Tesla's Best Legal Defense Is That You Shouldn't Have Believed the Name

In October 2016, Tesla promised that one of its cars would drive itself from Los Angeles to New York with no human touching the wheel - a coast-to-coast demonstration, by the end of 2017.⁷ The drive never happened. Nearly a decade later, the car you can buy today still asks you to keep your hands ready and your eyes on the road, because in Tesla's own regulatory filings it is an SAE Level 2 system that requires 'a fully attentive driver who is engaged in the driving task at all times.'³ The product is called Full Self-Driving. It does not, by the company's own definition, fully drive itself.

The official story is that Full Self-Driving is the world's most advanced autonomy program, perpetually one large software update away from arrival. The real story is plainer and stranger: it is a decade-long sequence of specific, dateable promises, each one missed, sold under a name that asserts the very capability the system does not have. And the most revealing evidence of that gap is not a missed deadline. It's the argument Tesla's own lawyers won.

The defense that admits the case

In September 2024, a court handed Tesla a victory. Investors had sued over Musk's autonomy timelines, alleging securities fraud. The court dismissed the suit, ruling that Musk's statements were 'corporate puffery' - vague corporate optimism, too forward-looking to count as an actionable misrepresentation.⁴ Tesla treated it as a win. It was. But look at what the win requires you to accept. To escape a fraud claim, the company needed a court to find that no reasonable investor should have taken the CEO's autonomy claims at face value. That is not a ruling that the promises were true. It is a ruling that they should not be believed.

To win the fraud case, Tesla needed a court to agree that its CEO's promises were not the kind of thing a reasonable person should rely on. That is the whole product in one sentence.

Here is the bind that makes this an A24 crisis rather than a marketing footnote. The same statements that are harmless 'puffery' to a sophisticated investor are something else entirely to a consumer who paid five figures for a feature named after the promise. 'Don't take it literally' is a fine defense against shareholders who read 10-Ks. It is a terrible one against a buyer who read the order page - and an even worse one against a jury looking at a fatal crash. Tesla escaped one courtroom by conceding the exact thing that sinks it in the others.

Why a name becomes a liability

The mechanism here is not a bad sensor or a buggy update. It is the wedge between what the system is engineered to do and what its name tells a human to expect. A Level 2 system is only safe if the driver stays fully engaged, because the system will, predictably, fail to handle some situations and hand control back. NHTSA's engineering analysis of Autopilot worked exactly that fault line: across 956 crashes where Autopilot was alleged to be in use, it found that 'foreseeable driver misuse of the system played an apparent role,' and that the driver-engagement warnings 'did not adequately ensure that drivers maintained their attention on the driving task.'² Read that twice. The regulator did not primarily fault the software's perception. It faulted the design's failure to keep the human paying attention - in a product whose name tells the human they don't have to.

That is the liability machine. Name a Level 2 product 'Full Self-Driving,' and you have manufactured the precise gap between expectation and reality that produces inattentive drivers - and then a court gets to decide whether that gap was foreseeable. Tesla's own recall history shows it understands the exposure. The December 2023 recall covered some 2 million vehicles with Autosteer for driver-monitoring deficiencies, and Tesla issued it while flatly stating it 'did not concur with the agency's analysis.'¹ It fixed the thing it said wasn't broken. That is not the behavior of a company confident in its position; it is the behavior of a company managing a regulator it cannot beat.

The bills come due in three courtrooms

For years the gap between promise and product cost Tesla nothing but credibility, which it had to spare. That changed when the gap stopped being a marketing question and became a damages question. In August 2025, a Miami federal jury in Benavides v. Tesla found the company 33% liable for a fatal 2019 Autopilot crash that killed a 22-year-old woman, and awarded $243 million - including $200 million in punitive damages.⁵ Tesla had been offered a chance to settle for $60 million before trial and turned it down.⁵ Punitive damages of that size are not compensation. They are a jury's verdict on conduct - the legal system's way of saying the gap was not an accident but a choice.

Then the consumer side opened. In August 2025 a federal judge certified a class of California buyers who paid for the FSD package between October 2016 and July 2024 and opted out of arbitration - and the court accepted a full-refund damages theory, meaning members could recover the entire $5,000 to $15,000-plus they each paid if misrepresentation is proven.⁶ And in October 2025, the regulator returned: NHTSA opened a preliminary evaluation covering roughly 2.88 million vehicles after FSD-engaged cars were reported running red lights, with six such reports and four resulting injuries.³ Three doors - a wrongful-death jury, a refund class, a federal investigation - all hinged on the same hinge: the distance between 'Full Self-Driving' and Level 2.

Isn't this just hard technology arriving slowly?

The fair objection is that autonomy is genuinely one of the hardest engineering problems of the era, and that ambitious timelines are how hard things get funded and built. Every frontier company over-promises; that's not fraud, it's how frontiers get crossed. There's truth in it. Tesla has shipped a Level 2 system that millions use, and the data flywheel is real - Musk has said the fleet needs 10 billion miles before safe unsupervised deployment, and as of the Q1 2026 earnings call it was approaching that figure with unsupervised consumer FSD pushed to Q4 2026 at the earliest.⁸ A company that meant to defraud doesn't keep grinding toward the milestone for a decade.

But the objection answers a question nobody is asking. The lawsuits are not about whether autonomy is hard. They are about the name. You can pursue a hard problem honestly without selling the unfinished product as if it were finished - the rest of the industry calls its systems 'driver assistance' and tells you to keep your hands on the wheel. Tesla chose 'Full Self-Driving,' collected up to $15,000 a car for it, and is now defending the choice by arguing nobody should have believed the words. The slow arrival isn't the problem. The premature christening is.

Tesla has spent a decade insisting Full Self-Driving was nearly here, and now insists - in the one forum where it counts - that no reasonable person should have taken the insistence seriously. Both can't be the load-bearing truth, and the company gets to pick a different one in every room. That works until the rooms start comparing notes, and in 2025 they did: a jury, a class, and a regulator all reading the same gap the same way. The most expensive thing Tesla ever shipped wasn't the software. It was a name that promised what the software couldn't, and a defense that proved it.