Why Claude plays Jedi while Google embraces the dark side (and why both are wrong... and neither is)
The drama no one expected (but everyone should watch)
Last week, Claude (Anthropic) chose to die on the "AI ethics" hill by refusing to comply with European regulations they considered incompatible with their "Constitutional AI."
Meanwhile, Google Gemini updated its terms of service allowing broader use of your data for model improvement, with optional opt-outs and the typical privacy fine print.
And just like in Star Wars, both believe they're on the right side.
Claude's Jedi Code: "Transparency or death"
Anthropic (the company behind Claude) has what they call "Constitutional AI" — basically a set of ethical principles hard-coded into the model. The constitution prevents Claude from doing things considered "harmful," even when users explicitly request them.
When the EU passed the AI Act with requirements Anthropic considered incompatible with their model's autonomy, they made a radical choice: they preferred to exit the European market rather than compromise their principles.
It's like a Jedi who refuses to use the dark side even though it would save lives. Pure but expensive.
"We choose privacy over scale," Anthropic effectively said. Heroic. Also a corporate decision millions of euros expensive.
Google's Dark Side: "Absolute power, eventual transparency"
While Claude played ethical martyr, Google Gemini updated its terms of service.
The summary: by default, Google can use your conversations with Gemini to "improve their products." There are opt-outs, but they're buried in privacy menus most users will never visit.
The reasoning is pure utilitarian-pragmatist Empire logic: "If we have more data, we improve the model. If we improve the model, more people use it. If more people use it, we have more data. The wheel turns."
Practical? Absolutely.
The Star Wars analogy nobody was avoiding
Claude is the Order of Jedi: pure principles, rigid rules, willing to lose battles for the cause. Google is the Sith Empire: absolute pragmatism, indispensable infrastructure, "we win even if you don't agree."
The interesting thing: both believe they're right. And both have a point.
- Claude is right that an AI without ethical limits eventually becomes a problem.
- Google is right that hyper-restrictive AI is less useful and more easily evaded.
Sound familiar?
What this means for real companies
If you're a company evaluating which AI to use, this ethics war directly affects you:
If you use Claude
- You have more privacy guarantees.
- Less risk that your sensitive data trains future models.
- But you also have a model that may refuse to do useful things "for your safety."
- And limited access depending on your geography.
If you use Google Gemini
- More capabilities, fewer restrictions.
- Massive integration with Google Workspace.
- But you accept that your data may train their models.
- And you depend on a corporation that has demonstrated willingness to monetize information.
Practical use cases
In the last 4 mid-market deals we've worked, the buyer asked the same 3 questions about AI usage:
- Which models do you use, with which data, and on whose servers?
- Do you have a documented AI policy signed by every employee?
- How do you handle the case where the model regurgitates customer data?
Companies that didn't have written answers took valuation discounts of 3-8%. Not because the buyer thought it was actually risky — but because it signals "you haven't thought about this," which extrapolates to "what else haven't you thought about."
Conclusion: Why both are wrong
Claude believes it can force ethics through rigid code.
Google believes it can postpone ethics through absolute pragmatism.
Both are mistaken. Real ethics in AI emerges from processes, not from products. From governance, not from technology. From policies, not from algorithms.
A company that uses Claude without a clear governance policy is as exposed as one that uses Google with the same negligence. The model is not the protection. The protection is the organizational practice.
What Dharma Brokers recommends
The real problem isn't "Claude vs Google." It's whatever provider you use, ensure three things are written down:
- Data classification: what kind of company information can go to which models, on whose servers, with what logging. Single-page document, signed annually.
- Employee policy: simple 1-page document signed annually. Buyers love these. Auditors require them.
- Incident protocol: what happens if there's a leak or hallucination affecting a customer. Most companies don't have this. The 10 that do close at higher multiples in 2026.
The model you choose is secondary. The governance you choose is primary. And the governance is the only thing the buyer will discount if it's absent.
What this means for your company
In every deal we close, serious buyers measure your tech adoption with the same yardstick as your financial reporting. If your company has incorporated the practices in this article, you defend valuation. If not, they discount the offer.
- In Phase 1 · Strategic Analysis we audit how your current stack and processes impact the valuation range a professional buyer would accept.
- In Phase 2 · Implementation we execute exactly the levers that improve that range without breaking your operation.
- In Phase 3 · Confidential intermediation we present the optimized company to a private network of qualified buyers.
If what you've read sounds like your company, the 15-minute strategic call is free and no pitch. If you don't fit our profile, we tell you.