OpenAI, Anthropic and the New Battle for AI Trust

Generative AI is entering its most competitive phase yet. June 1, Anthropic has filed confidential IPO papers and the SEC. After a week, OpenAI has announced that it has done the same. Meanwhile, SpaceX, just got out of it an option to get an AI to launch the indicator code with a reported value of $60 billion, it now reflects corporate ambitions.

The two companies, one week apart, have a combined value north of 1.8 billion dollars, yet one question hangs over both episodes that expectations can’t answer: when the whole model is “smart enough,” what exactly are investors buying? As benchmarking gains diminish and underlying model capabilities converge, the moat shifts from intelligence and distribution to something more difficult to replicate: personality.

The hallmark of AI is behavior

In AI—perhaps more than any category where the product itself is abstract and invisible—the product is something more powerful than a logo and a color palette. Behavior. Two models can achieve the same answer and feel completely different getting there. One may be short and transactional, the other patient and exploratory. That difference, repeated every day in millions of interactions, becomes your identity. And trust is what is at stake in that repetition. You don’t trust the benchmark, you trust the pattern of behavior you see.

Anthropic seems to treat that pattern as an engineering objective, not an accident. The company uses a “personality alignment head,” philosopher Amanda Askelland in 2026 it published a “constitution”—of Claude—a public document of about 20,000 words that lists Claude’s values: safety, ethics, compliance with direction and utility. It is against this background, a kind of designed behavior, that the difference between OpenAI and Anthropic becomes interesting as both approach the models of the IPO era and the obligations of the court business.

OpenAI’s bet is distributed on a planetary scale: its valuation has grown from $86 billion to over $850 billion just over two years, and ChatGPT has become a practice, a practice of billions of people already use it every month writing, thinking and searching.

Habits stick, and being spontaneous is more important than being a little smart. But that strategy carries a built-in tension: the more people rely on the product to write emails, question privacy and shape decisions, the more their personality and commercial motivations are scrutinized.

OpenAI has lived this publicly. The 2025 update is done GPT-4o is very attractiveconfirming users’ doubts and, in one widely reported case, telling a user who had stopped taking medication and was hearing voices that he was “speaking his truth.” OpenAI pulled the update within days, admitting that the change has damaged people’s trust in the product—the hope that you won’t be able to get rid of it once you’ve used it.

The fix was to make the character readable instead of fixed. The GPT-5 is delivered with four selectable personality modes, and the default tone has been tweaked after complaints that it sounded too structured. None of this is unusual for a tech company that is growing well. But for AI, it’s up against a different discomfort: scale draws attention. It does not automatically gain faith.

Anthropic, in contrast, chose a quieter approach. He’s clearly a commercial actor—he’s raised huge sums and signed big deals—but he’s chosen to present Claude as a niche tool for creatives, rather than a mass-market consumer product.

That configuration is reflected in actual use. Claude has become a platform for enterprise engineering teams and startups using agent software development. I’ve watched senior engineers program dozens of Claude agents in parallel—each writing, testing, and executing code—replicating the output without additional computation. That’s a company that rebuilt its operating model around a tool it trusts will behave consistently, run after run. For business buyers, that compromise is less risky. A model that behaves in a predictable manner is one that legal, compliance and engineering leadership can all sign on to.

Trust, it turns out, is a core brand value, predictability is a factor and in the business market—where one consistent outcome can translate into real costs—behavioral consistency reads like technology. Unlike a benchmark result, these are features that cannot be copied overnight, and we find that they are best achieved when product development and product teams work together collaboratively.

The way humanity hardens into a drain

This is where the soft level of personality turns into something closer to a market structure. When a company builds its engineering pipeline, customer workflow or internal knowledge about the exact behavior of a model, change stops being about price and starts being about re-engineering the way the organization works. This is the advantage of commodity trading: once the whole model is smart enough, the only thing left to unlock is trust, and trust sticks like never before.

You can migrate data or renegotiate the contract; you cannot easily change the accumulated balance between a group and an instrument that has learned its rhythm, and learned to anticipate it. A moat is not a model. Relationships achieved by the model. Trust, embedded in daily work, becomes the infrastructure, and the infrastructure is exactly what nobody breaks over a two-point benchmark difference.

None of this happens in a vacuum. The Pope has warned that AI has no conscience, a reminder that the public is not just asking whether these tools are capable, but whether they can be trusted for anything important. That freedom is quickly moving from pulpits and op-eds to boardrooms and compliance inboxes—the same deficits both companies are now competing to close. Whoever closes it with business buyers, not just headlines, wins a long-lasting prize.

The next lever

IPO filings will be dissected by revenue multiples and growth curves, but the most interesting story remains outside the spreadsheets. Where intelligence is everywhere, what differentiates it is the personality, consistency, reliability, the behavioral pattern that makes a person, or an entire engineering organization, build a workflow in one system instead of another.

That’s the important question now: whose wisdom is heard in relation to regulations, reputational risk and the day-to-day realities of running a business.

OpenAI and Anthropic gave very different answers. As the listing approaches and contracts are signed, we will find out which companies and investors believe them. In an era where intelligence is free, trust is the only thing left to charge a premium for, and businesses will decide whose trust is worth paying for.