ByteDance's Seed team steers the company into autonomous driving
Physical AI is pulling the big platform players toward the road: ByteDance is lining up a driving effort built on its world models, and the companies already putting robotaxis on city streets are now fighting about the rules that govern them.
ByteDance is moving into autonomous driving, with its world-model research team spearheading an early-stage push that sources say is aimed at unmanned logistics. The effort is being led by the world-model group inside Seed, ByteDance's flagship large-model division, under Zhou Chang, who also oversees the Seed Robotics team, according to 36Kr's exclusive report. ByteDance gave a cautious denial when asked — it does "a great deal of early-stage research in frontier AI areas, including physical AI, but has no plans to develop an intelligent driving business" — and the project is described as being in preparatory stages, with the company already approaching senior talent from the assisted- and autonomous-driving industry. The timing makes sense: world models have emerged as the dominant technical paradigm for driving in 2026, with Geely, Momenta, Xiaomi, and XPeng converging on the approach, and ByteDance's existing world-model work could be retrained for vehicle use given its GPU firepower and traffic data. Industry observers see a deeper motive too — real driving data is increasingly treated as critical feedstock for embodied intelligence, the same playbook that Tesla's FSD road data supposedly powers for its Optimus robot, so autonomous driving may double as a proving ground for ByteDance's broader physical-AI ambitions. The project would land under Volcengine, which has run an automotive vertical since 2020 and already ships the Doubao cockpit assistant.
Robotaxis are real now — and so is the pushback over how they're governed. In The Verge's survey of a fight that has escalated from science fiction into city policy, New York Governor Kathy Hochul withdrew a proposal that would have opened the door to driverless service, Washington DC lawmakers are weighing a 200-vehicle cap and a per-mile fee, San Francisco's mayor is calling for tougher rules after Waymos snarled traffic at emergencies, and NHTSA warned companies that vehicles which can't safely interact with first responders are a danger to the public. The market leader Waymo now operates more than 3,500 vehicles across 11 US cities and logs over 500,000 paid rides a week; Zoox just began charging for rides in Las Vegas; and Tesla is growing its unsupervised fleet ahead of the Cybercab launch this month. Notably, even the robotaxi companies want federal rules — Zoox's CEO said on Fox Business that "we need to be regulated" — partly to avoid the patchwork of state and local laws now under construction. Robotaxis are still a rounding error next to Uber's 40 million daily trips, but the fights underway are about the terms of expansion, not the tech's existence.
AI-generated appeals are starting to flood government services — a new paper maps where the pressure will hit hardest. The study, headed to the AAAI Conference on AI, Ethics, and Society, documents 84 potential cases of what the authors call "agentic flooding" across 11 jurisdictions, driven mostly by LLMs generating text cheaply for benefit claims, policy queries, and public comments. Their risk matrix suggests near-term exposure is greatest for financially attractive but complex services, where the payoff of an AI-generated application is highest relative to the friction of filing it. The uncomfortable finding is that the fastest fixes governments reach for — friction-inducing measures like application fees — directly trade off equitable access to public services. The authors recommend near-term mitigation that avoids that tradeoff, but the deeper question is whether public institutions can scale their intake before AI scales the demand against them.
If every driver can script a thousand benefit applications with a chatbot, should governments answer with better service — or barriers? Tell us in the comments.
