In the world of Tesla’s fast-paced software development, the release notes often lag behind the code itself, and Tesla often doesn’t actually reveal everything that’s under the hood in every update.
Noted Tesla hacker Greentheonly recently discovered a new, curiously named geofence region buried in the public builds of update 2025.44 called the “Bay Unsupervised CA DMV.” This discovery, accompanied by a map encompassing the entirety of the San Francisco Bay Area, is more than just an internal geographic boundary with a suspicious name - it is Tesla’s move forward for regulatory compliance.
For informed observers, linking Unsupervised with the CA DMV (California Department of Motor Vehicles) suggests that Tesla is using this geofence to gather fleet-wide data intended to help validate and satisfy the state’s requirements for driverless vehicle operation permits.
Shadow Mode
The inclusion of the DMV in the geofence title is the giveaway. It implies that this isn’t just an internal testing parameter for Tesla’s engineering team, but a specific data-logging filter tailored for regulatory reporting.
The primary hurdle for any AV company in California is the transition from a “Testing with a Driver” permit to a “Driverless Deployment” permit. To bridge this gap, an applicant must demonstrate their system is statistically safer than a human driver within a specific design domain.
This new geofence likely helps Tesla aggregate data: when a customer vehicle enters this zone, or when a Bay Area Robotaxi completes a trip, the software will tag those miles as simulated unsupervised, even when a human is in control.
That’s what Shadow Mode does - it runs behind FSD, making a comparison, and checking whether the trip would have been completed safely without a human supervising. That’s the dataset that Tesla needs: “We drove X million miles in the Bay Area. In 99.9% of those miles, our Unsupervised FSD logic made the correct decisions without human intervention.
Slowly at First…
Green’s analysis also revealed that Tesla didn’t expand this to a gigantic area all at once. The firmware has evidence of a more methodical, expanding testing strategy that likely has been running quietly for some time.
There are several smaller earlier zones, centered around Tesla facilities in Palo Alto - growing in size from just a parking lot, to the immediate block, then to the city of Palo Alto, and now the entire Bay Area.
Then Everywhere All At Once
The scope of the map matches Tesla’s Robotaxi Bay Area zone and is quite ambitious. While competitors like Waymo seek permits for restricted neighbourhoods or particular times of day, Tesla’s Unsupervised geofence appears to cover the complex, high-traffic arteries of the entire region, from San Francisco down to San Jose, and then up to the East Bay.
By hard-coding the geofence into customer vehicles, Tesla can collect data when a vehicle is within that geofence area and ensure that the data is relevant. They won’t be asking the CA DMV to look at highway miles in Montana or urban miles in New York - this is a set of data specifically for the Bay Area.
The Path to the Permit
It is important to mention that the existence of this geofence doesn’t mean Tesla will be flicking a switch for customers in that area, enabling Unsupervised FSD tomorrow. Rather, it can be seen as a sign that Tesla is confident in gathering the data to prove that Unsupervised FSD is at least as safe as a human driver.
Tesla will be gathering data, making its regulatory arguments, and then working to enable Unsupervised FSD as it progresses, step by step.
Hopefully, we’ll soon see the Bay Area use safety monitors like in Austin, or have no one in the car at all. In Austin, Tesla recently removed safety monitors on test vehicles, and they’re currently driving around without any passengers.
