The Trust Gap: Inside the Internal Skepticism Surrounding Tesla’s Full Self-Driving Ambitions

The discrepancy between the public marketing of autonomous technology and the private reality of its development has reached a critical inflection point. For years, Tesla CEO Elon Musk has championed "Full Self-Driving" (FSD) as a revolutionary leap toward a world without human drivers. However, a recent and extensive investigation by Reuters has pulled back the curtain on the internal culture of the company, revealing a profound lack of confidence among those closest to the technology.

The investigation, which featured interviews with nine former Tesla data labelers and a former self-driving engineer, suggests that the individuals responsible for training the artificial intelligence do not trust the product they are building. This revelation poses a fundamental question for consumers and regulators alike: If the architects of the system are afraid to use it, is the public being used as a massive, unpaid testing cohort for a fundamentally flawed product?

Main Facts: The Insider Verdict on FSD

The most striking takeaway from the Reuters report is a simple, damning statistic: seven out of nine former Tesla data labelers interviewed stated they would not ride in a Tesla operating in FSD mode. These individuals are not casual observers; they are the specialists whose entire professional lives revolved around watching thousands of hours of FSD footage, identifying errors, and teaching the software how to navigate the complexities of the physical world.

One former employee went as far as to state they would not ride in a Tesla robotaxi "if you f**king paid me." This sentiment was echoed by a former self-driving engineer who cautioned against believing the CEO’s optimistic projections. "Definitely don’t trust Elon on this," the engineer noted, specifically referencing Musk’s repeated claims that Tesla’s vehicles are on the verge of achieving "safe unsupervised" autonomy.

The data labelers’ primary concern stems from what they saw on their screens daily. Tasked with combing through terabytes of proprietary driving data, these workers routinely witnessed the software failing in real-world scenarios. At least five of the interviewed specialists reported that they frequently saw Teslas operating on FSD exceed speed limits—a behavior that was reportedly treated as a low priority by management. While engineers focused on "edge cases"—rare events like unusual road configurations or extreme weather—routine safety violations like speeding were deprioritized, leading to a system that essentially "learned" to break traffic laws.

Chronology: A History of Missed Deadlines and Rebranding

To understand the weight of these internal warnings, one must look at the timeline of Tesla’s autonomous driving promises. The gap between expectation and reality has been widening for nearly a decade.

• 2016: Elon Musk announced that all Tesla vehicles being produced would have the hardware necessary for full autonomy. He famously predicted a Tesla would be able to drive itself from Los Angeles to New York without a single human touch by the end of 2017.
• 2019: At "Autonomy Day," Musk promised that by 2020, Tesla would have over a million "Robotaxis" on the road. He suggested that owners could turn their cars into income-generating assets while they slept.
• 2021–2023: As the technology struggled with basic navigation and safety, Tesla shifted its terminology. The "FSD Beta" program expanded to hundreds of thousands of users, but the system remained firmly classified as Level 2 autonomy, meaning the driver must remain fully attentive at all times.
• 2024: Tesla rebranded the software to "FSD (Supervised)," a move seen by critics as a legal hedge against mounting lawsuits. Despite this, Musk continued to push for a "Robotaxi" unveiling and announced plans to bring the software to the Chinese market.

Throughout this timeline, the software has been implicated in a series of high-profile incidents. While Tesla maintains that its cars are safer than human-driven vehicles, real-world footage and data labeler testimony tell a story of cars driving into lakes, striking emergency vehicles, and failing to recognize oncoming trains. Each year, a new "v12" or "v13" update is touted as the "mind-blowing" solution, yet the fundamental requirement for human supervision remains.

Supporting Data: The Mechanics of Disbelief

The data labelers occupy a unique position in the Tesla ecosystem. While marketing teams focus on polished demos and Musk focuses on stock-moving projections, labelers see the "raw" performance. They see the near-misses that never make it to the evening news.

The Speeding Problem

The revelation that FSD routinely speeds is significant because it highlights a systemic failure in the AI’s "reward" structure. In machine learning, the system is often optimized for "smoothness" or "human-like" behavior. If human drivers in the training data frequently speed, the AI may perceive speeding as the "correct" way to navigate. Former employees noted that when they flagged speeding as an error, it was often ignored by higher-level engineers who were more concerned with preventing the car from "phantom braking" or hitting curbs.

The Hardware Divergence

A major point of contention within the industry is Tesla’s "Vision" approach. Unlike competitors such as Waymo or Zoox, which use a "sensor fusion" approach combining LiDAR, Radar, and Cameras, Tesla relies almost exclusively on cameras.

• Waymo: Uses expensive LiDAR to create a 3D map of the surroundings, allowing for precision in the dark or fog.
• Tesla: Relies on AI to "interpret" 2D images into 3D space.

The data labelers’ lack of trust often stemmed from seeing the "Vision" system fail to perceive depth correctly or misidentify objects in low-light conditions—failures that a LiDAR-equipped system might have avoided.

The "Safety Statistics" Dispute

Tesla frequently releases safety reports claiming that drivers using FSD have fewer accidents per million miles than human drivers. However, the former engineer interviewed by Reuters joined a chorus of independent analysts in disputing these figures. Critics argue that FSD is primarily used on highways—the safest type of road—whereas human driving statistics include high-risk environments like parking lots and narrow city streets. By comparing FSD’s highway performance to general human driving, Tesla may be creating a statistical "apples-to-oranges" fallacy.

Official Responses and the Corporate Stance

Tesla, which famously dissolved its public relations department years ago, did not provide a detailed response to the Reuters investigation. Historically, the company’s defense of FSD rests on three pillars:

1. Driver Responsibility: Tesla’s user agreements explicitly state that FSD (Supervised) is not a self-driving system. It requires a "fully attentive driver" who has their hands on the wheel and is prepared to take over at any moment. This legal shield allows the company to blame "misuse" for accidents.
2. Iterative Improvement: The company argues that the system learns from its mistakes. By having millions of cars on the road, Tesla collects more data than any other manufacturer, which they claim will eventually lead to a "solved" autonomy problem.
3. Safety Parity: Tesla continues to insist that its internal data proves the system reduces the likelihood of a crash. During earnings calls, Musk has reiterated that it would be "morally wrong" not to release FSD, as he believes it already saves lives.

However, the disconnect remains. While the company tells regulators the system is a driver-assist tool, it tells consumers (and investors) it is a precursor to a fully autonomous robotaxi. This "double-speak" is currently the subject of several investigations by the National Highway Traffic Safety Administration (NHTSA) and the Department of Justice.

Implications: The Future of Autonomous Trust

The testimony of the data labelers has implications that extend far beyond Tesla’s stock price. It touches on the broader societal acceptance of artificial intelligence.

Regulatory Scrutiny

Regulators are increasingly wary of "autonomy washing"—the practice of making a product appear more autonomous than it actually is. The Reuters report provides a roadmap for investigators to look into internal "priority lists." If it can be proven that Tesla knowingly deprioritized safety issues like speeding or failed to address known perception errors, the company could face massive recalls or even criminal liability.

The China Expansion

Tesla’s recent push into the Chinese market adds another layer of complexity. China is a leader in autonomous driving, with companies like Baidu and Xpeng operating advanced systems. If Tesla introduces FSD to a market with different driving norms and higher density, the failures witnessed by data labelers in the U.S. could be magnified.

The "Labeler’s Paradox"

Perhaps the most profound implication is the "Labeler’s Paradox." AI is only as good as the data used to train it. If the people training the AI are disillusioned and skeptical, it suggests a breakdown in the feedback loop. When workers feel that their flags for safety violations are being ignored by management, the quality of the training data suffers, leading to a stagnation in the software’s capability.

Conclusion: A Product Built on Precarious Ground

The Reuters investigation serves as a sobering reminder that "Full Self-Driving" remains a misnomer. The people who spend eight hours a day looking at the "brain" of the Tesla AI are the least likely to trust it with their lives.

As Tesla moves toward a dedicated Robotaxi launch and expands into global markets, the gap between the vision of a driverless future and the reality of a camera-based Level 2 system continues to widen. For the consumer, the takeaway is clear: the technology is an impressive feat of engineering, but it is far from the "unsupervised" savior promised on stage. Until the people who build the system are willing to sit in the back seat, the public would be wise to keep their hands firmly on the wheel.