The safety standards governing robots that share a workspace with people were written for stationary robotic arms bolted to a factory floor, not walking, balancing humanoid robots carrying energy-dense batteries — and 2026 has become the year the certification world scrambled to catch up, updating its foundational standards, folding old rules into new ones, and openly admitting that some of the hardest questions, like who's liable when a robot's AI misperceives something and causes damage, still don't have settled answers. This is a good-news, bad-news story: real progress is happening fast, but genuine gaps remain while robots are already being deployed.
Why Old Cobot Standards Don't Fit Humanoids
The existing framework for robots working near people, ISO/TS 15066, was built around collaborative robots (cobots) — typically stationary robotic arms — and defines biomechanical injury thresholds, speed limits, and separation-monitoring requirements for that specific kind of machine. Under those rules, a robot operating in close proximity to a human is limited to a maximum speed of 250 millimeters per second, a deliberately conservative constraint designed for a robot that isn't going anywhere, just reaching and gripping. Humanoid robots break that assumption entirely: they walk, they balance dynamically on two legs, and they carry batteries dense enough to pose their own separate hazard category, none of which the original cobot-safety framework was ever designed to address.
The 2025 Standards Overhaul
Recognizing the gap, standards bodies moved quickly through 2025. ISO 10218-1:2025 now includes basic cybersecurity requirements for the first time in the foundational industrial robot safety standard's history — a direct response to what the International Federation of Robotics has flagged as a rising number of hacking attempts targeting robot controllers and cloud platforms, a genuinely new risk category for humanoids that receive software updates, training data, and operational commands over network connections rather than operating as sealed, offline machines. Separately, the former standalone collaborative-robot specification, ISO/TS 15066, was fully absorbed directly into the main standard as ISO 10218-2:2025, consolidating what used to be two separate documents. Perhaps the most operationally significant shift: certification moved from covering hardware alone to covering specific applications — meaning a humanoid robot certified safe for one deployment and task isn't automatically certified for a different one, even with identical hardware, since the certification now has to account for the specific environment and task the robot is actually performing.
The Standard That Doesn't Exist Yet
Here's the honest gap worth naming directly: ISO 25785-1, the standard specifically meant to address the dynamic stability challenges unique to walking humanoid robots — how a bipedal machine's balance and fall risk should be assessed and certified — was still under development as of 2026, meaning it doesn't yet formally exist even as humanoid robots walk factory floors today. Until it's finalized, deployments are working from a patchwork of adjacent standards (ISO 10218 for industrial contexts, ISO 13482 for personal-care and service robots operating around the general public, ISO 13849 for the functional safety of control systems like emergency stops) rather than a single purpose-built framework for the specific hazard walking robots introduce.
The Liability Question Nobody Has Fully Answered
One robotics industry analysis framed the honest state of things bluntly with three scenarios: when a production worker shares a body shop with a humanoid for a ten-hour shift, who's responsible if the robot falls? When a company deploys 1,000 humanoid units across a factory, what certification did each individual unit actually receive before powering up alongside human workers? And when an autonomous robot running an AI model acts on a misperception and damages equipment, is the liability with the hardware manufacturer, the software developer, the AI model's training company, or the facility operating it? As of 2026, the honest answer to all three is "it depends," with the legal frameworks needed to give a definitive answer still being actively constructed rather than settled. That's not a reason deployments have stopped — workers are collaborating with humanoid robots today, under real if imperfect guidance — but it's a genuine, acknowledged gap enterprise buyers and facility operators need to understand rather than assume away.
New Certified Hardware Filling the Gaps
Some of the most concrete recent progress has come from specific component-level certifications rather than the broad standards themselves. Sonair's ADAR One sensor, described as the world's first safety-certified 3D ultrasonic sensor built specifically for human-robot collaboration, achieved SIL 2 and Performance Level d ratings — safety certification tiers historically met only by laser scanners and vision-based systems, the same category of sensors covered in an earlier entry of this series. A certified ultrasonic option gives facility integrators a genuinely useful alternative in environments where dust, glare, or visual occlusion make optical sensors unreliable, exactly the kind of real industrial condition a demo-video-perfect showroom doesn't reveal. Separately, Yaskawa secured ISO/IEC 27001:2022 certification, the global benchmark for information security management systems, addressing the cybersecurity side of deployment rather than the physical-safety side.
Humanoid Robot Safety Standards at a Glance
| Standard | What it covers | Status (2026) |
|---|---|---|
| ISO 10218-1:2025 | Industrial robot safety, now including cybersecurity requirements | Active, updated 2025 |
| ISO 10218-2:2025 | Collaborative robot safety (absorbed former ISO/TS 15066) | Active, updated 2025 |
| ISO 25785-1 | Dynamic stability for walking/bipedal humanoid robots | Still under development, not yet finalized |
| ISO 13482 | Personal-care and service robots in public/non-industrial settings | Active |
| ISO 13849 | Functional safety of machine control systems (e-stops, safety zones) | Active |
Frequently Asked Questions
Are there safety standards specifically designed for walking humanoid robots?
Not yet fully. ISO 25785-1, which addresses the dynamic stability and fall-risk challenges unique to bipedal humanoid robots, was still under development as of 2026. Current deployments rely on a patchwork of adjacent standards originally designed for stationary collaborative robots.
Who is legally liable if a humanoid robot causes an accident?
As of 2026, there's no fully settled answer. Industry analysts describe the honest state as "it depends," since legal frameworks addressing liability among hardware manufacturers, software developers, AI model trainers, and facility operators are still actively being constructed.
What changed in the 2025 update to ISO 10218?
ISO 10218-1:2025 added basic cybersecurity requirements for the first time, addressing rising hacking attempts against robot controllers and cloud platforms. ISO 10218-2:2025 absorbed the former standalone collaborative-robot standard, ISO/TS 15066, directly into the main framework.
How fast can a collaborative robot move near a human worker?
Under the traditional collaborative robot framework, robots operating in close proximity to humans are limited to a maximum speed of 250 millimeters per second, a conservative limit originally designed for stationary robotic arms rather than walking humanoids.