On July 8, the head of the National Highway Traffic Safety Administration sent a letter to the autonomous vehicle industry saying that AV developers had shown a “clear pattern” of driverless cars blocking ambulances and fire trucks, ignoring flares and flashing lights, and in some cases driving directly into active emergency scenes.
While not technically naming Waymo, it was undeniably a focal player in the robotaxi reckoning - saying there was “a functional insufficiency” among automated vehicle developers regarding “a pattern of interference with first responders”.
At the same time, a different conversation has been unfolding inside the AI industry. Some of its leading figures have begun looking beyond engineering for answers. Anthropic recently launched its Faith & AI Covenant Roundtable to discuss how best to infuse morality and ethics into AI, while OpenAI's Head of Strategic Futures, Dean Ball, sparked debate after saying he had begun studying the Talmud to better understand AI policy.
Dr. Tal Cohen, co-founder of Drive TLV and managing partner of Next Gear Ventures, joined me to discuss these issues. In an interview recorded just after the NHTSA letter, Cohen used Waymo as the clearest example of what he calls “The Habitat”: the missing institutional layer of trust, permissioning, and accountability that must exist around an AI system before it’s allowed to act with consequential impact in the world.
“Two, three years ago, I was irrelevant, because there was no capability to talk like that,” Cohen said. But AI ability is expanding, and it is clearly starting to outpace the governance structure meant to contain it. “The capability is expanding,” he added, pointing to Waymo’s rapid deployment. “Suddenly, you have a gap between what the capability can provide and what we are lagging as a society.”
Cohen’s main argument is that the industry has spent its energy on the wrong bottleneck. Public debate about AI has focused overwhelmingly on physical infrastructure constraints related to chips, energy, and data centers, or on the capabilities of the models themselves. But he told me the actual constraint is a broader and less visible vacuum around the governance that authorizes an autonomous system to act, who reviews what it did, and who has the standing to revoke its authority when it gets something wrong.
“So then the question is: who’s gonna own the habitat that’s gonna authorize Waymo to go into crime scenes or not?” He argues society will hand over billions of consequential driving decisions before governments build the institutional framework capable of supervising them.
In his framing, regulators, municipal transportation departments, and NHTSA itself are unlikely to build that infrastructure fast enough on their own. He expects, and believes the moment demands, some hybrid of public and private coordination to define a constitution of sorts for autonomous systems. “We’re gonna live in the centuries or decades of habitat construction,” he said. “People really don’t get it yet.”
A manuscript he's been circulating, The Case for Habitat, argues this point exactly: That without a system or code in place, organizations face an uncomfortable choice. “You either put somebody in the basement, don’t let it do what it can do… or let it destroy your business,” he concluded.
In Cohen's words, the capability is “shiny,” but the Habitat is “boring.” Yet history suggests that the boring is often what determines which transformative technologies succeed.
