On October 30, Kodiak AI said its Autonomy Readiness Measure had reached 99% for planned driverless long-haul operations. The company defines ARM as the proportion of materially completed claims and evidence in its safety case. The percentage does not measure driving accuracy, the probability of a safe journey or independent certification.
After completing the safety case, Kodiak plans operations without a person in the cab on Interstate 45 between Dallas and Houston by the end of 2026. That remains a future company plan. Approaching an internal milestone does not mean the highway service has already started.
How to read the percentage
A measure like this answers a question about progress in preparatory work. It does not independently tell us how difficult the remaining tasks are, whether every task has equal significance or how long completion will take. Consequently, 99% cannot be used to calculate a launch date or residual risk.
It is useful to separate three kinds of statement: what has been completed, what evidence supports safety and when commercial operation is expected. They can be connected, but one does not replace another. This is our editorial framework for reading the announcement, not an external vehicle assessment.
Where the evidence comes from
In a June 3 technical article, Kodiak describes Probabilistic Risk Assessment through scenario exposure, collision likelihood within that scenario and severity of consequences. BreakPoint searches for problematic cases in simulation and feeds results into the assessment. These are methods described by the developer, not calculations verified by SIExpo.
The activities answer different questions. Searching for a difficult scenario may identify a weakness; estimating its likelihood establishes how it fits the wider picture. Neither the number of discovered cases nor the mere existence of a simulator automatically provides public proof of acceptable risk.
Our proposed reading of a future report would look for a clearly defined operating environment, vehicle configuration and boundaries of the conclusion. A result for one route should not silently be extended to every road, load or condition. That is not a claim that Kodiak is making such an extension.
A meaningful assessment would also distinguish the certainty of a model's estimate from the amount of evidence behind it. Two neatly presented numbers can rest on very different assumptions. Keeping assumptions and unresolved questions visible would make an operational claim easier to assess; this is an editorial criterion rather than a finding about undisclosed company data.
What would strengthen the next announcement
A verifiable service launch, description of its actual scope and traceable operational results would add information different from a readiness percentage. Readers could then compare the promised milestone with what happened, without confusing testing and operation.
The distinction between company assessment and independent confirmation also matters. A detailed technical explanation is useful, but its authorship remains with the supplier. We do not present reading its pages as an audit, regulatory permission or proof of superiority over human drivers.
The development represents progress according to Kodiak's own reporting system. For the AI sector, the interesting element is an attempt to express evidence and remaining work as a traceable process. Practical safety of the planned service must be assessed through specific evidence and conditions, rather than the impressive percentage alone.
*Cover: AI illustration of a planned freight service and unfinished evidence dossier, not a photograph of an actual trip.*
*Editorial verification: 31.10.2026, 09:12:27 +0200 (Europe/Sofia).*
