Tesla Chose Self-Certification Over a 2,500-a-Year Exemption. NHTSA Is Now Auditing That Choice.

On September 3 — the day Tesla began charging for Cybercab rides in Austin — the National Highway Traffic Safety Administration's Office of Defects Investigation opened Audit Query AQ26002, logging it at 2:42 that afternoon, Austin time. The file examines, in the agency's words, "the process and technical data on which Tesla relied when certifying the Cybercab" — a car with no permanently attached steering wheel, brake pedal, accelerator pedal or mirrors — as compliant with every applicable Federal Motor Vehicle Safety Standard (FMVSS). The agency announced it the next day, and Tesla's pre-event rally reversed: $376.37 at Thursday's close, $354.08 at Friday's, down 5.92%.

That much is fact. Most of what travelled with it on X over the weekend is not, and the errors cluster around one number: the 1,000 printed on the audit file. Before Monday's open, it is worth being precise about what this action is, what that number is, and what Friday's move can and cannot be blamed on — because we went to the primary documents, and all three of the popular readings fail there.

But the 1,000 is the shallow end of this story. Six weeks before the launch, on Tesla's July 22 earnings call, an analyst asked the company's vehicle-engineering chief whether federal rules on steering wheels and pedals would gate the Cybercab ramp. "The short answer is no," Lars Moravy replied. "I feel like we have a partner there." The partner has now opened a file on exactly the question the analyst asked. So this piece is the long version: what the audit is and is not, the bet Tesla made when it certified a car with no controls under the current federal rulebook, what happened to the one company that made the same bet before it, and what each way this can end is worth on our Tesla model. The short version of that last part: delay is survivable, and the fallback — the capped exemption path Tesla has publicly said it will not need — prices within $1.11 a share of deleting the robotaxi business entirely.

What the record actually says

• It is an audit of paperwork, not a defect investigation. The file's own action line reads "Open Audit Query (AQ)". There is no crash, no defect allegation, no finding of non-compliance and no order to stop or pause the service. The agency's open FSD investigations are a different track entirely: those ask how the software behaves on the road, this one asks how the vehicle was certified.
• The scope is one question. NHTSA "will consider the extent to which Tesla's certification depended on determinations that certain FMVSS are inapplicable to the Cybercab." Tesla self-certified; the audit tests the theory behind the self-certification.
• "Population: 1,000 (Estimated)" is a standard field on the file — the agency's estimate of the product class, not a count of anything on a road. In the same document, NHTSA describes what deployed on September 3 as "a small number" of vehicles.
• The Texas registry holds 45 Cybercabs. We pulled the state's public autonomous-vehicle registry on Saturday, September 5: 432 Tesla vehicles registered as autonomous in Texas, of which 45 are Cybercabs and 387 are Model Ys, all model-year 2026.
• The public docket holds exactly one document — the opening file itself. Tesla's certification package is not public, no information-request letter has appeared, and no response deadline has been published.
• The agency's framing is deliberate. Administrator Jonathan Morrison, in the announcement: "NHTSA fully supports the safe development and deployment of automated vehicles. But as the federal regulator, we need to ensure that all of our laws are followed." Existing standards remain in force while the agency's rule updates for vehicles without manual controls are unfinished.
• Tesla has said nothing of record. No statement, no comment to the wires, and no current-report filing with the SEC since its July 22 quarterly report. Operationally it is proceeding at full speed: the rider guide is dated September 3 and public app rides were being advertised in Austin on Friday afternoon.

The three misreadings, in ascending order of cost

First: "NHTSA is auditing 1,000 cars on Austin streets." The 1,000 is the file's estimated product population. The cars that exist in any registry are the 45 above; the cars that carried riders on day one are NHTSA's "small number". The most likely reading of the 1,000 is the class of vehicles Tesla can field, not the vehicles fielded — and it happens to sit almost exactly on our Cybercab production ramp, whose base case has roughly 900 Cybercabs built by the end of September, most of them parked. An estimate of what has been built is not a count of what is deployed, and the gap between those two numbers has been the finding of every piece we have written on this product.

Second: "1,000 is Tesla's certified ceiling — like Nevada's 5,000." No. Nevada's 5,000 was a permit ceiling issued by a state transportation authority; a ceiling is the instrument there. Federal self-certification carries no unit cap at all — that is precisely why manufacturers prefer it to the exemption route, which does. The 1,000 is an estimate on an audit file, written by the auditor. Reading it as a cap imports the shape of the Nevada story into a document that has no cap in it.

Third: "Tesla fell 6% because of the audit." The print is real — down $22.29, and Friday closed about 0.8% below even the level the stock traded at before Thursday's 5.4% pre-event rally. But the after-close wraps led with the launch itself: invite-only, not livestreamed, no Elon Musk, and little incremental information on deployment numbers or locations. Wells Fargo's note was titled "Launch Event Underwhelms", per press reports. The honest attribution is a disappointing launch with a federal audit layered on top, and no available method separates the two inside one session.

What the audit actually turns on

Tesla told the agency the Cybercab complies with all applicable standards. The load-bearing word is "applicable". Several federal standards are written on the assumption that a steering wheel, pedals and mirrors exist — they specify where controls sit, how they behave, what the driver must be able to see. Tesla's certification, per the file, rests on determinations that some of those standards simply do not apply to a vehicle that has none of the hardware they regulate. The audit examines whether that reasoning and its technical data hold.

It also lands on the joint the Texas service stands on. Texas authorises autonomous vehicles on the strength of the manufacturer's federal self-certification — no hearing forces Tesla to state anything — so the paperwork NHTSA is now auditing is the same paperwork the Austin deployment rests on.

Note what the agency's description does to one popular caveat: the file states flatly that the deployed vehicles lack a brake pedal, gas pedal, steering wheel and mirrors. Claims that some Cybercabs may carry brake pedals appear nowhere in the federal record for the launched cars. The innocent origin of pedal sightings is in Tesla's own words: Elon Musk said on the July call that the company accumulates calibration miles with Cybercabs "retrofitted with steering wheels and acceleration, braking pedals, that kind of thing, to calibrate to the Cybercab chassis." Retrofitted test cars are not a production configuration, and we are not going to assert one that the auditing agency's own description contradicts.

The bet, in Tesla's own words

Federal law offers the maker of a car with no manual controls two doors. Door one: certify, yourself, that the vehicle complies with every applicable standard — no application, no waiting, and no limit on how many you build. Door two: petition for a temporary exemption from the standards the design cannot meet — and the ceiling on that door is not agency policy but statute. Title 49, Section 30113(d) allows the relevant exemptions "only if the Secretary determines the exemption is for not more than 2,500 vehicles to be sold in the United States in any 12-month period," granted "for not more than 2 years" at a time, renewable on re-application. Congress wrote the cap; NHTSA cannot waive it from inside that path.

Tesla took door one, and said so plainly. Asked on X in April whether the 2,500-a-year cap applies to the Cybercab, Moravy answered "No", as reported by Reuters and Electrek — no cap, because no exemption: the company's position is that the Cybercab complies with the standards as written, the missing-controls rules simply not applying to a car with nothing for them to regulate. The audit examines precisely the determinations that position rests on. Lose them, and the only door left is the one with the statutory cap — the cap Moravy said the Cybercab "will not be subject to".

Which gives the July 22 exchange its before-and-after quality. Walter Piecyk of LightShed asked:

"The NHTSA Administrator, Jonathan Morrison, was just on CNBC. He already is trying to get rid of the pedal and said, I forget his exact term, but it seemed like he wanted to get rid of the steering wheel. It seemed like that would be a gating factor, is there other things that you need at the federal level to really unlock your ability to ramp Cybercab?"

Moravy's answer:

"The short answer is no. I think we have a great relationship with NHTSA and Administrator Morrison especially. … We've been open and honest with them for the last couple of years about what our plans were and what we're doing, and I don't want to say we're in lockstep, but I feel like we have a partner there, and we're working together on it."

Note what the question was about. Not the 45 cars, not the launch date — the ramp, the rate at which Cybercabs can enter service. Six weeks and a day later, Administrator Morrison's agency opened a file on the certification that rate stands on, and Morrison's framing — "we need to ensure that all of our laws are followed" — is not the language of lockstep. Our analysis of that call wrote in July that the Cybercab launch would invite "exactly this kind of scrutiny at a larger scale" of the gap between how Tesla markets autonomy and what it can demonstrate. We did not expect the invitation to be accepted within six weeks.

The only company that has walked this road

There is exactly one precedent, and NHTSA's own closing file on it contains a sentence that now describes Tesla word for word: "Zoox was an outlier in the industry, having self-certified these vehicles rather than seeking an available exemption."

Zoox — Amazon's robotaxi unit — announced in July 2022 that it had self-certified its no-controls vehicle. NHTSA sent a special order that September demanding the basis of the certification, opened its own audit query in March 2023 — eight months after the announcement — and inspected two of the vehicles. Its December 2024 inspection report found "multiple apparent noncompliances with applicable FMVSS." Then, "following dialogue with the agency," as the file puts it, Zoox stopped defending the bet: it applied in June 2025 for a demonstration exemption, which the agency granted that August on condition that Zoox "obscure or remove all existing statements that the purpose-built vehicle complies with or conforms to all applicable FMVSS." Some of the vehicles were "discontinued from public road operation." The audit closed the same day. A separate commercial exemption — the one that finally allowed paid rides — came in July 2026, on press-reported terms of 2,500 vehicles a year for two years, covering parts of eight standards, with speed and weather restrictions. Paid Las Vegas service began weeks later.

Read as a map: the audit ran twenty-nine months, and the full path from self-certification to sanctioned paid service ran about four years. At no point did the agency ground the fleet — even after documenting apparent noncompliances — and at no point did the self-certification survive. It was not merely set aside; the company was made to erase the claim.

Three details sharpen the precedent. First, the personnel: Neil Dold, who investigated the Zoox audit, is the reviewer on Tesla's file; Otto Matheke, who reviewed the Zoox closure, approved opening Tesla's; and the sentence describing each investigation's scope is word-for-word identical between the two documents. The same office is running the same playbook, thirteen months apart. Second, the scale is not the same: Zoox's audited population was 64 vehicles; Tesla's file estimates 1,000 — roughly fifteen times larger, opened the day commercial deployment began rather than eight months after an announcement. Third, the money: Zoox could not charge a fare until its exemptions came through, so no precedent says whether paid service may continue through an open audit. Tesla is the first company collecting fares while its certification is examined. Whether that continues is genuinely open — the audit orders nothing stopped, and nothing published answers it in either direction.

The cap also has a Tesla-specific echo: 2,500 a year is the number Tesla's own Cybercab chief engineer told Nevada regulators would be a satisfying twelve-month outcome for Clark County alone.

Delay is survivable on our model. The cap is not.

An audit has three endings worth pricing. NHTSA accepts Tesla's determinations, and the bet pays off in full. Or the process drags and the ramp starts late — the middle of the Zoox map, whose audit slowed its commercialisation per press accounts. Or the determinations fail, and the Cybercab lands on the exemption path with the statutory cap Tesla said it would never need.

Our model carries robotaxi as a capacity line: 1,000 vehicles in commercial service, additions of 600 a quarter beginning in the December quarter, compounding at 22% a quarter, at $30,000 of revenue per deployed vehicle per quarter. Every one of those is our assumption, none is a Tesla disclosure — and the model's starting 1,000 is an August assumption that coincides with the audit file's estimate by accident, not by shared source. Against the $272.39 base case:

Delay barely registers.

A nine-month regulatory pause costs $3.79 a share; even a two-year freeze costs $6.44 — both less than the $8.47 the entire line is worth, because the compounding survives on the far side of the pause. Our ramp scenarios already carried this case before the audit existed: their low case was defined as "a software or regulatory pause of a month or more".

The cap is different in kind.

A permanent 2,500-a-year regime — 625 additions a quarter with the compounding dead — prices at $265.03, down $7.36 a share, or 87% of what deleting the entire robotaxi business costs. The last $1.11 is the frozen trickle of 625 cars a quarter. And the bite is entirely the compounding, not the number: 625 a quarter with the 22% growth intact is worth +$0.35 against base, and 600 versus 625 with growth dead moves four cents. On our model, the capped path does not trim the robotaxi case — it is the delete case, minus the trickle.

This is the same asymmetry Nevada taught: the full 5,000-vehicle permit was worth $0.33 a share, the additions growth rate $11.06. Levels are worth cents and rates are worth dollars. Tesla has disclosed levels and has never disclosed the rate — and on our model the rate is the entire business. The audit is the federal government examining the legal basis of the rate.

Two honesty notes on those tables. The asymmetry is partly a fact about our model's structure: a growth rate compounds across twenty-odd quarters while a level does not, so any model built this way will price rates far above levels — the ordering is robust, but the exact dollar figures are ours. And the flat-cap row assumes the cap never lifts across the model's horizon, when the statute's exemptions run two years at a time and are renewable, and Zoox's own story kept moving after its first grant — read −$7.36 as the bound of the cap world, not its expected value.

The model itself does not change on this news. An audit query moves no operating assumption; the instrument that would — a numeric cap, Zoox-style — is exactly what the watch list below is for.

Friday erased 2.6 times what the robotaxi line is worth on our model

Here is the arithmetic that makes the causation question worth caring about. At 3,528 million diluted shares from the last annual report, Friday's $22.29 decline removed about $78.6 billion of market value in one session. The entire robotaxi line — every market, every future vehicle, compounding for years — is $8.47 a share on the model above, or about $29.9 billion on the same share count. Friday's move was 2.6 times the modelled value of the whole business the audit touches — a business whose registered fleet is 45 cars, whose revenue our fleet-size arithmetic says would need roughly 14,800 cars to move by a single percent, and whose parent earned a 1.4% operating margin on cars in the June quarter. Every figure in this paragraph after the share count is ours, not Tesla's — and that is the point. There is no set of disclosed numbers on which a compliance audit of this product justifies a $79 billion repricing. What moved is the autonomy expectation embedded in the multiple, which was never a number Tesla disclosed either.

The registry step change happened — and it is still Model Y

Our piece ahead of the launch said the earliest observable signal of a real ramp would be a step change in the Texas registry, which had been adding 1.65 vehicles a day. The step change happened: 186 registered vehicles on August 21 to 432 on September 5 — 246 added in 15 days, about 16 a day, nearly ten times the prior measured rate. Two caveats keep it honest: a launch-fortnight registration batch is not a run rate, and 201 of the 246 additions are Model Ys. The Cybercab — the product the audit covers — is still 10% of the registered fleet.

What to watch

1. The docket's second document. The file holds exactly one today — the opening record. The next to land sets the clock: an information-request letter with a response deadline, Tesla's reply showing which standards it determined inapplicable, or a closure. The entire case turns on whether NHTSA accepts those determinations.
2. Whether paid rides continue through the audit. Open in both directions today. The agency saying nothing while the service expands is itself an answer; so is any request to pause charging.
3. A Part 555 petition or an SEC filing. Nothing has been filed since July 22. A petition would mean accepting the capped path Tesla has publicly said the Cybercab will not take — and would make the Zoox timeline, and the cap table above, the base case.
4. The Texas registry moving off 45. Cybercab registrations are the first observable deployment signal, and whether they stall while Model Y registrations keep climbing is the cleanest read on whether the audit is slowing the actual product.

The audit's dates, scope, population estimate, quoted language and the missing-controls description are from NHTSA's published file for AQ26002; the Zoox account, including every quoted phrase, is from the agency's published closing file on that earlier investigation; Administrator Morrison's statement is from the agency's September 4 announcement via press reports. The 2,500-vehicle and two-year exemption limits are federal statute, 49 U.S.C. §30113(d). The Piecyk, Moravy and Musk quotations are as delivered on Tesla's July 22, 2026 earnings call. Texas counts are our September 5 pull of the state's public registry; the August 21 figure is from Robotaxi Tracker, which aggregates the same registry; registration is not service. The closes are the September 3 and 4 prints and will not match today's quote; every market-cap, ratio and registry-rate figure is our arithmetic on them, with the diluted share count from Tesla's FY2025 annual report. The base case and every fair-value sensitivity are our model's assumptions, not Tesla figures. Zoox's 2026 commercial exemption terms, analyst-note characterisations and the Moravy cap remarks are press-reported; Tesla has said nothing of record.