They Were Speaking Different Languages — Why Great Technology Fails When It Doesn't Translate

#What happened over five days in June

Here's how Axios's June 2026 reporting laid it out. On June 9, Anthropic released its frontier models, Fable 5 and Mythos 5, to the public. There was no advance briefing to the administration. Three days later, on June 12, the U.S. Department of Commerce issued an export control order: block all foreign nationals, inside the U.S. or out, from accessing Fable 5 and Mythos 5. One day after that, on June 13, Anthropic broadly suspended access to both models to comply with the legal directive. The stated justification was jailbreak vulnerabilities and national security, but the bottom line was this: the company had to pull its own strongest, freshly launched product just four days after release.

This clash didn't erupt out of nowhere. Back in February of the same year, the Pentagon asked Anthropic to strip the guardrails off Claude so it could be used for autonomous lethal targeting and mass surveillance, and Anthropic refused. The defense contract fell apart. On March 9, the administration designated Anthropic a 'supply chain risk' (a designation Anthropic is contesting in court). The safety-first philosophy of the company and the administration's demands had been misaligned from the start, and the five days in June were where that misalignment finally landed.

#"They speak a different language than we do"

One assessment that came out of the administration's side is striking. It's admittedly one side's view, but here's what a source close to the administration told Axios.

The key phrase is 'speak in different languages.' Releasing a model without advance notice, flatly refusing a guardrail request, fighting a supply-chain-risk designation in court — from Anthropic's side, every one of these choices likely flowed from a single, consistent safety philosophy. But the other side didn't read it as philosophy. Looking at the exact same actions, one side called it 'principle,' and the other called it 'disregard.'

Another administration official put it more coldly still.

Here's the thing worth sitting with: it's hard to say Anthropic made the 'wrong' call every single time. Refusing to strip guardrails from a system that could be used for autonomous killing is, by most people's standards, the right judgment call. The issue wasn't right versus wrong — it was translation. The company was right in its own language, without ever accounting for how its choices would sound in the other side's.

#Great technology and righteous conviction don't automatically persuade anyone

Anthropic had frontier technology. Its safety-first philosophy was legitimate in its own right. It's hard to fault a company for refusing to enable autonomous killing and mass surveillance. And yet its strongest product still got pulled from the market.

Granted, this episode — tangled up with national security and export controls — operates at a different scale and with different stakes than a demo day or a government grant review. But the failure pattern rhymes. The company couldn't translate its own values into the other side's decision-making criteria.

The uncomfortable but clear question this raises for founders is this: does great technology persuade people on its own? Does a righteous conviction automatically win agreement? The five days in June answer no to both. Good technology and the right conviction are the raw materials of persuasion — they are not persuasion itself. If the other side can't understand that value in their own language, even the most brilliant asset gets translated back to you as 'risk' or 'disregard.'

#A founder's stubbornness deserves respect — but

Let's be clear so this isn't misread. A founder having their own stubborn conviction and a way of seeing the market nobody else sees isn't a flaw — it's an asset. The ability to see what others miss, and to push forward when everyone else says it can't be done: without that, you wouldn't have started the company in the first place. Just as Anthropic's safety philosophy was legitimate in its own right, a founder's conviction can be legitimate in its own right too.

What's dangerous isn't conviction — it's the complacency that tends to ride along beside it. Confidence and complacency look alike. Both start from the exact same place: 'I'm right.' That's precisely what makes them hard to tell apart. Conviction stops at 'I believe my judgment is correct.' Complacency adds one more line: 'so the other side will understand eventually, too.' That single extra line is the trigger that quietly costs founders the very opportunity that came their way.

In a founder's head, 'they'll understand eventually' can feel like self-respect. On a judge's desk, it just reads as an unhelpful document. If a judge can't understand it within five minutes, that business plan gets filed under 'not yet validated.' Not because the technology is bad — because it was never translated.

#Investors and judges speak a different language

Even when a founder and a reviewer are looking at the exact same business, what they ask, what they weight, and what they trust are different. Founders want to talk about how innovative this is. Reviewers want to know first whether it actually works, who has already paid for it, and why you're the one to build it. Same language on the surface, effectively a different one underneath.

The left column isn't wrong. It's just not yet translated into the reviewer's language. Translation means turning the sincerity on the left into an answer to the question on the right. It's not about abandoning your conviction — it's about carrying that same conviction over in a form the other side can actually receive.

Anthropic's case illustrates this exact point. The company pushed back, arguing that the jailbreak vulnerability in question was comparable to what's commonly found in competing models like GPT-5.5, and that applying this standard across the industry would halt every new model launch. Taken purely on technical merit, that's a reasonable rebuttal. Other issues — the fairness of the regulation, the political context — are tangled up in this conflict too, so it would be unfair to pin all the blame on Anthropic's communication alone. But one thing is clear: even a technically sound rebuttal failed to persuade, because the other side was listening by a different standard entirely.

#Persuasion isn't shouting louder — it's translating

Founders who fail to translate share a pattern. When the message isn't landing, they say it louder, longer. They just turn up the volume on their own language. But if the other side is speaking a different language, no amount of volume gets the meaning across. It's the same reason a better-built model didn't repair Anthropic's relationship with the administration.

Real persuasion runs in the opposite direction. It's not shouting louder in your own language — it's carrying the message over into the other person's. Failed translation tends to look the same every time: pages full of adjectives like 'groundbreaking' and 'overwhelming' with no verifiable numbers behind them, or answers only to the questions you feel like answering while the ones the reviewer actually cares about go untouched. Just as Anthropic kept talking about 'safety' while the administration kept asking about 'security,' both sides end up answering different questions entirely.

1. Distill your business's core value into a single sentence.
2. Rewrite that sentence as a direct answer to the question the reviewer actually wants answered (who has already paid you money).
3. Don't throw out your conviction and vision — place them after the evidence. Earn trust first, then talk about the dream.
4. Before you submit or present, check how that translation actually reads to the person receiving it.

One misconception needs addressing here. Translation isn't about changing the facts. This is absolutely not a license to invent traction that doesn't exist or inflate your numbers. It's about carrying the same facts over into a form the other side can accept as grounds for trust. The table below illustrates the difference (the figures are illustrative only, not real data from any specific case).

The left and right columns aren't the difference between a lie and the truth. They're two expressions of the same fact. It's just that the right side has numbers, a basis for comparison, and a path to verification. What a reviewer trusts isn't adjectives — it's verifiability.

Most founders read what they've written only in their own language before hitting submit. What's missing is the step of reading it in advance through the language of the person who will actually read it. So it's only once they're on stage — like Anthropic in June — that they realize, too late, that they'd been speaking a different language all along.

#Read it in the "reviewer's language" before you submit

The first step of translation isn't dramatic. It's this one thing: check in advance how your business plan reads in the reviewer's language. Not on stage, but before you ever get there — not after the pitch ends, but before you submit.

The problem is that, at that point, it's hard to find someone who will read your writing through a reviewer's eyes. The people around you are usually already fluent in your language, or hesitant to give you honest feedback. So right when you need a mirror the most, there isn't one.

That mirror is exactly what OpenSeed is trying to be. AI review agents built on real evaluation criteria and the question structures investors actually use read your business plan in the reviewer's language before you submit it. They show you, in advance, where something is left untranslated — which sentence was written as 'innovation' but reads as 'lacking evidence.' It's not a good/bad scoreboard. It's a tool that shows you exactly where your language got stuck on its way to becoming theirs.

Anthropic had the best technology in the world and still failed to translate it. Fortunately, your business plan can still be translated again before you ever step on stage.