The thread running through the week
Four stories logged on this desk in two days look unrelated, but they share one pattern: artificial intelligence is moving from software into the physical world, and the institutions that govern physical risk are not keeping pace. That mismatch is now the defining problem for US technology companies. The questions being asked are no longer about model quality; they are about who is accountable when an AI system drives a car, runs a factory, or shapes a market.
Antitrust as a safety question
On Decoder, The Verge interviewed Jonathan Kanter, the former antitrust chief for the US Department of Justice in the Biden administration, now a professor of law at WashU and professor of technology policy at Carnegie Mellon. The framing of the episode - whether AI needs an antitrust exemption so it does not kill everyone - is deliberately extreme, and Kanter's presence gives it weight. A former enforcement official discussing exemptions is a signal that the policy debate has shifted from whether concentration is a problem to whether the usual remedies are too slow for the technology.
The subtext matters for US companies. An exemption would be a bet that scale and coordination produce safety faster than competition does. The alternative, which Kanter has spent a career enforcing, is that concentrated control over critical infrastructure creates its own risks. For US technology firms, the practical consequence is that antitrust review is becoming a venue for AI safety arguments, whether or not anyone intended it that way.
Robotaxis and the cost of physical failure
Waymo restarted its San Antonio service five months after flooding troubles, as TechCrunch reported. The company suspended service in April after one of its robotaxis got swept away. The pause is the important part. A software bug can be patched in an afternoon; a vehicle lost to a flood cannot, and the reputational and regulatory consequences of a visible physical failure are orders of magnitude larger.
This is the compliance test in its purest form. Waymo did not merely fix a defect; it had to persuade a city, a regulator and a riding public that the system understood a hazard it had already misjudged. That is a slower, more expensive and more public process than shipping a model update. For US consumers, it means autonomous mobility will arrive unevenly, city by city, shaped by weather events and local tolerance for risk rather than by technical readiness alone.
A startup factory betting on industrial AI
TechCrunch reported that UP.Labs, now doing business under the name Vantora, raised $100 million and is all-in on physical AI, building startups for industrial corporations. The funding figure is notable, but the structure is more interesting. Vantora is not selling software to factories; it is manufacturing companies alongside the incumbents that own the factories.
That model is a direct response to the accountability problem. Industrial customers will not deploy AI into a production line on the strength of a benchmark. They need a counterparty with a balance sheet, a maintenance contract and someone to call when a machine stops. By wrapping AI in a corporate vehicle co-owned with the industrial partner, Vantora is effectively selling institutional trust as much as technology. If the model works, it suggests the winners in physical AI may be defined by their corporate structures as much as their models.
