ITMAX Embeds NVIDIA Cosmos 3 into SamurAI for Kuala Lumpur
ITMAX System Berhad said on September 23, 2026 that it is integrating NVIDIA’s Cosmos 3 world models and the Metropolis Video Search and Summarization blueprint into SamurAI, its computer-vision platform now rolling out in phases with Dewan Bandaraya Kuala Lumpur. Cosmos 3 powers SamurAI Inspector so roadside detections—stopped vehicles, restricted-area intrusions, shoulder smoke—are verified with contextual reasoning before operators see them; ITMAX claims more than a seventy-percent lift in alert-verification accuracy in the initial corridors. SamurAI Shogun agents on NVIDIA Nemotron are said to cut incident-search time by over ninety percent, while City Maestro inside City OS lets staff query traffic and incident systems in plain language via Model Context Protocol.
Filed under Infrastructure and dated September 23, 2026, this AI4Malaysia briefing treats the Cosmos showcase as Malaysian smart-city AI news distinct from election-manipulation probes or turbine-power analysis. ITMAX fine-tuned Cosmos 3 with TAO recipes on Malaysia-specific road footage and runs inference across H100, L20 and RTX PRO 6000 Blackwell servers spanning central halls and on-premise control rooms—an architecture meant to scale beyond Kuala Lumpur once DBKL phases validate.
Why it matters: Malaysian cities already drown in camera alerts. Reasoning layers can restore operator trust—but only if false-positive claims, privacy rules and human override remain auditable.
What it means in practice
Malaysian municipal OT and privacy leads should demand before/after false-positive dashboards; map which corridors enter phase two; assign an owner for footage retention and resident notice; run time-boxed shadow modes before auto-dispatch; and prefer kill switches that keep humans on enforcement actions. Place the rollout beside CelcomDigi’s Sophia agents and nationwide AI literacy drive.
Caveats come first. Vendor accuracy claims need independent audit; agentic city queries can hallucinate; and surveillance expansion raises rights questions. AI4Malaysia therefore presents the SamurAI–Cosmos stack as directional infrastructure context until published DBKL KPIs appear.
What to watch next: phase-two corridor lists; third-party accuracy tests; and other state councils adopting City Maestro. Readers can continue on the AI4Malaysia homepage, or browse the Newsroom for additional briefings.
Bottom line: treat this update as orientation, not instruction. Malaysian smart-city AI is shifting from detection to reasoned verification and remains under municipal supervision. Organizations that benefit most will insist on audit trails, keep humans on enforcement, and refuse to confuse a showcase with finished public safety.