General Motors is taking a big, very public step toward its “eyes-off” self-driving future, moving its newest automated driving tech from closed tests to real highways in California and Michigan. The company is still keeping a human in the driver’s seat for now, but this is the first time its next-gen system is mixing with everyday traffic at scale, not just collecting data in the background.

GM says these development cars have already logged more than a million miles across 34 U.S. states just to train the system, on top of millions more run through simulation, where engineers can effectively “fast‑forward” about 100 years of driving scenarios every day. Now, that homework phase is giving way to supervised public-road testing with over 200 vehicles, each with a trained safety driver ready to grab the wheel at any moment if something looks off. It’s a deliberate, safety‑first rollout designed to show regulators and drivers that the tech can handle the chaos of real traffic, not just carefully controlled test loops.

This push is part of GM’s larger plan to launch what it calls “eyes-off” driving in 2028, starting with the all-electric Cadillac Escalade IQ before expanding to more Cadillac, Chevrolet and other GM models. “Eyes-off” is important wording: unlike today’s hands-free systems that still require the driver to watch the road, GM is targeting a higher level of automation where the car can handle driving in its defined domain—initially, highways—without constant human supervision, at least under certain conditions. The backbone for that will be a new centralized computing platform that bundles everything from powertrain to safety into a single high-speed “brain,” promising up to 35× more AI performance and 1,000× more data bandwidth compared to current setups.

GM also has a quietly powerful advantage: real-world experience. Its Super Cruise system, already on 20‑plus GM vehicles, has logged more than 700 million hands‑free miles on about 600,000 miles of mapped roads in the U.S. and Canada, with no crashes officially attributed to the tech, according to the company. Add in roughly 5 million fully driverless miles from the now‑restructured Cruise robotaxi program in dense city environments, and GM is training the new system on a massive library of both human-driven and computer-driven behavior.

For everyday drivers, none of this means you can let your car chauffeur you to work tomorrow and watch Netflix from the driver’s seat. In the near term, these supervised tests are about refining the AI, improving how the system handles edge cases, and convincing regulators that GM’s approach is responsible. But if the company hits its 2028 target, the tech now running quietly under the watchful eye of test drivers on California and Michigan highways could evolve into a consumer feature that lets future Cadillacs—and eventually mainstream Chevys—take over more of the driving load on long highway stretches, with driveway‑to‑driveway capability on the roadmap after that.