Tesla FSD Safety Data, Disengagement Rates, and Regulatory Status: What the Numbers Actually Tell You

Tesla’s Full Self-Driving (Supervised) is now installed on more than three million vehicles across North America, generating headlines ranging from glowing owner testimonials to federal safety investigations. For a buyer trying to cut through the noise, the fundamental questions are concrete: what does the safety data actually show, is the “disengagement rate” you keep reading about a meaningful number, and what can regulators in your state or province legally require of you?

This article addresses those questions directly — with the honesty the data demands. Tesla publishes striking safety comparisons, but those figures have been criticized by independent researchers and federal regulators. FSD (Supervised) is a Level 2 driver-assistance feature, not an autonomous vehicle system. The driver must remain attentive and ready to take control at all times. The difference matters legally and practically, and no amount of impressive video clips changes it.

Disclosure: This post contains a referral link. If you order a Tesla using it, the author may receive referral benefits at no extra cost to you.

What Tesla’s Own Safety Statistics Say — and Why Critics Push Back

Tesla regularly publishes a vehicle safety report comparing crash rates for Autopilot-active miles, FSD-active miles, and miles driven without any driver assistance. For many recent reporting periods, Tesla’s published figures have shown FSD-active miles resulting in fewer accidents per million miles driven than the national average. Some Tesla-aligned commentary has framed this as FSD being significantly safer than a human driver — in some cases by a wide margin.

These are Tesla’s own published figures, drawn from Tesla’s internal telemetry. They have not been independently verified, and they have drawn pointed criticism from NHTSA, academic researchers, and automotive safety analysts for several reasons worth understanding before you treat them as settled science:

• Selection bias in the driving environment: FSD is used most heavily on highways and in favorable conditions — the same miles where any driver’s crash rate is lower. The comparison is not between equal driving populations under equal conditions.
• How “accident” is defined: Tesla’s metric counts incidents in which an airbag deployed or a property damage claim was filed. Driver interventions that prevented a crash — the moments FSD almost made a serious error — do not appear in the numerator, but they are arguably the most safety-relevant events.
• Denominator inflation: Every mile the system is active counts toward the denominator, including low-risk freeway miles where the base crash rate is already very low. A system used mainly on easy roads will look statistically impressive against an average that includes dangerous intersections, bad weather, and distracted driving.
• No independent audit: No third-party safety organization has validated Tesla’s methodology. NHTSA has not certified these figures as proof of safety superiority. They remain self-reported corporate data.

None of this means FSD is necessarily unsafe. But it does mean that claims proclaiming Tesla’s self-driving to be multiples safer than a human driver should be read with appropriate skepticism. The data may be directionally interesting; it is not a controlled study. Researchers have consistently called on Tesla to release raw incident data for independent analysis — something the company has not done.

To understand how FSD compares in terms of features and capabilities to Autopilot and Traffic-Aware Cruise Control — not safety specifically, but what each system actually does — see: Tesla FSD vs. Autopilot vs. TACC: What’s Actually Different in 2026.

Why “Disengagement Rate” Is a Deeply Misleading Metric

When people look for objective FSD performance data, they often land on “disengagement rates” — the number of times per mile a human driver had to take control from the autonomous system. California’s DMV requires companies testing autonomous vehicles on public roads to report disengagements annually. The comparison looks alluring: Waymo reports very low disengagements per mile; surely Tesla’s headline numbers must be comparable?

Here is the fundamental problem: Tesla does not report disengagement data to the California DMV, and it is not legally required to do so.

California’s AV testing reporting requirements apply to companies operating SAE Level 3 or higher vehicles in autonomous mode — vehicles where the system itself is the legal operator during automated driving. Tesla FSD (Supervised) is classified as a Level 2 driver-assistance system. The human is always the legal driver. Tesla is therefore categorically exempt from these reporting requirements, and its vehicles do not appear in the state’s annual autonomous vehicle disengagement report. For the most widely deployed driver-assistance AI system in America, there is no regulator-mandated, publicly filed disengagement figure at all.

What Independent Tracking Shows — and Its Limits

Because no regulatory-mandated FSD disengagement data exists, observers rely on community-run trackers — volunteers who log their own FSD drives and record every critical intervention (a moment where they had to take over to avoid a problem). These trackers produce a “miles per critical intervention” figure that is widely cited in FSD discussions.

Community tracking data for FSD v14.1, released in late 2025, reached a peak of roughly 4,000 city miles per critical disengagement — a figure Tesla and its proponents highlighted as evidence of dramatic improvement. After subsequent software updates, that figure dropped sharply, falling to below 1,000 city miles per critical intervention according to the same tracker community. By late 2026, figures from owner-run trackers hovered in the range of 1,000–2,000 miles per critical human takeover for the newest hardware, with significant variation across versions and driving environments.

Compare that to Waymo, whose most recent official California DMV disengagement filing showed roughly 63,000 miles per disengagement — a figure republished in 2026 industry statistics roundups. (Note that California is phasing out mandatory annual disengagement reporting, so the historical series is ending.) The difference sounds dramatic — but Waymo is a fully driverless Level 4 system, operated by a professional fleet on mapped geofenced routes, held to a fundamentally different standard. Equating community-tracked Tesla FSD interventions with officially reported Waymo disengagements is a category error. The two metrics are not measuring the same thing under the same rules.

What a Meaningful Safety Benchmark Would Actually Require

Tesla’s own head of FSD has stated publicly that the internal target for enabling unsupervised driving is reaching a miles-between-critical-intervention figure equivalent to the human average miles between crashes — a figure NHTSA places around 700,000 miles. By any community-tracker metric currently available, FSD (Supervised) is not close to that target. The gap is not a reason to dismiss the technology, but it is a reason to dismiss claims that FSD has already achieved human-level safety. It has not, by its own developers’ stated benchmark.

Until Tesla publishes audited, independently verified data under a consistent, publicly disclosed methodology — or until regulators mandate such reporting for Level 2 systems — the disengagement rate remains an incomplete and easily misread signal. If you want a balanced view of what FSD owners actually experience versus what the marketing suggests, see: Tesla FSD Myths Debunked: What Owners Actually Experience in 2026.

NHTSA’s Investigation Into Tesla FSD: Where Things Stand in 2026

The most consequential regulatory development for FSD buyers in 2026 is NHTSA’s active, escalated investigation into the system. Here is the timeline and what it means in plain terms for the roughly 3.2 million vehicle owners now covered.

NHTSA opened a preliminary investigation (PE25012) into Tesla’s FSD (Supervised) in October 2024, following reported incidents in which the system failed to respond appropriately in reduced-visibility conditions — specifically glare, dust, and fog. In March 2026, NHTSA escalated that investigation to an Engineering Analysis (EA26002), covering an estimated 3,203,754 vehicles across model years 2016–2026 for Model S and X, 2017–2026 for Model 3, and 2020–2026 for Model Y and Cybertruck variants equipped with FSD hardware.

What “Engineering Analysis” Means in Practice

An Engineering Analysis is the highest-intensity investigative phase NHTSA conducts short of formally ordering a recall. At the EA stage, NHTSA can compel Tesla to produce internal documents, engineering data, test results, and crash reports. NHTSA has reportedly demanded internal Tesla communications — including documents referencing radar-related system behavior — as part of this investigation. It is the final investigative step before the agency either closes the case (if it determines no safety defect exists or that prior corrective actions were sufficient) or formally proceeds toward a recall.

As of November 2026, EA26002 has not resulted in a recall, and NHTSA has not issued a formal defect determination. This matters for buyers: an open engineering analysis is a flag, not a verdict. However, it also means the matter is unresolved. Buyers of affected vehicle years should monitor NHTSA’s public docket at nhtsa.gov and register their vehicle with NHTSA’s recall notification system to receive updates if the status changes.

Red Lights, Fog, and Glare: The Core Allegations

Two categories of behavior are under scrutiny in the current investigation. The first is the low-visibility issue: nine incidents, including at least one fatal crash in the initial filing, in which the system allegedly failed to detect objects or hazards in conditions where camera visibility was impaired by glare, dust, or fog. Tesla’s vision-only architecture — which removed radar from most vehicles starting in 2021 in favor of a camera-only system — has been a persistent point of criticism from independent automotive safety engineers precisely because optical cameras can be physically blinded in ways that radar cannot.

The second area of concern, separately documented by NHTSA, involves more than 80 reported instances in which FSD allegedly violated basic traffic laws, including running red lights. Tesla has pushed over-the-air software updates targeting specific behaviors in this category, and NHTSA is continuing to gather data on whether those updates resolved the underlying failure modes or whether the problem persists across new edge cases.

Neither investigation has resulted in a finding that the system is categorically unsafe. But the combination — an escalated engineering analysis and documented traffic law violations — is precisely the kind of regulatory environment that warrants attention from anyone deciding whether to subscribe to FSD today.

How US States Regulate FSD: A Patchwork, Not a Standard

Tesla FSD (Supervised) is legal to use in all 50 US states. Because it is classified as a Level 2 driver-assistance system — not an autonomous vehicle — the driver is always the legal operator, and no special permit is required to use it on public roads. This is worth stating plainly because regulatory discussions about autonomous vehicles can create the misleading impression that FSD faces legal barriers in certain states. For supervised use, it does not.

The more complex picture involves state frameworks for fully autonomous vehicles (Level 3 and above) and how those frameworks interact with Tesla’s commercial ambitions for future unsupervised FSD. That distinction matters because what is legal today (supervised use everywhere) is different from what Tesla is trying to expand into (unsupervised commercial deployment, which requires specific state authorization).

What all 50 states do require: compliance with existing distracted-driving and hands-free laws. If your state prohibits handheld device use while driving, you cannot use your phone while FSD is active, regardless of what the car is doing. You are the driver. The system does not change that legal reality.

Texas: The Most Permissive Jurisdiction for Autonomous Operations

Texas has enacted some of the nation’s most permissive autonomous vehicle legislation. Senate Bill 2807, which took effect September 1, 2025, created a formal legal framework for commercial automated vehicle operations on public roads. Administrative rules were finalized in February 2026, with enforcement beginning in late spring 2026. This framework enabled Tesla’s limited unsupervised robotaxi operations — a small fleet operating in geofenced areas of Austin, Dallas, and Houston under commercial AV permits. For everyday FSD (Supervised) users in Texas, the state presents no additional regulatory barrier beyond standard traffic law.

California: Oversight and Permits Required for Unsupervised Operations

California maintains the most demanding regulatory environment for autonomous vehicles in the country. The California DMV and the California Public Utilities Commission both maintain oversight of autonomous vehicle testing and commercial deployment. Any vehicle operating without a human safety driver requires separate testing and commercial deployment permits. Tesla has obtained limited authorization for unsupervised operations in a small geofenced area in Palo Alto but has not secured the permits required for statewide commercial unsupervised FSD. For supervised FSD users in California, no additional permit is needed — but the state’s distracted-driving and hands-free laws are among the most strictly enforced in the country.

Other Notable State Developments in 2025–2026

Across the country, 2025 saw a surge in state-level autonomous vehicle legislation: 25 states introduced 67 related bills. Arizona, Louisiana, Montana, Nevada, and the District of Columbia enacted autonomous vehicle legislation that year. As of late 2026, 18 states have established frameworks that allow fully driverless commercial operations under permit. Active bills are under consideration in California, Illinois, Massachusetts, New Jersey, New York, and Pennsylvania. The absence of a comprehensive federal standard means regulatory requirements will continue to differ meaningfully across state lines — particularly relevant if Tesla’s commercial robotaxi service expands nationally.

Transport Canada and the Regulatory Landscape for Canadian Buyers

Canadian buyers operate in a parallel but distinct framework. Transport Canada does not consider fully self-driving vehicles to be commercially available in Canada at this time. Tesla’s own Canadian documentation consistently and explicitly describes FSD as requiring a fully attentive human driver — there is no ambiguity in how Tesla presents the system to Canadian customers. For regulatory purposes, FSD (Supervised) is a Level 2 driver-assistance system in Canada, identical to its US classification.

In March 2026, Transport Canada opened a formal public consultation on a proposed United Nations Global Technical Regulation (GTR) for automated driving systems. This signals movement toward a national regulatory framework for higher-level automation, but no binding federal rules specifically governing consumer automated driving features have been finalized. The consultation process itself reflects that Canada is at an earlier stage of autonomous vehicle regulation than some US states that have already enacted specific AV frameworks.

At the provincial level, the supervision requirement for FSD is universal: every province that has addressed driver-assistance and automated driving technology requires a licensed driver to remain attentive and in legal control when using Level 2 systems. This is not an FSD-specific rule — it applies to all driver-assistance technology and to all drivers across Canada. FSD (Supervised) in Canada is priced at $99 CAD per month, the same subscription-only model that replaced outright purchases in February 2026 across North America. For a full breakdown of provincial rules and practical guidance, see: Tesla FSD Supervised in Canada: 2026 Guide to Rules, Costs, and What to Expect.

The key difference between the Canadian and US regulatory pictures is not the supervision requirement — both require attentive human oversight — but the fact that no Canadian province has established a commercial pathway for unsupervised FSD deployment comparable to Texas’s framework. Unsupervised Tesla robotaxi service is not available in Canada, and no concrete timeline for Canadian commercial autonomous deployment has been announced.

The Gap Between FSD (Supervised) and True Autonomy

Much of the confusion surrounding FSD safety data stems from a naming problem that Tesla itself created. The phrase “Full Self-Driving” implies a capability the system does not currently have. The current official name — Full Self-Driving (Supervised) — is more accurate, but the original name persists in public understanding, advertising context, and casual conversation. That gap between name and reality shapes how buyers interpret both the safety statistics and the regulatory status.

FSD (Supervised) can handle a wide range of highway and urban driving tasks in favorable conditions: navigating intersections, making turns, responding to traffic signals, changing lanes, and managing speed in traffic. It is genuinely capable and continues to improve with over-the-air updates. It can also make unexpected errors — sudden inappropriate braking, missed stop signs, incorrect lane selections, red-light violations as documented by NHTSA — that require immediate driver intervention. This is the stated design: the driver is a supervisor, not a passenger.

Tesla’s unsupervised robotaxi operations in Texas and Palo Alto, California, represent a tightly controlled pilot operating in geofenced areas — a small fleet of vehicles, not a broadly available consumer product. Consumer-available unsupervised FSD has been pushed to Q4 2026 at the earliest by Tesla’s own public statements, and that timeline has shifted multiple times over prior years. The safety data, independent audits, and regulatory approvals needed to close the gap between supervised and unsupervised operation do not yet exist in a form that regulators or independent researchers have fully validated.

For a comparison of how FSD behaves across Tesla’s current lineup — what the hardware differences mean for the system’s performance — see: Tesla FSD on Model 3, Model Y, and Cybertruck: A Real-World Comparison.

What This Means for Buyers: A Neutral Summary

If you’re buying a Tesla in the US or Canada and deciding whether to subscribe to FSD (Supervised), here is a clear-eyed summary of where the safety and regulatory picture actually stands:

• It is legal to use everywhere you can legally drive. FSD (Supervised) is a Level 2 feature. Using it requires no special permit in any US state or Canadian province. Standard traffic laws and distracted-driving laws still apply fully.
• Tesla’s safety statistics are real data points but are not independently verified. They are self-reported, methodologically contested, and reflect Tesla’s own definitions of what counts as an incident. Treat them as directionally interesting, not as proven superiority over human driving.
• The NHTSA engineering analysis is worth monitoring, not panicking about. EA26002 covering 3.2 million vehicles is not a recall. It is the final investigative phase before NHTSA either closes the case or proceeds toward one. Check nhtsa.gov for updates on your VIN.
• Disengagement metrics you read online are not standardized or audited. Community trackers provide useful trends for comparing FSD versions over time, but the numbers are not apples-to-apples with Waymo’s official figures, vary significantly across software versions, and are not independently verified.
• You are always legally the driver. No US state or Canadian province currently permits unsupervised consumer FSD for private buyers. Regardless of what the car is doing, you are legally responsible for safe vehicle operation.

FSD (Supervised) currently costs $99/month USD in the United States and $99/month CAD in Canada. Tesla eliminated the outright purchase option in February 2026, making the subscription the only path to FSD for new buyers. If you’re evaluating the long-term financial picture of subscribing versus the former purchase price, see our break-even analysis: Tesla FSD Subscription Cost and Break-Even Analysis for 2026.

One practical note: new buyers who order using a referral link receive one free month of FSD (Supervised). One month is enough time to form a genuine firsthand opinion about how the system performs in your specific driving environment — something no amount of third-party data can replicate. You can claim the free month by ordering through this Tesla referral link (code: willy327274).

The safety and regulatory picture around FSD is genuinely mixed: a system that improves measurably with each software generation, publishing data that is promising but not independently validated, under active federal investigation for specific failure modes, and legal to use everywhere as long as the driver remains attentive. That nuance is worth holding as you evaluate the subscription — and worth keeping in mind every time you engage the feature on a real road. Order your Tesla and claim the free first month at tesla.com/get-app/referral?code=willy327274.