Who pays after an Uber or Lyft crash?

It depends on what the app was doing. Offline, only the driver's own policy applies. Waiting for a request, a smaller contingent policy applies. From accepting a ride until the passenger is out, the platforms publish at least $1 million of cover. The screenshot showing which period applied is what an Uber accident lawyer needs first.

The three coverage periods the platforms publish

Your own policy, and nothing else

Driving for nobody is ordinary driving, whatever the sticker on the windshield says.

Uber published driver insurance, coverage when offline

$50,000 per person, $100,000 per crash, $25,000 property

Third-party liability for the injuries and damage the driver causes while waiting.

Uber published driver insurance, online and available

At least $1,000,000

From accepting the request until the rider is out, covering riders and everyone else hit.

Uber published driver insurance, en route or on a trip

The same three tiers

Lyft publishes the identical structure, with named markets where the limits differ.

Lyft published driver insurance

Limits vary by state

Uninsured and underinsured cover in particular, so the state page decides the real number.

Uber published driver insurance, state variation notice

Rideshare platform coverage periods, checked 2026-09-11 by second-pass check agent, independent of the author.

Who answers, and why the driver is only part of the answer

The driver is liable the way any driver is. What differs is that behind them sits a platform whose published cover changes with the app's state, and the platform decides what its own record says about that state. So the question "who pays" is answered by data neither party at the scene can see.

A passenger, another driver, a cyclist and a pedestrian are in different positions here even in the same crash. Someone in the car was on a trip, which is the period with the largest published cover. Someone hit by a car that was waiting for a request is looking at a much smaller one.

None of this is federal law, and the pages on this site say so. These figures are what the platforms publish they carry and what a state's own transportation network company statute requires them to carry, which is why the number moves when you cross a state line.

The record the platform holds, and the one on your phone

The single most useful thing you can do is photograph the driver's screen, or your own trip receipt, before anyone leaves. It establishes which period applied, and the periods are the whole dispute. A receipt emailed minutes later does the same job.

The platform holds the authoritative version: when the driver went online, when a request was accepted, the route, and the time of drop-off. That record is obtainable, and it is what an insurer's early position gets measured against.

The ordinary evidence still matters. The police report, the other driver's details and photographs of the vehicles are what they always are, and the rideshare layer sits on top of them rather than replacing them.

Why an Uber accident lawyer starts with which policy was live

Offline, the sticker in the window means nothing. Driving for nobody is ordinary driving, and the driver's personal policy is the only one in play, subject to whatever exclusion that policy carries for commercial use.

Online and waiting, the published third-party cover is $50,000 per person, $100,000 per crash and $25,000 of property damage. It is contingent, meaning it sits behind the driver's own policy rather than in front of it.

From the moment a ride is accepted until the passenger is out, the platforms publish at least $1 million, covering the passenger and everyone else involved. Uninsured and underinsured cover inside that period is the part that varies most by state, which is why the state page decides the real number.

Photograph the trip before you leave

The app is the only witness that cannot be talked out of what it saw, and it is also the one that disappears on its own. A completed trip stays in a rider’s history for a long time. The screen showing a driver online and waiting does not, and a driver whose account is deactivated after a crash can lose access to their own record of the night.

So before anyone drives away: open the app and photograph the trip. The driver’s name, the vehicle and the plate, the pickup, the drop-off, the timestamps. If you were the passenger, save the receipt as well. If you were in the other car or on foot, the rider standing beside you has all of it and will usually send it to you; failing that, the plate and the platform’s sticker in the windshield are enough to start from.

None of this replaces the platform’s own record, which is obtainable later and is what actually settles which of the three periods was running. What it replaces is the weeks in between, when nobody can yet tell you whether the app was even on.

When the platform’s timeline is the only witness left

The driver’s app writes a record whether anyone looks at it or not: the moment they went online, every request offered, which ones they accepted, the route the phone reported, and the moment a trip was closed out. That telemetry sits on the platform’s servers under the platform’s own retention schedule, and no regulator sets how long it lives there.

A letter asking the platform to preserve the trip record is therefore an early step rather than a late one. Asked once positions have hardened, it becomes a request about a dispute. Asked in the first weeks, it is a request about a file nobody is arguing over yet.

A driver who has been deactivated after a crash has a harder version of the same problem, because deactivation removes the in-app access they would otherwise use to pull their own history, and it is a decision the platform makes quickly and alone. Their agreement with the platform usually routes disagreements into arbitration rather than a courtroom, which changes the route to the record but does not close it.

A delivery is not a ride, and dual-apping is neither

The same three tiers are published for couriers carrying food and parcels, but the shape of the work decides which tier was running in a way that rides do not. A courier is often carrying several orders batched together, and the stretch between handing over one and collecting the next can read as waiting rather than as engaged. Nobody standing at the roadside can tell which it was, and the gap between the two is most of the money.

Then there is dual-apping, meaning a driver logged into two platforms at once, taking whichever offers first. It is ordinary practice among people doing this for a living. What it produces after a crash is two published policies each with a reason to say the trip belonged to the other one’s waiting period. Which app held an accepted request at the moment of impact is the whole of that argument, and it is answered from two timelines rather than from anything visible at the scene.

What the platform checked before it let the driver drive

Neither platform employs a driver in the ordinary sense, and neither trains one. What they do run is a screening process, and what it does and does not look at is the ground for a claim aimed at the platform rather than at the person behind the wheel. A screening vendor pulls a motor vehicle record and a criminal history search, applies a list of disqualifying entries within a stated lookback period, and re-runs monitoring afterward so that a new conviction or a license suspension surfaces without waiting for an annual cycle.

Around that sit the vehicle rules. Both platforms set a model year floor, require an inspection at a stated interval, and require the driver to carry their own policy, which is ordinarily a personal auto policy with a commercial-use exclusion in it. No commercial license is required and none of this is a commercial fleet program. A driver can be approved to carry strangers for money on an afternoon.

Where that becomes a claim rather than a grievance is when a specific entry was reachable by the process the platform says it runs and was not acted on. That is a narrow argument and it depends entirely on what the screening actually covered, which is why the published criteria and the date the driver was approved are worth establishing early rather than assuming either way.

Airport queues, pooled rides, and the small things that fix the period

A handful of ordinary situations decide which period was running, and none of them is obvious from the road.

Airports are the clearest. Drivers wait in a designated staging lot and join a virtual queue held by the platform, and the boundary of that lot is a geofence the app is watching. A driver leaving the lot, circling the terminal loop and returning is doing something the queue records precisely, and a crash on an airport access road is usually datable to the second against it.

Pooled and shared rides complicate the passenger side rather than the driver side. Two riders in the same car can be at different points of their own trips, one collected and one not yet, and a third request can be accepted mid-journey. The driver is plainly engaged throughout, but which trip a given passenger’s claim attaches to is not a single answer.

Then there is simply identifying the car. Platforms have moved away from large decals toward small windshield trade dress that is easy to miss and easy to remove, and a driver who has stopped, taken the placard down and driven off has left a vehicle that looks entirely private. The plate, and a rider standing on the sidewalk with the trip open on their phone, are what recover it.

The driver’s own policy, and the exclusion most of them have not read

Almost every personal auto policy excludes carrying passengers for a fee. That exclusion is what the platform’s contingent cover is built around, and it is also the reason a driver’s own insurer can decline a claim outright, and in some cases decline to renew the policy afterward, once it learns what the car was being used for.

Personal insurers now sell an endorsement to close that gap, covering the waiting period the platform covers only contingently and, on some products, the period before a request is accepted at all. Whether the driver bought one is knowable from their declarations page, and it materially changes what is available where the platform’s waiting-period limits are too small for the injuries. It matters to a passenger and to the other driver as much as it does to the driver, because it is another policy in the file rather than another argument.

The mirror of this is worth saying plainly to anyone hurt: the driver being an independent contractor rather than an employee is a fact about their tax status, not a shield around the platform. What the platform’s own published cover reaches, it reaches regardless of how the driver is classified.

The app may have noticed the crash before anybody reported it

Both platforms run sensor-based detection on the driver’s phone. An unusual deceleration, a long unexplained stop mid-trip or a route that halts where no destination is can trigger an automated check: a prompt to the driver, a prompt to the rider, an offer to call emergency services, an in-app report form. Whether anybody answered those prompts, and what they tapped, is logged.

That matters for a reason people miss. A driver who told an adjuster weeks later that nothing much happened may have dismissed a prompt at the roadside that records the phone measuring otherwise. And a rider who used the in-app emergency button, or shared their trip with somebody who then watched it stop, has created a timestamped account of the crash before anyone had a position to protect.

When there is nobody in the front seat

Driverless vehicles now take passengers on rideshare networks in a growing list of cities, and they change the shape of a claim completely. There is no driver whose attention wandered, no personal policy underneath, and nothing to photograph on a dashboard. What there is instead is a machine that recorded everything: lidar returns, camera feeds, radar, the map it was matched against, the trajectory it planned and the one it executed, and every disengagement or remote assistance request logged to the millisecond.

That record is vastly richer than anything a human-driven crash produces, and it belongs entirely to the operator. Nobody at the roadside can capture any of it, which inverts the usual advice: the photograph that matters is of the vehicle, its identifying number and the street, and the substantive evidence is reached by asking the company that owns the fleet rather than by anything done in the first hour.

The defendants change too. Without a driver, the claim runs against the operator of the fleet, the developer of the driving software, and potentially the manufacturer of the vehicle or a sensor in it, which is a products case with experts and a timeline unlike anything in an ordinary collision. Several jurisdictions now require these operators to file collision reports with a state regulator, so a public record of the incident may exist that has no equivalent for a human driver.

What gets counted, and what nobody counts

No federal file records a rideshare crash as a rideshare crash. The national fatal-crash record counts every death on a public road and codes the vehicle, not the app that dispatched it, so a trip that ended in a death is inside that total and cannot be pulled back out of it.

What does exist is held by the platform and, in states whose transportation network company statute requires it, reported to the state regulator: when the driver went online, when the request was accepted, and when the passenger got out. That record decides which policy applies, and it is obtainable. Our fatal crash tool carries the public totals for the state and county it happened in.

Rideshare rules, state by state

What each platform publishes is national. What a state requires of it, and what deadline you have to act inside, is not.

Where to go next, including finding an Uber accident lawyer

Take this page with you

Download this page as a one-page kit (PDF)

Sources

Want to discuss your accident?

Answer a few questions to request an attorney review. You choose whether to hire anyone.

Check my case

Free to request. No obligation to hire anyone.