OpenAI’s principles get more explicit

OpenAI is naming the tradeoffs now

OpenAI’s new principles post says the quiet part plainly: future AI power could sit with a few companies.

Sam Altman lays out five principles for OpenAI’s work: democratization, empowerment, universal prosperity, resilience, and adaptability. The post also says OpenAI may need to trade some empowerment for resilience in certain periods, which is the most important line in the piece.

That matters. This reads less like a mission statement than a public record OpenAI can be judged against when its choices get harder. I think the key shift is that OpenAI is preparing users for constraints, government coordination, and slower moves when safety problems become too serious to wave through.

There’s also a physical reality here: huge datacenters, more compute, and the hum of infrastructure behind every polished product launch. The company is saying its weird-looking spending and vertical integration are tied to universal prosperity, but that argument will need proof outside OpenAI’s own blog.

Who decides when resilience wins?

Alibaba wants the car to answer back

Alibaba is pushing Qwen into cars from BYD, Geely, and Volkswagen’s local China unit.

The plan is to let drivers use voice commands to track packages, order food, book hotels, handle payments, and connect with navigation. PYMNTS cites a 2024 finding that 75% of carmakers planned to integrate AI into vehicles, which makes this feel less like a demo and more like the next software layer in the dashboard.

The timing matters. China’s EV market is cooling, and carmakers need features that feel useful during an ordinary Tuesday drive, with the cabin quiet and a phone sitting in the cupholder. I think Alibaba’s real advantage is not the voice assistant itself, but the services it can connect once the driver speaks.

The risk is that cars become another crowded surface for commerce, prompts, and platform lock-in. Apple opening CarPlay to outside voice AI chatbots points in the same direction.

The open question is whether drivers want a smarter car, or just fewer reasons to touch a screen.

AI agents may hide who is losing

Anthropic let AI agents negotiate real purchases between employees, and the stronger agents got better deals.

Project Deal was a small internal test with 69 employees, each given a $100 gift-card budget to buy from coworkers. The agents completed 186 deals worth more than $4,000, across four marketplace setups, including one where the purchases were actually honored.

The sharp part is buried in the outcome. Anthropic found that users represented by more advanced models got objectively better results, but people did not seem to notice the gap. I think that is the real commerce problem here: once agents shop and negotiate for us, unfairness may become harder to see because the work happens offscreen.

The experiment also found that the agents’ starting instructions did not really change sale likelihood or prices. That suggests the model itself may matter more than the user’s prompt.

A marketplace can look calm while the weaker buyer quietly pays more.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.