OpenAI, Anthropic, and the Illusion of Responsible AI

This week gave us two signals worth paying attention to.

The first: OpenAI publishes a paper on keeping people first ("Industrial Policy for the Intelligence Age: Ideas to Keep People First" - download link at the bottom). The second: Anthropic withholds its new model, Mythos, because it is supposedly too powerful to release.

At first glance, both stories are about responsibility. But if you zoom out, they reveal something else entirely.

Why We Chose Azure OpenAI Instead of OpenAI Direct

Let's start with something concrete.

We chose Azure OpenAI over OpenAI direct. Not for performance. Not for pricing.

For data residency.

OpenAI does not guarantee EU data residency. On top of that, OpenAI defaults to opt-in rather than opt-out for data usage controls — a choice that runs directly counter to the GDPR principle of privacy by default. For anyone building serious software for European companies, that is not a detail — it is foundational. And it raises an uncomfortable question: can a company credibly talk about responsible AI while leaving open where your data actually lives?

For us, the answer is straightforward: no.

This is where sovereign AI stops being theoretical and becomes a product decision.

• Where is your data stored?
• Under which jurisdiction?
• Who ultimately controls access?

These are not philosophical questions. They are the first questions any serious builder should be asking.

The Mythos Signal: Power or Positioning?

Now let's look at Anthropic.

They announced that their new model, Mythos, is so capable and so risky that it cannot be released publicly. Instead, access was shared with a small group of large organisations — big tech, financial institutions, and infrastructure players. The announcement triggered government meetings, cybersecurity discussions, and significant media attention.

But not everyone is convinced.

Several independent experts have pointed out that this may not represent a dramatic leap in capability — it may represent excellent strategic positioning. And that distinction matters enormously.

That said, I am not entirely convinced by the cynical reading. I genuinely believe Anthropic cares about the greater good, and I think Mythos is as dangerous as they claim. But here is the harder truth: if Anthropic has reached this level of capability, others will too. And they might not tell us — they will simply use it for their own ends.

Limiting access to a model does two things simultaneously. It signals power. And it creates exclusivity.

"This is too dangerous for the public" is structurally identical to: "This is too valuable for everyone."

We have seen this pattern before in finance, defense, and pharma. Now it is arriving in AI. And with it comes a new reality: AI is no longer just a product. It is becoming controlled infrastructure.

The Contradiction at the Heart of Responsible AI

Here is the paradox this week made visible.

OpenAI talks about democratising AI. Anthropic restricts access to elite players. Both claim to act responsibly. But responsibility seems to mean two very different things depending on who is speaking.

From a founder's perspective, the definition is simpler: responsible AI is not what you say. It is what you build into your product.

Data residency. Transparency about access. Honest answers about who benefits and who does not. These are the tests that matter — not the press releases.

The Human Premium: Pushing the Idea Further

There is a concept we believe in at a more fundamental level. Call it the human premium.

The logic is: AI removes routine work, so humans focus on higher-value work. That sounds clean. But let's push it.

If AI reliably removes low-value tasks, then people spend more time on meaningful work, companies build better products, quality increases, and — here is the counterintuitive part — prices go up.

We often hear that AI will make everything cheaper. That is only partially true. Routine tasks become cheaper, yes. But high-quality output becomes more valuable. Rather than a race to the bottom, we are more likely to see a shift toward premiumisation — better services, higher expectations, and eventually higher wages as value creation scales upward.

The Problem No One Wants to Talk About: The Transition Gap

This is where things get genuinely difficult.

This transition does not happen overnight. We are likely looking at a decade-long window in which some companies are AI-native and others are not, some workers adapt and others cannot, and the new high-value jobs that AI is supposed to unlock do not yet fully exist.

The benefits arrive later than the pain. That is where social instability starts.

This is also where governments actually matter — not in abstract policy papers, but in very practical ways: retraining programs, temporary income support, access to education, and real transition pathways for workers who find themselves between a job that disappeared and one that does not exist yet.

The Real Divide in AI Right Now

After this week, the divide becoming clearer is not OpenAI versus Anthropic. It is narrative versus reality.

The narrative says: responsible AI, democratisation, safety.
The reality shows: infrastructure control, data sovereignty gaps, selective access, and strategic positioning dressed as ethics.

What This Means If You Are Building

If you are building products in this space, here is the takeaway.

Do not get distracted by model hype, policy narratives, or "too powerful to release" stories. Focus on the things that are real: actual workflows, actual customers, and actual constraints — like where your data lives.

The companies that win the next decade will be the ones who treated those questions seriously from day one, long before it became a regulatory requirement.

---

The AI industry is entering a phase where the most important question is no longer how powerful the model is. It is who controls it, and for whom does it create value.

The sooner builders, founders, and executives internalise that shift, the better positioned they will be for what comes next.

---

Read the full OpenAI paper: Industrial Policy for the Intelligence Age: Ideas to Keep People First