Tesla's FSD Improves, But Report Warns of a 'Double-Edged Sword'

LOS ANGELES, CA – July 29, 2026 – Tesla's Full Self-Driving (Supervised) system is caught in a fascinating and perilous paradox. A new independent evaluation by AMCI Testing reveals that the latest update, version 14.3.4, has achieved dramatic improvements in vehicle control, yet it simultaneously introduces a subtle but significant danger: lulling drivers into a false sense of security. The report, based on over 1,000 miles of real-world driving, serves as a critical reality check on the state of consumer-grade autonomous technology, tempering celebratory confetti screens with a dose of sober analysis.

While the electric vehicle giant continues its aggressive push toward a driverless future, AMCI's findings highlight fundamental navigation errors and unanswered questions about adverse weather performance, casting a shadow over the company’s near-term Robotaxi ambitions and underscoring the immense gap that remains between advanced driver assistance and true autonomy.

The Psychology of Partial Automation

The core of AMCI Testing's findings is what its Director, Guy Mangiamele, calls a "double-edged sword." During extensive fair-weather testing, the camera-only system demonstrated strong, confident lane positioning and vehicle handling. This progress, however, is precisely what worries safety experts. As the system becomes more competent, it risks exploiting a well-known human cognitive flaw: confirmation bias. Drivers, seeing the system perform flawlessly 99% of the time, may become complacent and less likely to supervise the vehicle's operation closely.

"These steps forward continue to be a double-edged sword as they will tend to lull a driver into a false sense of security and not closely supervise the vehicle's driving while still not error free," Mangiamele stated in the release. This concern is echoed by years of human-factors research, which shows that drivers using advanced driver-assistance systems (ADAS) often overestimate the technology's capabilities, leading to reduced situational awareness.

This psychological trap is particularly hazardous because, from a legal and regulatory standpoint, Tesla's system remains firmly at SAE Level 2. This classification mandates that the driver must remain fully engaged, monitor the driving environment, and be prepared to intervene at a moment's notice. The responsibility for the vehicle's operation never shifts from the human to the machine. Despite the "Full Self-Driving" branding, the driver is not a passenger; they are an active supervisor, a role that becomes harder to maintain as the system's performance smooths out.

The Robotaxi Dream Faces a Reality Check

While Tesla celebrates milestones like 1.48 million FSD subscriptions and the rollout of features like "Actually Smart Summon," AMCI's report zeroes in on persistent weaknesses that challenge the viability of a near-term Robotaxi network. According to the independent firm, FSD (Supervised) made "multiple route planning and navigational errors that were simply unacceptable, particularly in a future Robotaxi application." A vehicle that cannot be trusted to navigate reliably to its destination undercuts the entire premise of an autonomous ride-hailing service.

Furthermore, the 1,000-mile test was conducted exclusively in fair weather. AMCI's David Stokols noted, "The next phase of our testing – less-than-ideal conditions like rain, fog, etc. will be the ultimate acid test of FSD." This is a critical point of vulnerability for any vision-based system. While Tesla's latest updates include an upgraded vision encoder to improve performance in low-visibility conditions, its real-world effectiveness on a massive scale remains unproven in independent, rigorous testing.

This stands in stark contrast to the company's ambitious goals, which include a target for unsupervised FSD in geofenced areas in select cities by the end of this year. The findings from AMCI suggest that while the underlying vehicle control AI is advancing rapidly, the foundational elements of reliable navigation and all-weather capability—prerequisites for a truly driverless service—still have a long way to go.

A Tale of Two Philosophies

The challenges highlighted in the AMCI report illuminate the starkly different philosophies competing to solve the autonomy puzzle. Tesla is pursuing a strategy of massive scale, leveraging data from millions of customer vehicles to train its neural networks with a camera-only hardware suite. This approach allows for rapid, iterative software improvements and a potentially more cost-effective path to a generalizable solution.

In the other corner is Waymo, Alphabet's autonomous driving unit, which is widely considered the current leader in Level 4 autonomy. Its approach is more akin to aerospace engineering: methodical, cautious, and reliant on a multi-sensor suite that includes expensive LiDAR and radar in addition to cameras. Waymo operates its fully driverless robotaxi service within carefully mapped and validated geofenced areas in cities like Phoenix and San Francisco, currently completing over 100,000 paid rides per week. Its safety record is formidable, logging over 12 million autonomous miles with only one minor at-fault incident.

This is where the role of independent evaluators like AMCI Testing becomes indispensable. In a market where a vehicle's value is increasingly defined by its software and intelligence, AMCI's unbiased, real-world analysis helps cut through the marketing hype. It provides regulators, investors, and consumers with a clearer picture of what a system can—and, more importantly, cannot—do today.

Navigating the Regulatory Maze

Ultimately, the path from Level 2 to a Level 4 Robotaxi is paved not just with technological breakthroughs but with regulatory approvals. To make that leap, Tesla must prove its system is robust enough to shift legal liability from the driver to the machine. This requires a comprehensive safety case that satisfies stringent requirements from bodies like the National Highway Traffic Safety Administration (NHTSA) and the California DMV.

Recent updates to California's regulations, for example, now require manufacturers to submit a detailed, structured safety case and complete hundreds of thousands of miles of testing at each phase of deployment. In Europe, new laws like the EU AI Act are imposing strict obligations on high-risk AI systems, demanding proven robustness, data governance, and human oversight. The bar for deploying a truly driverless vehicle on public roads is exceptionally high, and AMCI's report suggests Tesla, for all its progress, has not yet cleared it.