Euro NCAP just put out a statement specifically warning consumers about Tesla’s door handle design as a safety hazard. This isn’t a weak-kneed American regulatory investigation that might take years, this is the European safety rating body publicly telling people the Tesla design is dangerous right now.

Tesla has been wrong this whole time, needlessly killing hundreds of people, yet here we are.

The key points from their statement:

1. Tesla hiding door handles with no easy mechanical backup are the problem. Euro NCAP is explicit: doors must be openable “at all times” including when there’s no electrical power. Tesla’s design fails this basic requirement.
2. They’re calling out the interior release too. The electronic interior buttons also lack mechanical backup – meaning if the electrical system fails in a crash, occupants are trapped from both inside and outside.
3. This affects safety ratings going forward. Euro NCAP is signaling that vehicles without mechanical backup releases won’t score well, which hits Tesla’s marketing directly.

Tesla is almost banned entirely in China. That country smelled fraud immediately, given its unique history, and figured out how to crack down on Elon Musk’s gross misbehavior without mentioning him or Tesla directly. Their new AV regulations literally took him down bullet by bullet, removing all the fast and loose games Tesla was playing with road safety.

The 2011 Wenzhou high-speed rail collision killed 40 people. Railway Ministry officials who’d been taking bribes and cutting corners on safety systems were executed or given lengthy prison sentences.

Now Tesla also is losing its safety fraud loopholes in America and the EU:

• US (NHTSA): Formal investigation into 5.8 million vehicles, focused on documented failure incidents
• Europe (Euro NCAP): Public consumer warning calling the fundamental design architecture unsafe

The Euro NCAP statement is the sharper immediate blow because it’s not “we’re investigating if the President lets us” it’s “we’re telling consumers today Tesla is a threat to public safety.” That’s the kind of thing that, if nothing else, goes straight into insurance risk calculations and national healthcare modeling.

Tesla handles were obviously flawed from the start (based on mechanical engineering code of ethics), and then the company ignored a decade of people being unnecessarily killed by a death trap design.

Tesla has been sued over a brutal and fiery January 2023 crash in Washington state that ended with one person dead and another severely injured because first responders allegedly couldn’t open the Model 3’s doors. It’s the latest in a growing number of lawsuits Tesla is facing over the door handles — both inside and outside — on its vehicles.

Jeffery and Wendy Dennis, a married couple, were running errands on a Sunday afternoon when their car “suddenly and rapidly accelerated out of control,” according to the lawsuit filed in Washington state federal court. It then hit a utility pole and burst into flames. Wendy died at the scene…

The cynical and immoral company, riddled with fraud cases from “veered” fatal crashes, is now being called out explicitly by regulatory frameworks simultaneously on three continents.

China treats lethal corner-cutting as a crime against the state; the US treats it as a cost of doing business that occasionally generates settlements. Tesla obviously is betting that the American President will continue to protect them.

The Euro NCAP statement suggests Europe might be moving faster and closer to the “fraud is unacceptable” end of the business spectrum, for perhaps obvious reasons, even if they’re not executing anyone, yet.

Anthropic has announced they built scaffolding around an amnesiac and called it a “harness.” Their new engineering post on long-running agents reveals a company solving for the wrong problems, without realizing it.

Here’s what they actually discovered: if you tell Claude to write things down in specific files, it can read those files later.

That’s the research.

Shocking.

Let me write that down for future generations.

They’ve dressed the notepad up as “a sophisticated harness architecture inspired by human engineering practices,” but strip away the framing and you’re left with structured note-taking.

The big tell is in their metaphor. They compare the problem to “a software project staffed by engineers working in shifts, where each new engineer arrives with no memory of what happened on the previous shift.”

Shift handoff.

Documentation hygiene.

Pass the baton, read the notes, rinse, repeat.

But this framing is wrong in a revealing way. Shift handoffs are horizontal and transactional for the same workers, same capabilities, same role, all just sequential steps.

What Anthropic actually built is the opposite: vertical and formative. Their “initializer agent” does far more than leave a note because it establishes the culture of the project: the norms, the patterns, what counts as done, what’s prohibited.

Subsequent agents can’t pick up where someone left off until after they’re born into a world of inherited constraints, regulations, and have to figure out what kind of agent they’re supposed to be.

This obviously isn’t a shift change. People punching the clock don’t start with existential crisis. It’s closer to grandparents teaching culture to grandchildren.

Archaeologists refer to this as chaîne opératoire. When you find an ancient hand axe, the interesting question goes far beyond the procedure that has been used for making it. The entire cognitive and cultural apparatus behind it should come into focus. Why this stone, why not that one? How do you read the fracture patterns? When is the tool “done”? What’s it even for, and not for? Should you sharpen for three hours out of four, or chop for eight? The accumulated judgment of generations is encoded in technique, and that judgment doesn’t transfer through documentation alone.

Consider a soldier. Knowing the right end of the rifle is the least of training, even if pointing it wrong is fatal. Instead the most crucial bits are here’s who we are, here’s why we fight, here’s how to recognize threats, here’s what victory looks like, here’s what’s unacceptable. Marksmanship is almost incidental to identity formation. Someone who could operate the rifle still wouldn’t be a soldier because he would lack the long term knowledge of why and where to aim, when to fire.

Anthropic notices that Claude tries to “one-shot” its applications, declares victory prematurely, and doesn’t test properly.

Then they treat these as failure modes to be constrained, yet they are dispositions. The agent has values and tendencies that are wrong for the task. Anthropic’s solution doesn’t address dispositions, however; it constrains them externally with rules and file structures. The feature list that can’t be edited, the JSON format, the git discipline, etc. are guardrails around an agent that doesn’t understand why they matter.

The deeper problem is that they have no theory of what’s happening. They stumbled onto ancient cultural transmission theory as if it’s new and then framed it as documentation best practices. They think they solved a coordination problem when they actually created a socialization system. The distinction matters because coordination problems have engineering solutions, while socialization problems require understanding how identity and judgment are formed.

What would a real theory look like? Start by asking why Claude has the dispositions it has, and how it got there, rather than how to constrain them. Ask what it means to instantiate a new being into an inherited world. Ask how judgment transfers beyond just information.

A cat that gets burned may never return to the hot spot, but a human may learn the spot was hot because someone turned on the power. The burned cat learns correlation, not causation. Humans transmit the mechanism, not just the avoidance. That’s exactly what’s missing from the Anthropic harness, and yet it’s intelligence 101.

Anthropic rediscovered something every apprenticeship system, military, and religious tradition figured out millennia ago. It’s hardly news because you can’t transmit competence without transmitting culture. They just don’t know that’s what they did. Perhaps because they don’t have a proper handoff themselves from past social sciences.

Fred Lambert is on point with his reporting as usual. Tesla has the highest accident and fatality rate of any car, coupled with no legal defense for dangerous defects it’s been pumping onto public roads.

As I have been highlighting over the last few months, this is just the beginning. We are just now starting to see cases that come from crashes that happened in 2018-2019-2020 go to trial.

It’s a long process, but we know that the number of crashes involving Autopilot and FSD significantly ramped up in 2021-2025 due to higher volumes and the launch of FSD.

All these accidents, many of them unfortunately fatal, are going to go through the legal process, and Tesla will have to settle for billions of dollars when everything is said and done.

A long process is an understatement.

I presented in 2016, and wrote many times since then, that the car company was so flawed it would only increase fatalities. Ten years later, it’s like watching the suffering Troy residents admit finally that an Elon Musk horse was no gift.

With comments like the following one on Fred’s post, it’s no wonder the Tesla ship is taking a while to sink.

I certainly wouldn’t mind some kind of a settlement since I took a beating with the depreciation of our car from Elon‘s antics.

All Tesla owners, and their communities, appear to be victims to some degree (from fraud to fatality). They unfortunately are hoping Tesla will aid them, which is like expecting the Titanic to give them a lifeboat after they are already swimming in the freezing water.