Physical AI is most fragile at the boundary between simulation and real-world consequence.
PRiTLabs is an open framework for scoring the safety and security of physical AI — from how accurately a system was trained, to whether it's ready to deploy, to what happens after it's live around people.
Industrial robots are scored on task success. Rarely on physical consequence.
Industrial robots and cobots can run codebases with millions of lines and an attack surface to match — in many deployments closer to industrial control systems than to the chatbots most AI safety work is built around. A safe-looking software state is not automatically a safe physical state when a machine operates around people. The gap between a training simulation and real-world consequence still lacks a widely adopted, open way to score realism, gate deployment, and keep testing once a system is live.
Existing standards address important, adjacent parts of this problem, but not this exact layer. ISO/IEC TS 22440 is a broad AI functional-safety framework still in development; ISO 25785-1 is a draft standard for industrial mobile robots with actively controlled stability; and ISO 10218 addresses safe design and protective measures for industrial robots. PRiTLabs is intended to complement that ecosystem by focusing on simulation realism, deployment evidence, and continuous post-deployment physical-reality testing.
Industrial physical AI can span edge, cloud, and hybrid architectures, each creating its own software, networking, and change-management risk.
Perception failures and state-estimation errors do not stay virtual. A system that loses track of a person or object can act on that mistake physically.
No widely adopted, open framework yet scores sim-to-real realism, gates deployment on that score, and monitors it continuously once a system ships — though PRS and PRiT have been proposed to help fill that gap.
Three instruments, one lifecycle
Physical Reality Score
Scores how accurately a digital twin or simulator reproduces real-world physics before it's ever used to train a system — an indicative number for expected sim-to-real gap downstream.
Physical Reality Interaction Test
A structured pre-deployment test, modeled on the site acceptance test used in industrial control systems. Produces a realism/risk score intended to inform deployment readiness.
Continuous Physical Reality Interaction Testing
Ongoing testing once a system is live, feeding real performance data back into the digital twin — closing the loop so the next system trained on it starts closer to reality.
Designed to complement, not replace
PRiTLabs is not intended to replace existing standards. It is an open, vendor-neutral framework designed to work alongside them by adding practical methods for evaluating simulation realism, documenting deployment-readiness evidence, and supporting ongoing validation of Physical AI systems across operation.
| Framework / Standard | Purpose | What it contributes | How PRiTLabs complements it |
|---|---|---|---|
| IEC 61508 | Functional safety lifecycle | Safety requirements for E/E/PE systems | PRiTLabs can be used alongside this standard to add simulation realism scoring and Physical AI validation methods |
| IEC 61511 | Process industry safety | Safety instrumented systems | PRiTLabs can be used alongside this standard for AI-enabled physical systems in process environments |
| ISO 10218 | Industrial robot safety | Robot design and safeguarding | PRiTLabs can be used alongside this standard, offering deployment-readiness and runtime-validation practices for AI-enabled behavior |
| ISO/IEC TS 22440 | AI functional safety | AI safety concepts | PRS, PRiT, and CPRiT can provide operational scoring instruments used alongside this framework for physical-AI evaluation |
| PRiTLabs (Open Framework) | Open engineering framework | PRS, PRiT, CPRiT lifecycle | Works alongside existing standards rather than replacing them |
Open framework. Open governance.
Read and cite the full PRS / PRiT / CPRiT specification
- Comment on drafts as they're published
- Adopt the framework internally, no permission required
Shape the specification before it's finalized
- Voting seat in the working group
- Early access to reference scoring tools
Foundational sources
PRiTLabs builds on prior published discussion of physical AI safety and security, including a 2025 Control Engineering article introducing the PRS/PRiT concepts, and on subsequent acknowledgment that testing and validation frameworks are still needed for public-facing robotics. These sources help ground the framework’s focus on real-world consequence, deployment readiness, and ongoing oversight.
Help shape this standard together.
PRiTLabs is inviting practitioners and researchers to collaborate on an open, practical framework for physical AI safety and realism — with industrial robotics as the starting point and other consequence-heavy domains on the roadmap.