The most consequential fraud in modern technology is not happening in the code. It is happening in the units.

If you ever studied the collapse of Soviet economics, you know exactly what I’m about to explain.

AI companies have built a billing infrastructure in which the seller defines the unit of measurement, counts the units, and invoices the buyer. All with no independent verification at any point in the transaction. All without any enforcement mechanism.

If you prompt AI to build something and it launches a dozen agents and burns an entire day worth of credits in an hour, that’s business as usual, especially if they delete their own work and complain they have nothing to show you for it.

The unit of fraud is called a “token.” It has no fixed definition. It varies by model, by provider, and by tokenizer version. It can be changed at any time, by the vendor, without notice. There is no regulatory body certifying token measurement. There is no weights-and-measures regime. There is no audit trail the customer can independently verify.

This is not a new problem, as I already hinted.

It is one of the oldest problems in commercial history, and every previous instance ended the same way. It won’t be different this time. It’s logic any five-year-old should be able to figure out.

Caps for Sale

Let’s start with clause 35 of the Magna Carta, 1215:

Let there be one measure of wine throughout our whole realm; and one measure of ale; and one measure of corn.

This was the language of liberty from oppression. It was a response to documented, systematic fraud by royal merchants who controlled their own measures. A bushel in London was not a bushel in York, and the difference was profit.

It took England six centuries to arrive at a proper Weights and Measures Act. Every iteration addressed the same structural deficiency: when the entity selling the goods also controls the unit of measurement, the unit will be corrupted. The entire history of metrology from the Bureau International des Poids et Mesures to NIST to the EU’s Measuring Instruments Directive, is the history of forcibly separating the measurer from the seller.

It’s fundamental to the rise of industrialization that the clocks had to run on universal time, even with time zones, such that trains could have externally judged arrival and departure times. The British and Dutch factories that invented assembly lines to defeat Napoleon (infamously copied by Ford) couldn’t work without shared units of measure.

Given this context it appears now that AI companies are the most historically illiterate and economically unsound ever.

Their “token billing” has undone a fundamental tenet against trivial fraud. We are back to the royal merchant having their thumb on the scale for every transaction, except the thumb is an algorithm and the scale is proprietary.

How dumb does the intelligent machine business think we are, seriously?

LIBOR for Compute

Let’s review, for example, the London Interbank Offered Rate (LIBOR) that underpinned roughly $350 trillion in financial instruments worldwide. LIBOR was calculated from self-reported borrowing rates submitted daily by the banks that profited from the number. No independent verification. No transaction-based measurement.

Just trust.

And it failed. Banks manipulated it for years. Of course they did. The entity producing the number was also the entity whose trading positions depended on the number. When the fraud was finally exposed, the fix was to replace LIBOR with SOFR (Secured Overnight Financing Rate) which is derived from actual observed transactions rather than self-reported claims.

Now consider the AI jar of pickles we are being told to get in.

OpenAI reports that average reasoning token consumption per organization has increased approximately 320 times in the past twelve months. This number was produced by OpenAI, about OpenAI’s product, using OpenAI’s proprietary tokenizer, and reported to the press as evidence of adoption. It is Barclays submitting its own LIBOR rate as if nobody knows why we stopped them from doing this.

The difference is that LIBOR at least had the pretense of multiple submitters. Token counts have one source: the vendor.

Intelligence machine vendors have truly produced their most cynical moment.

Gosplan of Sand Hill Road

Soviet central planning failed not because the planners were being stupid. Many were brilliant, which probably made everything worse. It failed because the information system was structurally corrupt, and compliant agents corrupted it further. Every layer of the reporting chain had an incentive to inflate their output numbers, and there was no independent verification mechanism capable of correcting the distortion.

The famous case study is the Soviet nail factory. Measured by weight, the factory produced fewer, heavier nails that nobody needed. Measured by quantity, it produced millions of tiny nails nobody could use. The metric became the product. Actual utility was irrelevant because utility was not being measured, only the unit was.

Here’s another token output example of fraud I was taught in college. Soviet window manufacturers measured weight and nobody could install the heavy, thick glass. They measured by size, and all the very large, thin glass broke before it even could be loaded for delivery. Actual utility was irrelevant because utility was not being measured, only the unit was.

Every day that I use AI it wastes unbelievable amounts of money and time, measured in units of tokens, as it tells me if I don’t like it there’s nothing I can do.

Jensen Huang’s proposal at GTC this month is the Soviet nail or glass factory at much larger Silicon Valley scale.

He suggested that every engineer should have an annual token budget, where these allocations could reach half of base salary in value. Consider what this fraud means structurally. You are telling workers they have an annual allocation of a unit that measures interaction volume, not outcome quality.

Record scratch.

So a notoriously wasteful industry already in trouble for water and air pollution will optimize entirely for high consumption. An engineer who solves a problem by thinking for ten minutes and never touching the AI has, under this framework, underperformed relative to one who burned through a million tokens generating refuse. Yet the engineer who still thinks, and conserves tokens, is undeniably the superior engineer to the ones that do not!

Pray and spray, running out of ammunition and begging for $200 billion to keep firing at ghosts, is so inversely proportional to the efficiency of Delta operators I can’t even….

Tokens are not a productivity metric, like ammunition is not even a kill rate, because Nvidia is incentivized inversely to what customers actually need. It is Gosplan announcing the Five-Year Plan for compute consumption, and every factory manager is about to start filing reports showing they exceeded their quota of tokens, meaning… nothing.

Shovel Seller Tithe

Huang’s position is particularly elegant because Nvidia does not sell tokens. They sell the GPUs that generate them. Every token consumed requires silicon to produce. If token budgets become a standard corporate expenditure pegged to payroll, Huang has created a permanent demand floor for his hardware.

Gross. Literally gross product.

He does not need to manipulate the token count himself. He just needs the token to become the unit that corporations manage against, and every dollar allocated to token budgets flows upstream to GPU purchases.

He skips actual measurement. He proposes that companies commit, in advance, to spending a fixed percentage of their payroll on his product for compute.

That is not a metric. It is a tithe.

And the structure insulates him perfectly. The AI providers already grossly inflate the token counts. The customers overpay the AI providers, given that most of the token count is for fixing things the tokens were spent on to begin with, like a protection racket. The AI providers buy Nvidia’s GPUs to service the consumption they have encouraged and caused without any accountability for outcomes. Nvidia never touches the books. They sell shovels to the people salting the mine.

The Arc

Every instance of self-reported commercial measurement in recorded history has followed the same progression: self-reported measurement, then market adoption of the metric, then discovery of systematic manipulation, then regulatory intervention mandating independent measurement.

Medieval grain measures. LIBOR. Credit ratings. Remember Facebook’s video metrics? The company admitted in 2016 to inflating view times by 60 to 80 percent, having defined the view, counted the view, and sold the view. The pattern is not debatable. It is one of the most thoroughly documented dynamics in economic history.

Token billing is currently at stage two: market adoption. Enterprises are building budgets around it. Analysts are publishing reports denominated in it. A CEO is proposing tying it to compensation.

Nobody is asking who audits the count.

Auditors are completely absent.

The harsh reality for every major AI provider on earth, like royalty before the Magna Carta, is that nobody has the independent authority needed to vouch for them. The merchant is being made the king who declares their own scale valid no matter what. And this time the scale is processing trillions of transactions per day, denominated in a unit that has no legal definition, no regulatory oversight, and no independent verification mechanism.

No kings.

We have eight hundred years of evidence for this bullshit. The only variable today is how much it costs before someone reads basic history of economics and enforces an honest measure.

The AI industry pretends to be terrified about regulation, but really they are in danger of transparency. Because the moment an independent third party can compare token billing against actual computational work performed, or the moment someone builds a SOFR for inference, every provider’s margins become visible. And if those margins look anything like LIBOR spreads or Facebook’s video metrics, the correction won’t be gradual.

I’m telling you, even the best of the best agents are a tragedy of token inflation and massive waste.

Nobody inside the Soviet system volunteered for glasnost. It was forced by the fact that the gap between the reports and reality had become so grotesque that the system could no longer function even on its own terms.

Token economics in Silicon Valley is rapidly approaching that threshold. Engineers know. We watch agents burn through whole budgets producing garbage, watch our token counts spike on failed reasoning chains we are billed for anyway, watch “reasoning tokens” appear on invoices for computation we never requested and cannot inspect.

The bigger the tool failure and productivity suck, the more the AI companies try to report a Soviet-sounding productivity “gain”. The more energy they burn, the more they claim to have a “big engine”, which means literally nothing useful.

Gorbachev didn’t reform Soviet economics. He revealed that it was dead inside.

The production numbers had been fraudulent for decades. Everyone inside the system knew. The factories knew. The ministries knew. Gosplan knew. But the reporting structure made it impossible to say so, because every career in the chain depended on the numbers going up. Glasnost (openness) didn’t fix fraud any more than exposure of Enron balanced its sheets. It made it permissible to say out loud the numbers meant nothing. The gap between reported output and actual value had grown so large that the moment anyone was allowed to measure honestly, the entire structure lost legitimacy overnight.

That’s the truth of the AI bubble. Token output is the absolute wrong measure and will only bring pain to those who adopt it without audit.

Sitting at my desk in San Francisco, sometime around 2016, I got an email from a master sergeant. A recording of a firefight. And then a phone call. Play it, he said. So I played it. Gunfire. Chaos. Then a voice: Grenade.

Play it again, he said. Listen.

I played it again. The voice wasn’t scared. It was factual. Grenade. Like hearing “pastrami on rye” at a deli counter. The same calm. The same precision. The same total awareness of what’s happening and what matters right now.

That’s fire discipline. That’s what Delta is world famous for. Sprint and fire at maximum efficiency, nothing wasted. The decisive calm that comes from knowing exactly how many rounds and what each one needs to do. Every shot placed. Every word functional. Nothing lost to bravado or panic, nothing spent on performance.

Play it again, he said. What’s really happening in this fight? What works? What’s blunder? Do you hear it?

I’ve been thinking about that “grenade” for two weeks. Because the sound coming out of Operation Epic Fury, and the war crime theatre of Hegseth, is the exact opposite of that voice. It’s all fluff. All blunder. Do you know what the Pentagon sounds like right now? 1950 Korea. And China is listening to the same tape.

…during the 2016 campaign, Trump repeatedly declared that Douglas MacArthur was his “favorite general.” At rallies, Trump would invoke MacArthur’s name almost as though he were in direct communication with his ghost. […] MacArthur had been outwitted and outflanked by a guerrilla army with no air force, crude logistics, and primitive communications, an army with no tanks and precious little artillery. As David Halberstam put it, MacArthur had “lost face not just before the entire world, but before his own troops, and perhaps most important of all, before himself.” All of this happened because MacArthur was almost criminally out of touch with reality.

And so here we are.

The Deterrence Illusion

The “China is deterred” narrative runs something like this: the US popped Venezuela in January with overwhelming ratios, like blackouts for 3 million people to arrest one guy. Then the US blockaded a weak Cuba, as it has no sea defenses, and then Trump launched a surprise attack on Iran while his own negotiations were actively ongoing, decapitated the supreme leader, and declared total air superiority within 72 hours, boasting “not a fair fight”.

Beijing must be so scared now.

Yet this analysis assumes China’s baseline expectation was about Iranian air defenses working. Nobody thought they would. Certainly not Venezuela. Certainly not North Korea, which has spent decades putting everything that matters under granite for exactly this reason. Certainly not China, whose military planners have been studying US strike capabilities since long before Desert Storm.

What China actually witnessed in American hamfisted pray-and-spray salvoes wasn’t just “America bomb things, America make fire and noise, America so nasty.”

They knew that. What they got was specifics about weakness: F-35 and F-22 electronic signatures under combat conditions, operational tempo sustainability, jamming profiles, cyber-kinetic integration patterns, kill chain logistics from ISR to strike. That’s the structured data leak you never get from exercises or satellite imagery. As technologist Amir Husain put it in the Jerusalem Post, the American rush into unilateral war is “a dataset goldmine for China” for building automated detection and threat classification models.

The US demonstrates the opposite of deterrence. It held an unnecessary live-fire exhibition with free admission to telegraph its entire playbook, revealing the entire spectrum of Trump’s options and thoughts.

MacArthur’s Ghost

Those who say China should fear American willingness to fight clearly forget Douglas MacArthur wasn’t fired for being unwilling to fight. He was fired for being so willing he nearly started a nuclear war with China. Truman understood that willingness without discipline is what the Greeks long ago classified as the most self-defeating capability a military can possess. It’s like running into a minefield.

The historic parallel maps cleanly for China, not least of all because among the regressive all-show-no-go white supremacists that Trump calls out as his role models, MacArthur is right up there. The thunder and lightning had worked once for MacArthur, so he thought escalate, go bigger, roll right up to the Chinese border. He pushed to the Yalu drunk on the Inchon success, without a clue. Then 300,000 PLA troops crossed the Yalu and pushed him back to roughly where he started. That willingness was the vulnerability. It showed Beijing exactly when and how to intervene.

The same mindset that made MacArthur unable to read Chinese capabilities in 1950 is operating in the analysts who think China is “deterred” by watching the US flatten Iran. They don’t see China as the disciplined party, because their framework doesn’t allow it. The French Generals turned off their radios while the Germans rolled tanks through the Ardennes. MacArthur’s intelligence staff stopped reporting Chinese troop movements because he’d made clear he didn’t want to hear it.

To be clear, MacArthur’s father was the general known for genocide in the Philippines.

…Arthur MacArthur … brought to the archipelago the genocidal mentality that accompanied their warfare against Native Americans in the American West. Filipinos were branded “n—ers” by U.S. troops, though another racist epithet, “gugus,” was also widely used for them. When Filipinos resorted to guerrilla warfare, they were dehumanized … to legitimize all sorts of atrocities against them. The war of subjugation was carried out without restraints with General Smith ordering his troops to convert Samar into a “howling wilderness” by killing any male over 10 years old.

Chinese political scientist Zheng Yongnian told the South China Morning Post that “America’s war-making capability depends solely on its will to deploy such power.” The deterrence pundits read that as a compliment, when it’s actually a cynical vulnerability assessment. A power that acts on will rather than calculation can be drawn into commitments that exhaust it into extreme embarrassment.

”I tried to stop it, but I don’t own the licensing rights.” [Stallone told movie critic Siskel that during the holidays, he had been asked to give away a “truckload” of Rambo toys to sick children in the hospital.] ”I told ’em, ‘Get this … the hell out of my driveway and burn it. Don’t give it away,’ It’s not for kids. The movie was not supposed to be for little kids, and I wouldn’t let my own children play with those toys.”

Mao understood this about MacArthur in 1950. Montgomery understood this about Rommel in 1942. The more willing, the further he extended, the more exposed he became, the weaker. China, like Britain, waited for the overreach, then moved.

Trump overwhelmed Venezuela as it sat quietly, Trump launched a blockade of empty waters around Cuba, and Trump tricked Iran into believing negotiation wasn’t a lie. None of that projects strength. Now he appears as just a tail on the Israeli dog, if not the flea, pulled into the largest invasion of Lebanon since 2006. With what to show for it? Emptied stockpiles and wreckage, requests going to everyone to help him get out. Taiwanese analyst Cheng-Yu Wu assessed that the PLA learned Trump’s administration “will do whatever it takes to achieve its own national interests, whether or not there are negotiations.”

Some analysts really think they can frame that dictatorship signal as deterrence. A drill sergeant would call it a fighter closing his eyes, throwing wild haymakers and claiming whatever he hit is losing. Both descriptions can be true. Only one of them is meaningful to a trained opponent.

Fire Discipline

That voice on the recording knew something everyone is supposed to learn before stepping down range. Don’t fire until you see the whites of their eyes wasn’t academic poetry. It’s always been scarce resource management. The NRA was founded by Union Generals who said 1,000 rounds per kill was unsustainable to defend American Blacks against the KKK. Every round you fire at nothing is a round you don’t have when it matters.

The US is emptying interceptors against cheap Iranian drones. Burning Tomahawks on targets that were already assumed destroyed, or worse, killing nearly 200 little girls at school. America is expending precision munitions far faster than the industrial base will replace them. And doing all of it on camera for everyone’s intelligence collection, let alone China.

The Heritage Foundation warned before the Iran war started that SM-3, SM-6, PAC-3 MSE, and THAAD interceptors would be exhausted within days of sustained PLA salvoes. Aggregate US vertical launch system inventories were insufficient for even one full fleet reload. CENTCOM officials thus warned of a “Winchester” scenario: complete ammunition depletion.

Now look at what draft-dodging Trump ordered since February 28. The US burned over 2,000 precision munitions against more than 3,000 targets, only to announce repeatedly it’s not done yet. Allies have fired hundreds of interceptors. THAAD components have been redeployed from South Korea. The Abraham Lincoln carrier strike group was pulled from East Asia. Patriot interceptors, which Lockheed produces at about 620 a year, were depleted in the hundreds over the first 36 hours alone. We’re still debating who shot down the three F-15E in one night so efficiently without a trace, evaporating the mythology of American Air superiority.

And on the other side of the ledger: China has barred the export of rare earth elements for military use, which means the exact materials required to build the missiles the US is burning through. China is watching the US deplete stocks it can’t rapidly replenish because China controls the supply chain for those replacements.

The Asymmetry

Compare the two sides and see what America has been doing to itself, unprovoked.

	United States	China
Munitions	Depleting	Accumulating
Intelligence	Broadcasting	Collecting
Carrier groups	Redeployed to Gulf	Positioned in Pacific
Strategic reserves	Drawing down	Building up (104 days coverage, projected 140-180 by year end)
Industrial base	Years behind demand	Expanding offensive capacity
Rare earths	Import-dependent	Export ban in place
Diplomatic posture	Overcommitted on four fronts	Restrained, summit-focused

That’s not passivity on China’s part. That’s basic discipline. China is doing what any wise fighter does when the opponent is dancing and swinging wildly to amp up the audience: cover and wait, read the rhythm, count the punches, feel the decline, watch for the opening.

Nixon’s Tar Baby

Foreign Policy drew a parallel to 1964, when the Peking Review described US interventions in the Congo as Washington’s “second South Vietnam”, about keeping American assets tied down far from China’s borders. But that reference is far too diplomatic. The actual historical pattern is worse, and it has a name Americans should be embarrassed about.

Kissinger and Nixon adopted NSSM 39 in 1969 called the “Tar Baby” option. Their policy was strengthening ties with racist white-minority governments in Rhodesia and apartheid South Africa to deny Blacks power. The premise was that apartheid was an unpleasant but permanent reality, and Washington should accommodate it pragmatically and prolong white nationalism. Its own internal review later concluded that its only real result was to mire the United States deeper on the side of obvious oppressors. The name told you the outcome of the strategy. You punch the tar baby and you get stuck.

The destabilization model ran in parallel. Congo’s leader Lumumba was assassinated in 1961. The CIA sent poison to its station chief; when that fizzled, Lumumba was deposed in a CIA-backed coup and shot by Congolese assassins. UN Secretary-General Dag Hammarskjöld was shot down by U.S.-backed mercenaries, as he flew in to negotiate a ceasefire. Mobutu seized power with US help and misruled the country for three decades. Congo never recovered. Similarly Mondlane, a professor in America positioned to become leader of Mozambique, was assassinated in 1969.

Same trajectory.

Stuart Reid has put it precisely: for the Congolese people, the events of 1960-61 represented the opening chapter of a long horror story. For the US government, they provided a playbook for future interventions.

The playbook: assassinate leaders, destabilize the country, back the regional oppressor as the “stable” partner, let the region deteriorate, use the chaos to justify permanent intervention. The people who live there never recover. The strategic partner gets a free hand. And the great-power competitor watches you get stuck to the tar baby.

Apartheid South Africa formalized this as doctrine. P.W. Botha’s “Total Strategy” required a buffer of deliberately failed states on their border. They used the ugly term cordon sanitaire. The logic was self-sealing: a thriving Black-governed neighbor would quickly falsify the racist claims that only whites were capable of self-rule, so the Black neighbor had to be destroyed. South Africa armed RENAMO to terrorize Mozambican civilians, backed UNITA through decades of Angolan civil war that killed half a million people, and used proxy forces to turn a drought into a famine that killed over 100,000. The manufactured chaos confirmed the racist ideology that manufactured it.

Substitute “Palestinians” or “Iranians” and the sentence requires no other edits. Gaza flattened. Lebanon invaded. Iran decapitated with no successor structure. A stable, self-governing neighbor is an existential ideological threat to an ethno-supremacist state. The vacuum is the feature. In the latest news from Israel troops call non-Jews dogs and stop their vehicles to shoot people in the head at close range.

Just like ICE in Texas.

Just like ICE in Minnesota.

It’s a procedural thing.

The US knows this model so intimately because it ran the active version against the Soviets. In 1979, Brzezinski’s explicit goal in Afghanistan was, in the words of Defense Department official Walter Slocombe, “sucking the Soviets into a Vietnam quagmire.”

It worked.

The Soviets bled for a decade and many argue the stickiness was what accelerated power collapse. Now the question is whether China needs to engineer anything at all, or whether the US has been punching into a tar baby scenario on its own initiative while Beijing simply watches.

Kissinger dismissed the cost of millions killed as “the unhappiness of a bunch of Africans and the self-righteous indignation of a few minor NATO allies.” That contempt is the through-line. The people destroyed by the policy don’t register as costs. They’re externalities. Then as now.

The Missing Synthesis

I bring all this up because the gap I keep seeing in the pundit class is that almost nobody is synthesizing all the threads simultaneously: the munitions math, the intelligence exposure, the fire discipline asymmetry, and the destabilization pattern that ties them together.

The Heritage guys get stockpile numbers. The intelligence community people get collection problems. The strategists get the overextension risks. But nobody puts the full picture together, perhaps because the conclusion is too uncomfortable? Trump is systematically degrading American capacity to fight the war it actually needs to deter, while running the same Nixon playbook that failed across southern Africa for three decades.

Oh, I remember the problem now. Nobody studies African history.

Khamenei was decapitated February 28. The Assembly of Experts was bombed while meeting to elect a successor. No governance structure is left. And within days, Israel launches its largest ground invasion of Lebanon since 2006, explicitly modeled on the genocide in Gaza. No really, an Israeli official told Axios: “We are going to do what we did in Gaza.” Already 800,000 Lebanese civilians are displaced. Nearly 800 killed. The imperfect ceasefire was at least something, until the US removed the one actor whose deterrent capability was constraining Israeli expansion.

Iran was a threat, but Israel reframed it into the leash. Remove the leash and the immediate result is genocide expanded into Lebanon, conducted openly, described in those terms by the people conducting it. The regional partner gets a free hand.

Korean War Arithmetic

One of the hallmarks of 1950s military failure was the US flattened every standing structure in North Korea. Ran out of targets, just as Trump says today. The war ended in a stalemate on roughly the same line it started. The shock and awe, air superiority, technological dominance, more bombs dropped faster than ever before, worked for a minute as domestic propaganda and not at all as strategy.

Same pattern now. Trump declared Iran had no navy, no air force, no radar and “just about everything’s been knocked out” within days. And then? Two weeks later, Iranian drones are still destroying billion-dollar radar systems and THAAD subsystems on video. American soldiers are dead. Domestically manufactured surface-to-air missiles are shooting down $32 million Reaper drones. Civilian trucks are launching ballistic missiles produced en masse.

The grind is on. Zero ground gained. No articulated end state. And every day it continues, China’s relative advantage in the Pacific grows, again not because China is doing anything at all, but because the US is spinning like a drunk doing everything, everywhere, all at once, to itself.

North Korea drew the simple conclusion from American air power decades ago: you can’t stop the bombs, you can make the bombs irrelevant. Eritrea knows exactly the formula too, as it used the same rubric to defeat the largest standing army in the world. China, full of ardent historians, has been watching that model. Hardening, dispersal, redundancy, underground facilities, quantity over quality in offensive systems. The Iran war is leaking all the exact parameters needed to calibrate against.

The analysts celebrating American willingness are celebrating their MacArthur heritage for all the wrong reasons. Drunken, stupid overreach hasn’t become a virtue just because the explosions look good with social media tricks. The tar baby was a disaster, in the way everyone learns the most powerful tiger in the world never escaped La Brea.

Play the tape again. Listen. Grenade. Calm. Factual. Disciplined. That’s the voice China recognizes, and right now the US sounds nothing like it.

A Tesla Cybertruck running Autopilot on the 69 Eastex Freeway in Houston reached a Y-shaped interchange last August and chose the wrong path. Where the road curves right, the vehicle drove straight into a concrete barrier. Driver Justine Saint Amour disengaged the system and grabbed the wheel. She was too late, because Tesla drivers always seem to be too late.

The resulting lawsuit, filed in Harris County district court and reported by the Houston Chronicle, seeks more than $1 million.

It claims a compound design failure: no LiDAR, no effective emergency braking, and a CEO who unsafely overrode his own engineers.

That last allegation is not rhetorical. It is a theory of the case. And it fits a pattern that runs well beyond the Eastex Freeway.

Tesla’s Own Taxonomy

As I’ve presented and written for over a decade, LiDAR in cars reads curves. It uses light to measure three-dimensional space ahead of the vehicle. Tesla’s own engineers recommended it. Every serious competitor — Waymo, Cruise — built their systems around it. Only Tesla’s CEO Musk objected, presumably on cost concerns. He called it “freaking stupid” and chose cheap consumer-grade low-quality camera arrays instead.

And then? NHTSA’s investigation EA22002 analyzed 467 Tesla crashes and found 111 roadway departures where Autosteer was inadvertently disengaged by the driver. Almost all occurred within five seconds. The agency also found that Autopilot resisted manual steering inputs: the system is setup as an unaccountable death-trap, actively discouraging corrections that it simultaneously requires.

Tesla’s engineers built an internal taxonomy for their CEO’s design flaws. They run a crash database query program allegedly called Cabana. Mode confusion is one of their own category labels: the driver believes the car is steering when it isn’t.

When Tesla denied in formal litigation that more than 200 such crashes existed, their own lawyer corrected the number in open court. Tesla then filed Recall 23V838 covering two million vehicles for exactly this failure — while explicitly stating in the filing they did not agree the defect existed. A quiet software patch. No redesign. NHTSA found crashes continued and opened an inquiry into whether the recall even worked or was an attempt to cheat safety regulations. Tesla is now fighting in federal court to keep expert testimony about the recall and the pattern away from juries.

Back to Saint Amour’s dashcam footage, provided to the Chronicle by her lawyers, it shows the death sequence clearly: ramp, fork, divider managed, turn begun, then blind and straight into the sidewall.

Hood open. Body panels separating. She was diagnosed with two herniated discs in her lower back, one in her neck, sprained wrist tendons, and neuropathy in her right hand. The next object past the barrier was the freeway below. She is lucky to be alive, given how many Tesla has killed so far.

The Man

The Saint Amour complaint alleges negligent hiring and retention of Elon Musk as CEO, asserting that his participation in product design decisions contributed to unsafe outcomes. Tesla’s own engineers recommended the sensor that improves safety. He overruled them. That is the obvious paper trail the lawsuit is walking.

To really understand this straight-line, no curves, case it is helpful to see the rigidity and commitment originates long before the current timeframe.

Musk grew up in Pretoria under apartheid. His maternal grandfather, Joshua Haldeman, was, according to Errol Musk’s own account, “fanatical” in support of apartheid and supportive of Nazism. Haldeman had fled Canada after being arrested as an enemy of the state, to help build apartheid South Africa instead. Errol Musk was elected to the Pretoria City Council in 1972 and ran a construction and engineering business wealthy enough to support two homes, a yacht, a plane, and five cars. He dealt in emeralds from a Zambian mine — confirmed in his own interviews. Elon left at 17, to avoid military conscription, but mostly because USAID had successfully brought down apartheid in 1988 and he couldn’t handle any curve to the political road ahead.

Apartheid was a system opposed to anything but the same, arguing against change while the white nationalist infrastructure murdered anyone who dared to diverge. Its architecture — pass laws, Bantustans, labor controls — was built to deny any need for change that was already visible and documented. The demographic and political trajectory of southern Africa was known. The system refused to see or change anyway. It finally crashed in 1994.

The beneficiaries of that system were not confused about what it was doing. They were experiencing, in the language Tesla’s engineers later coined, mode confusion in reverse: trusting a system that had already disengaged from any viable future.

Dismantling the Steering Infrastructure for Curves

The U.S. Civil Rights Division was established in 1957 precisely to institutionalize accountability for straight-line power — to build enforcement mechanisms that would force institutions to negotiate with legal and demographic reality. Since the Trump takeover January 2025, 70% of its attorneys have departed. Roughly 250 lawyers — the voting rights section, the police accountability section, housing enforcement — are gone. The new leadership has redirected the division toward investigating noncitizen voters and “anti-Christian bias.” The Fair Housing Act no longer appears in its housing guidance. The Voting Rights Act barely appears in its voting guidance.

Across the federal government the anti-humanitarian DOGE operation, which Musk funded and ran, closed civil rights offices at USAID, the Social Security Administration, the Department of Education, and agencies throughout the executive branch. The latest estimates are at least 500,000 children have died as a result of DOGE. And then, in terms of this one Tesla tragedy in Texas, the agency that should pull Saint Amour’s telemetry and formally count this crash among ADAS failures — NHTSA — now operates under an administration run by the CEO of the company it regulates.

Waymo launched driverless commercial service in Houston last month using LiDAR-equipped vehicles. It stands in direct comparison to the Saint Amour lawsuit made in front of a Harris County jury. It should be about a system built to read the actual road ahead versus one built by a man with a documented, multigenerational commitment to only looking backwards and rigidly refusing to adapt.

Attorney Bob Hilliard’s statement lands exactly where Tesla’s own documents already sit:

This company wants drivers to believe and trust their life on a lie: that the vehicle can self-drive and that it can do so safely. It can’t, and it doesn’t.

Tesla’s engineers called out the failure. It has a name: mode confusion. They thought they were describing a car, but it’s a man in denial of history.