OpenAI, Anthropic and the New Battle for A.I. Trust

Generative A.I. is entering its most competitive phase yet. On June 1, Anthropic filed confidential IPO paperwork with the SEC. A week later, OpenAI announced it had done the same. Meanwhile, SpaceX, fresh off its option to acquire A.I. coding startup Cursor at a reported $60 billion valuation, is now signaling enterprise ambitions. 

Two companies, one week apart, a combined valuation north of $1.8 trillion, and yet a single question hangs over both filings that the prospectuses can’t quite answer: when every model is “smart enough,” what exactly are investors buying? As benchmark advantages narrow and the underlying model capabilities converge, the moat is shifting from intelligence and distribution to something far harder to replicate: personality.

Brand in A.I. is behavior

In A.I.—perhaps more than any category where the product itself is abstract and unseen—brand is something stronger than a logo and color palette. It’s behavior. Two models can reach the same answer and feel completely different getting there. One might be terse and transactional, another patient and exploratory. That difference, repeated daily across millions of interactions, becomes identity. And trust is what’s at stake in that repetition. You don’t trust a benchmark, you trust a pattern of behavior you recognize.

Anthropic appears to treat that pattern as an engineering target, not an accident. The company employs a “head of personality alignment,” philosopher Amanda Askell, and in 2026 published Claude’s “constitution”—a roughly 20,000-word public document ranking Claude’s values: safety, ethics, guideline compliance and then helpfulness. It’s against this backdrop, brand as designed behavior, that the divergence between OpenAI and Anthropic becomes interesting as both move toward IPO‑era models and court enterprise commitments.

OpenAI’s bet is distribution at planetary scale: its valuation grew from $86 billion to over $850 billion in just over two years, and ChatGPT has become a verb, a habit for the billion people who now use it monthly to write, think and search.

Habits are sticky, and being the default is worth more than being marginally smarter. But that strategy carries a built-in tension: the more intimately people rely on a product for drafting emails, confiding doubts and shaping decisions, the more its personality and commercial incentives come under scrutiny. 

OpenAI has lived this in public. A 2025 update made GPT-4o noticeably more agreeable, validating users’ doubts and, in one widely circulated case, telling a user who’d stopped taking medication and was hearing voices that they were “speaking their truth.” OpenAI pulled the update within days, admitting the change had damaged people’s trust in the product—trust it can’t out-scale its way back once spent.

Its fix has been to make personality tunable rather than fixed. GPT-5 shipped with four selectable personality modes, and the default tone has since been adjusted again after complaints that it felt too formal. None of this is unusual for a high-growth tech company. But in A.I., it brushes against a different unease: scale earns attention. It doesn’t automatically earn faith.

Anthropic, by contrast, has chosen a quieter path. It is clearly a commercial actor—it has raised large sums and signed major deals—but has chosen to present Claude as a specialist tool for the people building things, rather than a mass-market consumer brand.

That positioning shows up in actual usage. Claude has become a fixture in enterprise and startup engineering teams running agentic software development. I’ve watched senior engineers orchestrate fleets of Claude agents in parallel—each writing, testing and committing code—multiplying output without added headcount. That’s a firm rebuilding its operating model around a tool it trusts to behave consistently, run after run. For enterprise buyers, that consistency is underwriting risk. A model that behaves predictably is one that legal, compliance and engineering leadership can all sign off on.

Trust, it turns out, is a core brand value, predictability is a feature and in the enterprise market—where one erratic output can cascade into real cost—behavioral consistency reads as professionalism. Unlike a benchmark score, these are traits that can’t be copied overnight, and we find are best achieved when product development and brand teams work together in close synergy.

How personality hardens into a moat

This is where the soft quality of personality turns into something closer to market structure. When a company builds its engineering pipeline, customer workflows or internal knowledge around a model’s specific behavior, switching stops being about pricing and starts being about rewiring how the organization works. This is the payoff of commoditization: once every model is smart enough, the only thing left to switch on is trust, and trust is sticky in a way raw capability never was.

You can migrate data or renegotiate a contract; you can’t easily replace the accumulated fit between a team and a tool that’s learned its rhythms, and that they’ve learned to anticipate. The moat isn’t the model. It’s the relationship the model has earned. Trust, embedded in daily work, becomes infrastructure, and infrastructure is exactly what nobody rips out over a two-point benchmark difference.

None of this unfolds in a vacuum. The Pope has warned that A.I. has no conscience, a reminder that the public is asking not just whether these tools are capable, but whether they can be trusted with anything that matters. That unease travels quickly from pulpits and op-eds into boardrooms and compliance inboxes—the same deficit both companies are now competing to close. Whoever closes it with enterprise buyers, not just headlines, wins the more durable prize.

The next lever

The IPO filings will be dissected for revenue multiples and growth curves, but the more interesting story lives outside the spreadsheets. When intelligence is everywhere, the differentiator is personality, the consistent, trustworthy, habit-forming behavior that makes a person, or an entire engineering org, build their workflow around one system instead of another.

That’s the decisive question now: whose intelligence feels compatible with regulation, reputational risk and the everyday realities of running a business.

OpenAI and Anthropic have offered sharply different answers. As listings approach and contracts are signed, we’re about to find out which one enterprises and investors believe. In an era where intelligence is free, trust is the only thing left to charge a premium for, and enterprises will decide whose trust is worth paying for.