MathWorks expands TÜV SÜD certification to AI-assisted safety workflows
MathWorks has expanded TÜV SÜD certification for MATLAB and Simulink to cover AI-assisted workflows under ISO 26262, claiming a first among tool vendors in functional safety.

Why it matters
- On 18 November 2026 MathWorks expanded TÜV SÜD certification for MATLAB, Simulink, Stateflow and System Composer to include AI tool classification guidance under ISO 26262.
- MathWorks says it is the first tool vendor to provide such guidance for selected AI-assisted workflows in functional safety, based on use cases evaluated by TÜV SÜD.
- The certification covers Model-Based Design workflows using the Simulink Agentic Toolkit for model development, model review and test-case generation.
- The toolkits are built on Model Context Protocol and support agent platforms including GitHub Copilot, Codex, Gemini CLI and Claude Code.
- MathWorks, founded in 1984, employs more than 6,500 people across 34 offices and is headquartered in Natick, Massachusetts.
MathWorks has expanded its TÜV SÜD certification for MATLAB, Simulink, Stateflow and System Composer to include generic AI tool classification guidance under ISO 26262, covering selected AI-assisted engineering workflows in functional safety applications. Announced on 18 November 2026, the move addresses a question that has so far lacked a vendor answer: can AI agents participate in safety-critical automotive development without breaking the audit trail?
MathWorks said it is the first tool vendor to provide this type of guidance for selected AI-assisted engineering workflows in functional safety, based on specific use cases evaluated by TÜV SÜD. That claim carries weight because ISO 26262 — the automotive functional safety standard governing the development of electrical and electronic systems in vehicles — imposes documented processes on every tool that touches a safety-related design. What MathWorks has not certified is AI output itself; the certification addresses how AI-assisted tools are used, not whether their results are inherently trustworthy.
What does the certification actually cover?
The expanded certification covers products included with MathWorks' IEC Certification Kit and focuses on Model-Based Design workflows that use the Simulink Agentic Toolkit for three tasks: model development, model review and test-case generation. ISO 26262 requires engineering teams to classify their software tools according to the risk a tool error could introduce into a safety-related development process. Tool classification guidance helps teams perform that assessment for AI-assisted tools, something the standard predates by roughly two decades.
The scope is deliberately narrow. It includes work on Simulink models, model reviews and the generation of test cases tied to engineering processes. MathWorks frames the guidance as a way to show how AI-assisted workflows can enter safety-critical development in a controlled, auditable way — a phrasing that concedes the alternative, uncontrolled and unauditable, is what safety teams fear.
The takeaway: engineers now have a vendor-endorsed, assessor-evaluated path to use AI agents on ISO 26262 projects, but only within the specific workflows TÜV SÜD examined.
How do the agentic toolkits work?
MathWorks said its approach links agentic AI systems directly to Model-Based Design workflows rather than limiting their use to text prompts. Instead of a chat window, the AI agents operate on the engineering artefacts themselves.
The Simulink Agentic Toolkit is designed to let AI agents:
- interpret model structure,
- create and modify designs,
- run simulations,
- generate tests, and
- connect requirements to models.
The toolkit combines Simulink-specific tools with skills developed by MathWorks engineers and domain specialists for Model-Based Design workflows. Those skills cover model development, algorithm specification, plant modelling (building simulation models of the physical systems a controller interacts with), testing, validation and requirements generation.
One line in the announcement does the heavy lifting for safety practitioners: engineers remain responsible for validating outputs produced with AI assistance. Certification of the workflow does not shift accountability from the human to the tool.
Alongside the Simulink toolkit, the Polyspace Agentic Toolkit extends AI-assisted workflows into software analysis and verification. MathWorks said this includes support for code robustness analysis, coding-standard compliance and defect detection. Notably, the TÜV SÜD announcement names the Simulink workflows, not Polyspace, among the certified use cases.
In short: the toolkits put AI agents to work on models, simulations and code, while the certification covers a defined subset of that activity.
Why does this matter for the wider AI market?
The move comes as engineering teams assess how AI can be used in sectors where traceability, validation and review are mandatory. Safety-related industries such as automotive typically require documented processes showing how designs were produced, checked and approved — a requirement that generic AI assistants, which operate on free-form text, struggle to satisfy.
By tying AI systems to executable engineering assets such as models, simulations and tests, MathWorks is positioning its software within existing governance structures. Organisations can automate more development activity while keeping final decisions and approvals with engineers, the company said. That division of labour — machine does the drafting, human signs the release — matches how regulators and OEMs currently expect AI to enter certified processes.
The underlying framework is built on Model Context Protocol (an open standard for connecting AI agents to external tools and data), which, according to MathWorks, supports AI agent platforms including GitHub Copilot, Codex, Gemini CLI, Claude Code and other compatible agents. This gives users the option to incorporate the toolkits through AI platforms already used in engineering environments, rather than adopting a proprietary MathWorks assistant. It also means the certification's practical value depends partly on behaviour of third-party agent platforms that MathWorks does not control — a dependency the announcement leaves unexamined.
The stakes are commercial as much as technical. MathWorks develops mathematical computing software used across automotive, aerospace, energy and medical devices. MATLAB handles programming and numerical computation; Simulink handles simulation and Model-Based Design of multidomain and embedded systems. Founded in 1984, the company employs more than 6,500 people across 34 offices worldwide and is headquartered in Natick, Massachusetts. For a vendor whose customers build braking systems and medical devices, being first with certifiable AI workflows is a defensible position against both general-purpose AI tools and simulation rivals.
The takeaway: certification turns AI assistance from a compliance liability into a procurement question, and MathWorks has moved first to answer it.
What did MathWorks say?
"Engineering teams are looking for ways to benefit from agentic AI without sacrificing the rigour that system development demands," said Michael Carone, Principal Product Marketing Manager, Simulink Platform at MathWorks.
"By connecting AI agents to Simulink models, simulations and testing workflows, MathWorks is helping engineers apply agentic AI within established engineering processes while preserving the accountability required for complex engineered systems," Carone said.
Both statements stress preservation — of rigour, of accountability, of established processes. The marketing language sells continuity, not disruption, which is exactly what a functional-safety audience wants to hear. What the statements do not address is how AI-generated models or test cases behave when requirements change mid-project, or how TÜV SÜD's use-case evaluation generalises to workflows outside the certified scope. Those questions will land on engineering teams first.
The next test for this approach is adoption: whether automotive suppliers accept tool classification guidance for agentic workflows in actual ISO 26262 projects, and whether competitors follow with certifications of their own.
Source: itbrief.com.au
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
722 articles
Related articles
- OpenAI Reports Progress with US CAISI and UK AISI Safety Bodies
- OpenAI Calls for Safety Cases Before Frontier RL Training Runs
- White House Wants First Look at New OpenAI and Anthropic Models
- Australia to Create AI Safety Regulator With Penalty Powers
- OpenAI Publishes Frontier Governance Framework for AI Risk