🧪 The Most Tested Brand in AI

Hey, I’m Isaac 👋 I founded Pistachio, a growth agency working with B2B brands like Atono and Clay to build trust, relationships and loyalty with their current and future customers.

If you want to chat about your own growth strategy, book a free call with me here.

Read time: 4 mins 48 secs

Interested in sponsoring these emails? See our partnership options here.

On 9 June, Anthropic released the two most powerful AI models it had ever built. By 12 June, both were gone, pulled from every customer in the world after the US government issued an export-control order. It was the first time a leading AI company had taken a publicly deployed model offline at the government's order.

For almost any company, losing your flagship product three days after launch is a disaster. For Anthropic, it landed differently. One more public test of the thing its whole brand is built on, and one more test it held.

There's a real question buried in that, though. Is it conviction, or the most convenient brand positioning in tech?

This is the gap Anthropic has spent two years walking into. Its bet was that a brand built on a real belief earns a kind of loyalty that features never will. Wise proved the same thing in banking, building an $11 billion business by standing for transparency while everyone else competed on rates. People commit to what a company stands for far harder than they commit to what it sells.

A stated belief and a proven one are worth very different amounts. Anyone can write the charter. The real question is what happens when holding the position starts to cost serious money, and whether the company holds it anyway. That is the test Anthropic keeps being handed.

The first real test came from the Pentagon. It wanted to use Claude for fully autonomous weapons and mass domestic surveillance, and it wanted the safety limits that prevented both removed. Anthropic declined, with Dario Amodei saying the company "cannot in good conscience accede to their request." Defense Secretary Pete Hegseth designated Anthropic a supply chain risk, a classification the government normally reserves for hostile foreign suppliers. Treasury, State and other agencies cancelled contracts. Anthropic filed federal lawsuits over the designation. OpenAI went the other way entirely, taking its own $200 million Pentagon contract and building out "OpenAI for Government."

Read that as a business story and Anthropic lost. It walked away from one of the largest customers on earth and watched a competitor lean into the work it had refused.

Read it as a brand story and something else happened. Until that moment, Anthropic's safety positioning could be dismissed as the same alignment talk every lab recites. Refusing the Pentagon at a cost of hundreds of millions turned the claim into a fact. And this is an audience that rallies when Anthropic picks a fight it believes in. Around the same time, its Super Bowl ads mocking ads in ChatGPT had sent Claude from number 41 into the App Store's top 10, with paid subscriptions more than doubling. People show up for a brand that draws the lines they care about.

For a company whose entire brand is safety, a public war with the Pentagon over safety is the best piece of marketing it could ever run. No campaign buys that credibility. Which makes it fair to ask whether Anthropic is a company of unusual conviction, or one that has worked out that conviction, demonstrated at the right moment, is worth more than any ad. The honest answer is probably both, and the fact that we can't cleanly separate the two is the whole game.

February's cost was chosen. Anthropic decided to walk. The Fable shutdown was a harder test, because this time the cost was imposed.

On 9 June, Anthropic launched Mythos 5 and Fable 5. Fable was the public version, the same model wrapped in extra guardrails to stop it being used to find and exploit software vulnerabilities. Three days later a researcher published a way around those guardrails, and the government ordered both models pulled for every user worldwide. Anthropic complied. It also disagreed in writing, arguing the standard being applied, if used across the industry, "would essentially halt all new model deployments."

The statement is where the craft shows. Anthropic surfaced it inside the app, to the exact users who had just hit the wall. The message apologised for the disruption, explained the disagreement on technical grounds, and named no villain. It spoke to two audiences at once. Users got the apology up top. Regulators got the rebuttal underneath. There was no call to action, no request for anyone to take a side.

That restraint is a choice, and a disciplined one. A louder brand would have rallied its base. Anthropic let the facts and the audience do that work on their own. It read as the composure of a company sure of its position, which is its own kind of marketing. The act of pulling the model was loud enough. The words about it could afford to be calm.

A brand this committed attracts a particular kind of customer, and that customer is hard to keep happy. The Fable shutdown made it obvious. The same community that rallied in February split this time. A loud share of developers drew the opposite lesson from the outage. A model a government can switch off is a model you don't really own, and the only safe option is open weights you run yourself. The episode that proved Anthropic's values exposed a genuine product weakness, and the believers were the first to say so.

That is the bill that comes with conviction-based loyalty. People who buy into what you stand for hold you to it, and they punish drift faster than a casual user ever would. Anthropic is already testing that patience in smaller ways. Over recent months it reduced Claude's default effort level to save compute, tightened usage limits during peak hours, and moved enterprise billing to per-token pricing, and users clocked the pattern before the company explained it. So far these are ordinary cost-cutting, a long way from the safety values at the core of the brand. They still matter, because they are the first signs of a company feeling commercial pressure, and Anthropic has staked everything on being the one that doesn't bend under it.

The trust-compounds story skips this part. A reputation for never blinking is the most valuable asset Anthropic owns and the most fragile. It holds only as long as the company never visibly trades a principle for a quarter. The first time it does, on something that actually touches the values, the structure that took two years to build comes down at once. For all the trust it has banked, Anthropic holds the most exposed brand position in AI.

Strip away the politics and the lesson for everyone else is uncomfortable. Claiming a value is free, which is why every brand does it. Proving one costs contracts, products, and standing, which is why almost none of them do. Anthropic has paid every invoice it has been handed, and in a market where every competitor recites the same promises, those receipts are the only thing customers actually believe.

That is also why it is close to impossible to copy. You can't manufacture the test, and you can't fake the payment. Pick a fight you don't mean and the audience smells it. Hold a line only until it gets expensive and you have proven the opposite of what you intended. Anthropic's position works because the cost is real and it keeps paying. The model will come back online. The harder question is whether Anthropic can keep meaning it as the bills get larger, because the brand it has built leaves it no room to ever stop.

If you enjoyed this post or know someone who may find it useful, please share it with them and encourage them to subscribe: brandchemistry.co/p/the-most-tested-brand-in-ai