
The rapid convergence of global AI capabilities underscores a shifting economic landscape for enterprise defence.
Jeremy Moke, EnSOC Director at Ensign InfoSecurity Malaysia, said: "In our testing, comparable offensive capability was available at roughly a fifth of the price, and defenders should assume the pool of adversaries will keep growing.
Jeremy Moke
"Frontier AI is changing the speed, scale and economics of cyberattacks with capability advancing on an estimated two-month cycle, and organisations need to continuously reassess how they measure cyber risk and resilience, not just how much they spend on controls." Jeremy Moke
