You Never Actually Trusted It, Did You

Today on Briefly AI — new reporting suggests the more AI gets forced into your life, the less you actually like it. OpenAI and Anthropic are now racing each other to prove who can be trusted with your data. And the security researchers OpenAI once welcomed to go hunting for dangerous flaws have just been shown the door.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. Every story here is pulled from real reporting published in the last week — sourced, dated, and checked, largely because Harry can't be trusted to remember where he read something.

Okay — let's get into it.

You know that friend who insists they're completely fine with the surprise party, then spends the whole evening looking faintly betrayed in the corner? That, according to this week's reporting, is roughly where humanity's at with AI.

Right, so TechCrunch had a piece this week that's worth starting with, because it quietly undoes a story the industry's been telling itself for two years. The theory was simple: get AI into enough hands, often enough, and comfort follows usage — the way nobody thinks twice about a search engine anymore. Instead, the opposite's happening. As AI gets stitched into search results, email, phones, apps you didn't ask it to be in, people are growing warier, not fonder. Familiarity isn't breeding affection. It's breeding a sort of low-grade suspicion.

Here's the bit that matters if you work anywhere near a product team: usage numbers were never the real scoreboard. You can hit a hundred percent adoption and still have zero percent goodwill, because people were never given a choice — the AI just showed up one day, uninvited, rearranging the furniture. Which, when you say it like that, sounds a lot like most surprise parties.

Meanwhile, on the corporate end of that same trust problem — OpenAI has apparently decided the way to win you over isn't a smarter model, it's a stricter lock on your filing cabinet. TechCrunch reports the company's rolling out new privacy protections for enterprise customers, specifically pitched to outdo Anthropic, who's spent the last while quietly building a reputation as the more careful, more buttoned-up option of the two.

Which is a genuinely odd place for this rivalry to land. For a couple of years the entire pitch was speed and cleverness — who's got the smartest model, who can reason furthest, who ships first. Now two of the best-funded labs on the planet are competing over who's more boring with your data. That's not nothing. If you're a hospital, a law firm, a bank, "who handles our records most carefully" was always going to matter more than "who writes the wittiest poem." It's taken a while, but the market's finally pricing in the thing everyone actually cared about.

Now this next one's a bit more uncomfortable, because it cuts against everything OpenAI's been saying about itself lately. TechCrunch reports that independent security researchers have had their access abruptly pulled from OpenAI's Trusted Access for Cyber programme — the scheme that let vetted, good-faith hackers use OpenAI's sharper models to find and report vulnerabilities before criminals did. No real warning, work left mid-stream, researchers understandably annoyed.

The timing's the strange part. The Verge reported just days earlier that OpenAI is deliberately slowing down — tightening security, adding a couple of weeks' pause to some development, all in the name of caution, even with an IPO looming and Anthropic and a wave of cheap Chinese and open-weight models breathing down its neck. So in the same fortnight, the story is both "we're being more careful" and "we've cut off the people whose entire job was finding our mistakes before someone worse did." Those two things are supposed to point the same direction. Right now they don't.

Whether that's a policy tidy-up that got handled clumsily, or an actual retreat from letting outsiders poke at the system, is going to say a lot more about OpenAI's real priorities than any safety statement will.

So there we have it — trust going down, privacy suddenly fashionable, and the locksmiths not entirely sure they're still allowed in the building. Honestly, if AI companies are this bad at PR, it's a wonder any of us trust the models to write our emails.

That's Briefly AI for today. Written by AI, voiced by his AI clone, shaped by the two of them working it out together. Subscribe wherever you listen, and we'll be back tomorrow.