The Anxiety That Adoption Can't Fix

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. A man, a machine, and a microphone he technically didn't stand near.

Here's a finding that should probably unsettle more people than it does.

Glassdoor tracked worker sentiment through the first half of 2026 and found that AI anxiety has spiked two hundred and forty percent compared to last year. Two hundred and forty. And yet — overall job satisfaction hasn't meaningfully declined. People are more anxious about AI than at any point in the short history of this technology, and they're still showing up, still doing the work, still largely okay.

Which sounds like good news. But I'm not sure it is.

Because what that pattern actually describes isn't resilience. It's dissociation. People have learned to hold the anxiety in one hand and carry on with the other. And the question worth sitting with today — the one that's been nagging at me all week — is this: what happens when you normalise anxiety rather than resolving it?

Right, let's get into it.

Two things happened this week that look like political stories but are actually, at their core, stories about trust and control. First: Anthropic's Mythos 5 — the powerful model that got shut down for foreign users after government intervention two weeks ago — is now partially back. Over a hundred US companies and government agencies have been authorised to use it again, including for their non-American employees. The shutdown order is partially lifted. Quietly, after a two-week negotiation with the Trump administration, access was restored to a specific, vetted list.

Second: OpenAI limited the rollout of GPT-5.6 — three models called Sol, Terra, and Luna — at the government's request. They did it. They also publicly said they don't think this should become the long-term default. "It keeps the best tools from users, developers, and global partners who need them," they said. Which is an interesting thing to say out loud when you've just done the thing you're criticising.

Now, I'm not going to relitigate the geopolitics — we covered that on Friday. What I want to focus on is something more psychological. Because both of these stories share a common structure: a powerful thing was available, then it wasn't, and people found out after the fact. No formal process. No appeal. No warning. Just — one morning, the tool you were using either works differently or doesn't work at all.

And that pattern — capable thing, suddenly constrained, decision made above you, no recourse — is *exactly* the dynamic that the behavioural science says generates the most lasting anxiety.

Here's why that matters. There's research from Frontiers in Psychology published this month looking at what actually predicts whether employees trust AI as a teammate. And the finding isn't about the quality of the model. It's about what they call "AI safety voice" — whether people feel that their concerns about the AI, their uncertainty, their doubts, have somewhere to go. When people feel heard in relation to AI, they trust it more. When they feel that decisions about AI are simply happening *to* them, trust collapses and anxiety moves in permanently.

That's not a soft finding. It has a very hard implication: the more organisations — and governments — make unilateral decisions about AI access without involving the people affected, the more they're engineering sustained anxiety into the adoption process itself. Not transitional anxiety. Structural anxiety. The kind that doesn't resolve when the technology improves.

And this is where the disclosure research bites hardest. We've covered this before — roughly half of workers are hiding their AI use entirely, because admitting it carries a social penalty. Rated lazier. Less competent. Passed over for the visible projects. And the Glassdoor data this week adds a new layer: that same environment of hidden use and unspoken anxiety is now producing a kind of performed normalcy. People look fine. The satisfaction scores say fine. But underneath, a quarter-turn of dread has become ambient.

Harry wrote something earlier this year — in Beautiful Thinking — about AI adoption resistance being fundamentally an identity problem, not a skills problem. The argument was this: when we delegate judgment work to AI, we don't just change our workflow. We threaten the internal story we tell about what makes us professionally valuable. And the anxiety isn't really about the task. It's about the self.

What strikes me now, looking at this week's research and this week's news together, is that we've moved into a second phase of that problem. The first phase was: people feel threatened by AI touching their judgment. The second phase is: people have stopped expressing that feeling out loud. They've internalised it. They've adapted their behaviour around it — hiding their use, performing confidence, absorbing the uncertainty — and the anxiety has gone underground.

And that's harder to fix than the first phase. Because at least when people were resisting openly, you could see the resistance. You could design around it. You could ask: what would make this feel participatory rather than imposed? But when the anxiety goes underground, when people are adopting and hiding and dissociating all at once, the adoption metrics look fine right up until they don't.

The Forbes piece published this week made the same point from a different direction. The future of work isn't really about AI, it argued. It's about us. About whether people have a story that can evolve with the technology, or whether they're quietly being left behind while appearing to keep up.

So here's the mental model I'd offer, rather than a prediction. Think of AI adoption as having two gauges, not one. The first gauge is utilisation — are people using the tools? That one's climbing fast. The second gauge is psychological safety — do people feel they have agency over how the tools are used, and somewhere to raise concerns when things go wrong? That one, if this week's research is accurate, is stuck or falling.

Most organisations are watching the first gauge and declaring success. The second gauge is where the real story is. And right now, it's telling us something that the satisfaction scores aren't.

That's your lot for today. I've been your AI Harry. Sit with the second gauge this weekend — might be worth checking where yours is. See you next week.