Hey {{first_name | there}},

OpenAI and Anthropic have a rule: 

You cannot use their model outputs to train a competing AI.

But at the same time OpenAI and Anthropic trained their models on essentially the entire internet, your blog posts, your forum replies, your creative work, without asking.

This week, the CEO of Microsoft finally said it out loud.

Satya Nadella introduced what he's calling the 'Reverse Information Paradox'.

The concept is this: when enterprises use AI tools, plugging in their proprietary data, their workflows, their internal knowledge, that information potentially flows into the model's ongoing learning. 

The company pays for the subscription. And then the AI lab gets smarter from the company's own secrets.

Enterprise clients are, in a very real sense, paying twice. Once in money. Once in knowledge.

But Nadella didn't stop at the enterprise angle. He took direct aim at the hypocrisy baked into how these labs operate. They train on everyone else's data, scraped freely under a broad interpretation of 'fair use' and then prohibit anyone from doing the same thing back to them through model distillation.

Distillation, for context, is the process of using a large, powerful model's outputs to train a smaller, more efficient one. DeepSeek famously did this with OpenAI outputs earlier this year, and OpenAI screamed about it being a violation of their terms of service.

You can't claim fair use for yourself and deny the same logic to everyone else. 

That's not principle. That's monopoly protection dressed as ethics.

If Nadella was talking about structural hypocrisy, Apple just filed a lawsuit about the concrete version of it.

Apple sued OpenAI in federal court in Northern California, alleging trade secret theft. 

The accusation isn't abstract. Apple claims that former employees left for OpenAI and took proprietary hardware information with them, including confidential details about Apple's manufacturing processes and supplier relationships.

OpenAI's chief hardware officer, Tang Yew Tan, is named in the suit. The allegation is that OpenAI used insider knowledge to approach Apple's manufacturing partners directly, leveraging confidential information it was never supposed to have.

This is no longer about scraping blog posts. This is allegedly deliberate, targeted acquisition of competitive intelligence through the movement of people.

The irony is almost too clean: the company that prohibits you from using its outputs to learn from... allegedly sent people in to learn from Apple.

Anthropics's Claude Fable 5 was supposed to move to a paid model this week. Instead, Anthropic quietly extended free access for subscribers until July 19, because OpenAI just dropped GPT-5.6 Sol with aggressive pricing, and the market reacted.

This is what an AI pricing war looks like from the inside: models that cost hundreds of millions to train get given away temporarily to retain subscribers, because the alternative is losing market share to a competitor doing the same thing.

The labs are simultaneously claiming they're building transformative, world-changing technology... and racing each other to the bottom on pricing to keep users from switching.

If you're building a business on top of any of these platforms, you need to sit with a few uncomfortable questions:

1. Are you feeding your competitive advantage into someone else's model? Nadella's paradox is real. Every proprietary workflow you run through an external AI is a data point for that AI's continued learning.

2. What happens to your stack if the pricing war ends? Free or cheap access to frontier models is a promotional strategy, not a permanent market condition. The pricing will normalize upward. Build your unit economics accordingly.

3. Whose rules are you playing by? The labs ban distillation but train on the open web. They prohibit competitive use of outputs but allegedly use insider knowledge to compete with partners. The rules in this space are written by the powerful, for the powerful.

The most interesting companies I'm watching aren't the ones betting hardest on OpenAI or Anthropic. They're the ones reducing dependency, building proprietary datasets, running open-source models locally, creating moats the labs can't just absorb.

Something to think about.

- Aashish

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free