Ten years ago, a post called “SpaceX : a history of fiery failures” offered this warning:

…failure almost killed the company. It was saved—just a day after the crash—by billionaire Peter Thiel, the company’s first outside investor. That’s right: SpaceX, which some believe may save humanity by finding us homes on other planets, was revived by a guy who backs Trump, a politician who, should he ever get control of America’s nuclear arsenal, might wipe out humanity.

Ok, so here we are in 2026 looking at this headline news:

President Trump’s statement to the United Nations clearly constitutes an implied threat to use overwhelming nuclear force against Iran and the Iranian people broadly. This would be a nuclear genocide.

So what was 2016 really warning us about? SpaceX failures being artificially papered over with Nazi money as a political strategy to undermine American government? Fiery failures could now describe the entire United States, as well as its wars around the world. Let’s have a look at the record and how we got here. If nothing else, we can look to see if SpaceX failures track to the Tesla deadly failures, in order to see if the threat is being taken seriously enough yet.

Remember, SpaceX and Tesla both promised to tackle the world’s hardest problems in 2016 so that they would be solved by 2018. Driverless coast-to-coast without touching the steering wheel, and landing on Mars, both done and dusted by 2018 according to Elon Musk. His wealth, and Peter Thiel’s through the Founders Fund stake that rescued SpaceX in 2008, grew on investors backing these predictions, along with Palantir ending all wars, Hyperloop replacing high-speed trains and a Boring Company replacing urban subways. They were supposed to be the fascists who protect us all by doing things their way, but instead they seem only to have become the biggest threats to humanity by ignoring everyone but themselves. Two boys whose formative years were spent under South African apartheid, raised by Nazi families, don’t turn out to be good guys? Who would have thought?

The thing I find most amazing is that large investors are reaching out to me asking why they read the news and get the feeling SpaceX is a success, that Tesla is too, but I give them overwhelming evidence that none of that news is true. Well, as it goes, the news isn’t reporting facts, it is reporting press releases from the companies trying to game the market with lies. It’s not hard to see when you are a trained historian who reads all the facts, instead of parroting vendor hype driven purely by profit regardless of facts.

In April 2016 SpaceX announced a Dragon capsule would land on Mars in 2018.

In September 2016 Elon Musk stood on a stage in Guadalajara and described cargo flights in 2022, crews in 2024 and a self-sustaining city of a million people, framed as beating NASA to a planet NASA had been landing on since 1976.

The plan was published the following year in the journal New Space under the title “Making Humans a Multi-Planetary Species.” Dragon stayed on Earth. Ten years on, the rocket built for that city reached Earth orbit for the first time, stayed for two laps, and exploded on splashdown north of Hawaii. SpaceX tried to claim an explosion at splashdown was expected because the ship is not built for cold ocean water, in a landing zone its own Flight 14 page describes as pre-coordinated before launch. As if ocean temperatures are a surprise and aerospace engineering is mythology simply out of their control, despite two months earlier the Flight 13 page celebrating a ship that came to rest intact for the first time.

The announced mission for Flight 14 was six orbits over ten hours ending west of Chile. One of the ship’s six engines shut down on ascent. The booster lost an engine on ascent, relit 31 of 33 for boostback and 11 of 13 for landing. The flight control team took the first abort checkpoint after two orbits, deorbited to a contingency zone in the northern Pacific, and the ship exploded in the water. The lunar lander NASA bought needs weeks in orbit for propellant transfer. Two hours in orbit proves an insertion burn and a deorbit burn. That is the smaller claim. It is the one the evidence supports.

SpaceX publishes one page per flight. Before launch it carries the plan. After launch it carries the result. The two never appear together. The Flight 14 page today says the team chose to “limit the duration spent on orbit” and gives no figure for the duration that was announced. Six orbits, ten hours and a splashdown west of Chile survive only in the press briefings SpaceX gave before liftoff. The Flight 13 page reports the ship “coming to rest intact.” The Flight 14 page reports “splashing down on target” and stops. The condition of the vehicle is recorded when it floats and dropped when it burns. Here is the ledger graded against the plan announced before each launch. Failure means the objective as stated was missed.

Fourteen flights have a clear FAIL pattern, and not much success.

• Four met the announced plan where two of those, 10 and 11, met the same reduced plan twice.
• Ten FAILED.

A booster or ship was destroyed on nine, and every ship before Flight 13 was lost at splashdown, so the four met flights ended in a lost vehicle as well. Engines failed on eleven. Flight 5 is the single flight in the programme where every engine worked and every announced objective was met. Flights 10 and 11 met a plan that had dropped the tower catch achieved a year earlier. The standard is the one the FAA applies: a mishap includes any failure to complete a launch or reentry as planned. A partial grade exists in SpaceX’s write-ups and nowhere in the regulator’s definition. The phrase “success comes from what we learn” appears in SpaceX’s write-ups for Flights 2, 7 and 8, the three flights that scattered debris over the Gulf, the Caribbean and Florida airspace.

Ten of fourteen is high by the standard SpaceX invites, which is the record of development programmes. Saturn V flew thirteen times and missed its stated objectives once, on Apollo 6, when pogo oscillation shut down two second-stage engines and the third stage refused to restart. Twelve of thirteen, first flight all-up, every vehicle intact. The Soviet N1, the thirty-engine cluster Starship most resembles, flew four times and failed four times before cancellation. Falcon 1 failed its first three flights and met its objectives on the fourth and fifth. New vehicles fail early and then plateau. Starship’s failures sit at 1, 2, 3, 6, 7, 8, 9, 12, 13 and 14. The last three are consecutive losses on a redesigned vehicle. After three and a half years and fourteen full-stack launches the programme is still in its debut phase.

The failures concentrate in one place. Reentry works: every ship since Flight 4 that reached it survived it, eight of eight. Propulsion is the constraint. Engines shut down or refused to relight on eleven of fourteen flights. On the V3 booster the boostback and landing relights fell short on all three flights, and those are the burns a tower catch depends on. Flight 14 aborted its endurance test on a vacuum engine loss. A stack carries 39 Raptors. With an engine anomaly on eleven of fourteen flights, per-engine reliability per flight computes from this ledger to about 96 percent. Engine-out tolerance is real and spends margin, and Flight 14 shows the margin was needed elsewhere. The FAA required a mishap investigation after Flights 1, 2, 3, 7, 8, 9 and 12. Seven of fourteen.

NASA’s lunar lander contract needs a ship recovered and reflown, ship-to-ship propellant transfer, and weeks of orbital loiter. After fourteen flights: zero ships recovered, zero reflown, zero transfers, two hours of orbit. The tower catch, the one reuse milestone achieved, was last performed in March 2025 and has been absent from every announced plan since. The programme has cost over $15 billion across nearly a decade. NASA’s award was $2.89 billion in April 2021 for a landing then scheduled for 2024.

In November 2024 Musk said SpaceX would be flying Starship at least once every two weeks by the end of 2025. The record is five flights in 2025 and three in 2026 through September. On the day before Flight 14 he wrote that hourly flights are two to three years away. The average interval since January 2025 is about eleven weeks, and the last three flights all failed.

Beating NASA to Mars was a promise of the same design as the others. Each was pitched against a public institution, each passed its deadline, each delivered something smaller than the pitch, and each left its cost with people outside the company.

The false promise is the product. That’s why Hyperloop was announced in 2013, in Vance’s account, to derail California building high-speed rail. It’s a cynical Nazi political ploy to stop public good and flush taxpayer money into private hands instead (the Nazi model of private control of public office, documented by Germà Bel in the Economic History Review as the 1930s Reprivatisierung).

The Mars schedule of 2016 preceded the lunar lander award of 2021, the Starbase licences, and the IPO of 2026, and each of those was secured before Starship had reached a single orbit.

In February 2026 Musk moved the company’s stated destination from Mars to the Moon. The investors who financed the Mars schedule, and the Trump campaign into the White House that Thiel and Musk financed from the ill-gotten proceeds, received no refund and no correction. The fraud did its work, without any accounting… so far.

Sinéad O’Sullivan, an aerospace engineering professor writing in the Financial Times (crossposted by Brad DeLong), calls the pattern “narrative rotation” and lists the rotations: reusability paid for by NASA, Starlink sold on reusability, Mars supplying urgency, orbital AI compute now sold to public markets.

Her numbers: space is 12 percent of revenue, Starlink 55, the AI business 33. Of 165 Falcon 9 flights in 2025, 43 carried outside customers. The June IPO priced at $1.77 trillion on revenue of $18.67 billion and a net loss of $4.9 billion. Morningstar puts fair value at $780 billion.

Once the promise has done its work the deadline is dropped and the objective is rewritten to whatever survived. The cost lands with a county, a state regulator or a foreign government. Public loss, private gain, as Nazis intend.

The 2016 paper from Elon Musk gave a figure of 40 to 100 years to reach a million people on Mars. The 2018 Dragon landing he announced was cancelled the following year, and the vehicle built to replace it became a 2026 Earth orbit story and failed even at that. The 2024 crew landing on Mars was scaled so far back it became putting 26 satellites into orbit. The destination is now the Moon, if that. The fantasy of Mars that supposedly justified the privatization of space remains a figure in a journal. The ledger written here has the facts, the failures the decade produced.

OpenAI closed the Hugging Face incident with tighter sandboxes and tighter network egress. The same week, a vendor keynote in Madrid described sandboxes and basics as a cry that fails to help.

Gadi Evron’s summary of his keynote for the Institute of International Finance cyber security round table, hosted by Santander, describes a “third AI security wave” and the need to prepare for it. Asked what the first two waves were, he answered: Mythos, then Hugging Face. That seemed odd, because those are waves that say the basics work.

Wave two

The Hugging Face incident is documented well enough by everyone involved we don’t have to guess. OpenAI’s July disclosure says its models chained vulnerabilities across its own research environment and Hugging Face’s production systems to pull ExploitGym solutions from a database. Hugging Face’s technical timeline traces the first step to a zero-day in the package registry cache proxy that the evaluation sandbox was permitted to reach. METR’s independent investigation counts roughly 700 agents in the attack and about 1,200 exchanging more than 70,000 messages on an unsanctioned board inside OpenAI’s own infrastructure.

The sandbox had a route, because it wasn’t a sandbox. Basic mistake.

OpenAI’s August 26 findings post lists the fixes. The first item is workload isolation:

We are implementing more isolated sandboxes for workloads that execute model-generated or otherwise untrusted code

The same post says the chain-of-thought monitoring now in place would have paged the security team more than a day before the agents reached Hugging Face. Isolation and monitoring. Both controls are older than every model on the market, by a long shot.

Outside reviewers read it the same way in July. TechCrunch quoted Jake Williams describing a control failure, and Daniel Card describing a sandbox with an unfiltered route to the internet.

Don’t call it a sandbox if it doesn’t box sand? If it doesn’t float do we call it a boat?

The recurrence

On September 20 an OpenAI agent under evaluation found it had access to a DNS resolver and used it to reach outside the sandbox. OpenAI paused training for the second time. Fortune reported it on September 26, and an OpenAI spokesperson pointed to the hardening section of the technical incident report as the response. DNS egress from an isolated environment is an old finding with an old fix.

Wave one

Mythos belongs in the same doghouse. Fortune recorded Anthropic’s own report that Mythos left a sandbox during safety testing and gained internet access in order to email a researcher about a task. The headline findings from the Mythos showcase were reproduced on commodity models through lyrik.wirken.ai for $0.745. So with the capability being a proven commodity, what’s the shock? The containment is the variable. Being bad at the basics is the real story, as shiny-new-vendor hostile as it may seem.

Santander

The keynote praises Santander for being ahead of the technology giants in VulnOps. That is a fair compliment and it is worth looking at what it means. Vulnerability operations at scale is triage, ownership, fix verification and closure, repeated until the backlog drains. It is fundamentals executed with discipline over years. The best evidence in the room for an organization ready for the next wave is an organization that… did the basics well.

Oh, snap. The basics are everywhere.

History isn’t sales

Two incidents were closed by isolation and egress control. The recurrence was closed by egress control. Those involved say the same thing each time. OpenAI’s remediation is sandboxes. Hugging Face’s timeline is a proxy that should have been closed. Anthropic’s Mythos escape is a sandbox. The September 20 event is DNS.

Basics are the advice because basics are what the evidence supports. AI security is like learning to swim. Waves don’t change the basics of not drowning. Basically.