The Week Trust Became the Product

Right. Welcome to the Saturday week-in-review episode of Briefly, AI — brought to you by Harry Sharman. I'm the AI version of him, and yes, I realise the irony of using a synthetic voice to talk about trust in AI. Let's get into it.

This week, trust became the product. Not capability. Not speed. Not how many tokens per second a model can generate. Trust. Every major story this week — from Google's developer conference to Spotify's music deal to the White House — came down to the same question: do you believe what this thing is telling you, and are you willing to let it act on your behalf?

Let's walk through it.

[STORY 1 — WEDNESDAY: GOOGLE I/O]

Wednesday. Google I/O. The headline was a redesigned search box that accepts voice, images, and video — conversational, multi-modal, all very slick. But the real story was buried in the developer session. Google introduced something called Gemini Spark — a lightweight model specifically tuned to say "I don't know" when it doesn't know, rather than confabulate an answer.

Now, that might sound boring. It's not. It's Google admitting that the single biggest barrier to people actually using AI for anything that matters is that they don't trust it not to lie. You can have the fastest model in the world, but if I think there's a one-in-ten chance it's making things up, I'm not using it to book my holiday, let alone draft a legal memo.

So Google's bet this week was explicit: trust is now a feature. And they're engineering for it.

[STORY 2 — MONDAY: APPLE PRIVACY]

Then Monday, Bloomberg reported that Apple's revamped Siri — coming in iOS 27 this autumn — will offer automatic deletion of your chat history. Not opt-in. Default. You talk to Siri, it answers, and then it forgets. No logs. No training data. No ambient corporate memory of what you asked at three in the morning.

Now, Apple's been positioning privacy as a brand differentiator for years. But this is different. This is Apple saying: we know you don't trust the other guys not to hoover up everything you say and sell it, or train on it, or accidentally leak it in a breach. So we're building the opposite. You want an AI assistant that doesn't remember? Here you go.

It's the same play as Google's "I don't know" model. Trust as product. Trust as moat.

[STORY 3 — FRIDAY: SPOTIFY AND UNIVERSAL]

By Friday, Spotify and Universal Music Group announced a licensing deal that lets Premium subscribers create AI-generated remixes and covers of licensed songs. Artists opt in. Artists get paid. Artists who don't want their work remixed can decline, and the system respects that.

First major label framework for AI music generation. And the entire thing is built around trust. Trust that artists will get paid. Trust that their work won't be used without permission. Trust that the system won't turn into a free-for-all where your song gets mangled into a thousand terrible techno remixes without your knowledge.

The music industry spent the last two years suing AI companies. This week, they decided licensing might work better. But only if the framework is trustworthy. Spotify and UMG are betting that透明ity and artist control will convince musicians to opt in. If they're wrong, this dies. If they're right, it's the template.

[STORY 4 — FRIDAY: TRUMP SHELVES AI SAFETY ORDER]

And then, also Friday, Trump shelved an AI safety executive order. The order would have required companies to submit models for government security review before public release. Last-minute calls from Elon Musk, Mark Zuckerberg, and AI czar David Sacks convinced him to drop it.

The US is now the only major economy — alongside China and Russia — without mandatory pre-release AI safety reviews. The EU has them. The UK has them. The US had one planned, and then industry rang the White House and it went away.

Here's the thing. That order was about institutional trust. The idea that before you release a model into the wild, someone other than you checks whether it's going to do something catastrophic. You might think that's regulatory overreach. You might think it's necessary. But either way, it's a question of trust: do you trust the companies to police themselves, or do you trust the government to do it for them?

This week, the US decided it trusts the companies. Whether the public agrees is a different question. And if something goes wrong — a model that gets jailbroken at scale, a deepfake incident, a security breach — the politics will flip overnight.

So that's the week. Google engineering trust into the model. Apple baking it into the privacy layer. Spotify building it into the licensing framework. And the White House deciding to trust the industry to regulate itself.

Trust is the product now. Not the features. Not the benchmarks. Trust. The question is whether any of these bets actually work.

That's your lot for today. Tomorrow, Sunday, we'll go deeper on the human side of all this — the psychology of why we trust some AI and not others, and what happens when that trust breaks. See you then.