May 25, 2026, 11:05 GMT | Comment
Autonomous AI systems are rapidly moving beyond software into robots, vehicles and critical infrastructure, creating governance risks that existing AI frameworks were never designed to manage. Discussions at a recent AI summit in Singapore highlighted growing concern that embodied AI introduces operational safety problems more commonly associated with aviation and industrial systems than traditional software regulation. Industry and academic leaders warned that failures involving autonomous physical systems could carry real-world consequences, pushing governments toward governance models built around continuous testing and operational monitoring rather than one-time certification alone.
Autonomous AI systems are beginning to move out of controlled software environments and into warehouses, delivery networks, transport systems and public spaces. That shift is creating governance problems that existing AI rules were not designed to handle.
Most current AI governance frameworks were built around online harms and model outputs — biased responses, misinformation, harmful content or disclosure obligations. Embodied AI systems introduce a different category of risk altogether: autonomous systems operating continuously inside physical environments where failures can disrupt infrastructure, damage property or injure people.
At an AI summit in Singapore last week, discussions around robotics and embodied AI reflected a growing realization that regulators are increasingly confronting operational safety problems more commonly associated with aviation, industrial systems and critical infrastructure oversight than conventional software regulation.
The central governance challenge is no longer simply whether autonomous systems can be built. It is whether they can operate safely and reliably inside unpredictable real-world environments over extended periods of time.
— Amplified risks —
Dr. Ya-Qin Zhang, founding dean of the Institute for AI Industry Research at Tsinghua University, warned that embodied AI systems amplify the risks already associated with autonomous software systems because failures can directly affect transportation systems, drones, logistics networks and critical infrastructure.
“Any risk in the digital domain will be amplified in the physical domain, and the physical domain will have a physical consequence," Zhang told MLex on the sidelines of the summit.
“Your car could be all controlled, your drone could be all controlled,” he added.
He warned that as AI systems become increasingly embedded inside smart grids, transportation systems and other critical infrastructure, failures in the digital domain could rapidly cascade into physical-world disruption.
The result is that reliability, operational monitoring and post-deployment assurance are becoming central governance concerns rather than secondary technical issues.
The discussions at the summit reflected a broader shift toward deployment-based governance models built around simulation, telemetry or operational data collection, operational monitoring and iterative testing rather than one-time certification alone.
That approach increasingly aligns with Singapore’s own positioning as a real-world deployment and testing hub for autonomous AI systems. Through initiatives including robotics testbeds in Singapore’s Punggol district and broader AI deployment sandboxes, policymakers are emphasizing operational experimentation and iterative governance alongside formal regulation (see here).
— Real-world deployment —
The governance challenge becomes particularly visible once autonomous systems move into real-world deployment.
Grab, which is piloting autonomous vehicles and delivery robots in Singapore’s Punggol district, described deployment governance as heavily dependent on simulation, testing and continuous operational monitoring.
“We do a lot of simulation, we do a lot of testing in closed courses and open courses in order to make sure our robots are reliable,” Suthen Thomas Paradatheth, the chief technology officer of Grab, told one of the panels.
“Before we scale to hundreds of robots, we make sure we crack it first in simulation and with a few robots.”
The company also emphasized continuous monitoring after deployment, including instrumentation systems designed to track robot performance and identify unexpected operational failures.
“There’s a long tail of issues that could emerge,” Paradatheth said.
That operational structure complicates traditional accountability models.
Embodied AI systems increasingly rely on layered ecosystems involving AI developers, robotics manufacturers, semiconductor suppliers, infrastructure operators and platform companies. The result is that operational responsibility becomes fragmented across multiple actors, blurring distinctions between product liability, software liability and infrastructure accountability.
Those questions become even harder where autonomous systems continue adapting after deployment through software updates, telemetry and real-world operational data.
For Applied Materials, the deployment challenge is equally tied to semiconductor economics and systems integration.
Om Nalamasu, Applied Materials' chief technology officer, said large-scale robotics deployment will depend on improvements in sensors, energy efficiency, advanced packaging and computational architectures, while also requiring safer and more reliable systems across different operating environments.
He said robotics systems would increasingly require purpose-built designs adapted to specific industrial ecosystems rather than “one silver bullet” solutions.
— Diverging approaches —
The discussions also showed how Asian governments are seen to be diverging in their approaches to embodied AI governance.
Zhao Yuli, the chief strategy officer of Chinese robotics startup Galbot, said Beijing is prioritizing deployment scale and industrial commercialization, supported by government-backed testbeds, industrial partnerships and long-term funding initiatives.
Galbot has already deployed humanoid robotics systems across retail, warehouse and pharmaceutical operations in China, including autonomous stores operating around the clock. Zhao said semi-structured industrial environments are likely to become the first major commercialization pathway because they provide more controllable operating conditions.
Japan, meanwhile, appears more focused on standards-setting, robotics datasets and safety governance. Professor Yutaka Matsuo of the Graduate School of Engineering at the University of Tokyo pointed to Japan’s efforts to collect and share large-scale robotics datasets through initiatives including the country’s "AI Association" project, which aims to gather "100,000 hours robotic data" to support development of robotic foundation models.
Matsuo also emphasized that “safety is very important,” referencing both Japan’s AI Safety Institute and the Hiroshima AI Process as part of broader efforts to develop governance standards for embodied AI systems together with Singapore and other Asian countries.
Singapore, meanwhile, is increasingly positioning itself as a coordination platform for international AI governance and deployment discussions.
Across the summit, labor shortages, aging populations and industrial productivity pressures repeatedly emerged as major drivers behind embodied AI deployment across Asia. But the governance architecture surrounding those systems remains significantly less mature than the technology itself.
That regulatory gap may become one of the defining problems of the next AI cycle: governing autonomous systems operating continuously inside physical environments where failures increasingly carry operational and real-world safety consequences.
*ATxSummit 2026, Capella Singapore, May 20-21, 2026
Please email editors@mlex.com to contact the editorial staff regarding this story, or to submit the names of lawyers and advisers.
Tags
Sections:
Artificial Intelligence
Industries:
Agriculture, Forestry, Fishing & Hunting, Automotive, Banking & Finance, Construction, Computing & Information Technology, Defense & Aerospace, Energy & Utilities, Environmental Industry, Health Care, Manufacturing, Media & Telecommunications, Professional Services, Transportation & Warehousing, Travel, Hospitality & Tourism
Geographies:
Asia, Singapore, South-Eastern Asia
Topics:
Cybersecurity