A safety case is not a claim that an autonomous vehicle can never crash. It is a documented argument, backed by evidence, explaining why the system’s risks are understood, controlled and acceptable for the intended use.
What is normally inside a safety case?
The exact regulatory format varies, but a safety case can include the intended use and operational design domain, system architecture, hazards and risk controls, validation results, fallback strategies, monitoring processes, cybersecurity and software-change controls, and evidence that safety will be maintained through the system’s lifetime.
Why is it important for automated driving?
An automated driving system has to operate in an open, changing environment. Regulators therefore need more than a demonstration video or a total number of test miles. They need evidence explaining what the system is designed to do, how its risks were assessed and how developers know its performance is sufficient.
Safety case vs safety standard
A standard may specify engineering requirements or processes. A safety case brings together the reasoning and evidence for a particular system and use case. It can reference standards such as ISO 26262 or SOTIF, but the safety case itself is the system-specific argument.
UK example
In its 2026 automated-vehicle safety consultation, the UK government describes a safety case as documentation setting out the reasoning for how an automated driving system meets regulatory requirements and is free from unreasonable risk. It can include intended use, validation testing and lifetime safety processes.
What is an Autonomy Readiness Measure (ARM)?
Autonomy Readiness Measure (ARM) is Kodiak AI’s internal progress metric for a specific driverless deployment safety case. Kodiak says ARM measures the percentage of claims and supporting evidence in that safety case that are materially complete. It is useful as a project-readiness indicator, but it is not an industry-wide SAE automation level or a universal regulatory score.
On October 6, 2026, Kodiak reported that its ARM for long-haul driverless operations reached 96% at the end of September, up from 93% at the end of August. Kodiak says it is completing its highway-launch safety case for a planned unsupervised Dallas–Houston long-haul service before the end of 2026.
96% ARM does not mean “96% safe.” It means Kodiak reports 96% of the weighted claims and evidence in this particular safety case as materially complete. It is not a probability of avoiding a crash, a regulator’s approval percentage, or proof that 96% of driving situations have been solved.
Safety-case completion vs commercial driverless operation
Development status matters. Kodiak’s Dallas–Houston long-haul program is still in pre-launch validation: the company reports intervention-free end-to-end preparation runs with a safety observer onboard and targets unsupervised service by year-end. That is different from Kodiak-powered industrial trucks in the Permian Basin, where customer-owned trucks already operate commercially with no human in the cab. As of June 30, 2026, Kodiak reported 35 customer-owned driverless trucks in that industrial deployment.
This distinction also applies to Level 4 autonomous driving and robotaxi claims: testing, supervised operation, commercial service and fully driverless operation should not be treated as interchangeable statuses.
What remains before Kodiak’s highway launch?
Kodiak says its remaining work includes engineering verification and validation needed to close the long-haul safety case before driver-out operation. The company has not said that a 96% ARM itself authorizes deployment. CarGlossary will update this section if the safety case reaches completion or the Dallas–Houston service changes from supervised preparation to unsupervised commercial operation.
ARM and SAE automation levels are different
ARM describes completion of evidence for Kodiak’s deployment safety case. SAE driving automation levels classify which entity — human or automated driving system — performs the driving task and fallback under specified conditions. A high ARM percentage therefore does not turn a Level 2 or supervised system into Level 4.
How validation fits
Validation produces the test evidence. The safety case explains how that evidence supports the claim that the system is safe enough for the proposed operational domain and deployment.
Related terms
Sources
- UK Department for Transport / CCAV — automated vehicle safety principles and safety case definition
- UK government — 2026 statement of safety principles consultation
- Wayve — Safety 2.0 framework
- Kodiak AI — October 6, 2026 ARM 96% and long-haul safety-case update
- Kodiak AI — Dallas–Houston driverless launch lane and validation status
- Kodiak AI — commercial driverless Permian Basin fleet
