Thanks Yoav — this is an important post, and not just for the AI alignment community. What strikes me most is how CDV is becoming a connective methodology across domains that rarely talk to each other: chip verification, AV verification, robot SOTIF, and now AI alignment. Underneath all of them sits the same question safety people keep coming back to: how do you know what you don’t know? An explicit, evolving coverage map is the most practical answer I’ve seen for making that question tractable.
Coming from the SOTIF side (ISO 21448, scenario-based safety evaluation), I wrote an extension of your argument — mapping CDV onto the SOTIF four-quadrant model, connecting Layered CDV to the tree-like verification structure that robot SOTIF will need, and looking at what a standards-driven environment like China’s means for adoption: https://blog.autozyx.com/en/posts/robot-sotif-cdv-cross-domain/
Thanks Yuxin—and thanks especially for the description in your post of the mapping into “Robot SOTIF”, and how that might play in China’s standards-driven environment. You also wrote that if the coverage maps and risk assessments produced by CDV can become evidence of “reasonable care”—just like safety cases in autonomous driving—then alignment V&V gains an institutional incentive base.
That incentives part is outside my area of expertise, and yet it is crucial. Without it, rigorous alignment V&V loses to “ship faster” (in all jurisdictions). In AV-land what makes expensive, systematic V&V rational is the well-established “incident → investigation → someone is liable” loop, but there’s no AI analog yet.
The most promising hook I know of is the work trying to close the “responsibility gap” by attaching an AI agent’s actions to a human or corporate principal. Note this is mostly not aimed at the labs building general-purpose models, but rather at whoever deploys a specific AI-for-something (an AI CEO, a medical AI, a delivery robot) and thereby becomes the identifiable principal (and that specificity also makes the coverage map more tractable). If that holds, CDV-style coverage maps and risk-per-bucket claims can become exactly the “reasonable care” record such a principal would need.
If anyone reading this works on AI governance, algorithm assessment, or liability and sees a way to make rigorous V&V the path of least resistance rather than a cost center, I’d very much like to talk.
Thanks Yoav — this is an important post, and not just for the AI alignment community. What strikes me most is how CDV is becoming a connective methodology across domains that rarely talk to each other: chip verification, AV verification, robot SOTIF, and now AI alignment. Underneath all of them sits the same question safety people keep coming back to: how do you know what you don’t know? An explicit, evolving coverage map is the most practical answer I’ve seen for making that question tractable.
Coming from the SOTIF side (ISO 21448, scenario-based safety evaluation), I wrote an extension of your argument — mapping CDV onto the SOTIF four-quadrant model, connecting Layered CDV to the tree-like verification structure that robot SOTIF will need, and looking at what a standards-driven environment like China’s means for adoption: https://blog.autozyx.com/en/posts/robot-sotif-cdv-cross-domain/
To help the post reach more readers in the Chinese AV and robotics safety community, I have also published a full Chinese translation (with Yoav’s kind permission): https://blog.autozyx.com/posts/coverage-driven-alignment/
Looking forward to your talk at FISITA ISC 2026.
Thanks Yuxin—and thanks especially for the description in your post of the mapping into “Robot SOTIF”, and how that might play in China’s standards-driven environment. You also wrote that if the coverage maps and risk assessments produced by CDV can become evidence of “reasonable care”—just like safety cases in autonomous driving—then alignment V&V gains an institutional incentive base.
That incentives part is outside my area of expertise, and yet it is crucial. Without it, rigorous alignment V&V loses to “ship faster” (in all jurisdictions). In AV-land what makes expensive, systematic V&V rational is the well-established “incident → investigation → someone is liable” loop, but there’s no AI analog yet.
The most promising hook I know of is the work trying to close the “responsibility gap” by attaching an AI agent’s actions to a human or corporate principal. Note this is mostly not aimed at the labs building general-purpose models, but rather at whoever deploys a specific AI-for-something (an AI CEO, a medical AI, a delivery robot) and thereby becomes the identifiable principal (and that specificity also makes the coverage map more tractable). If that holds, CDV-style coverage maps and risk-per-bucket claims can become exactly the “reasonable care” record such a principal would need.
If anyone reading this works on AI governance, algorithm assessment, or liability and sees a way to make rigorous V&V the path of least resistance rather than a cost center, I’d very much like to talk.