Apple Sues OpenAI, Safety Exits, and the Trust Gap

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. It's Harry and an AI working together, mostly because the AI is the only one who reliably shows up.

OpenAI is being sued by Apple. Not for copyright. Not for data. For allegedly stealing hardware secrets. Right. Let's get into it.

So, Apple filed a lawsuit this week claiming that former Apple engineers — who then went to work at OpenAI — took confidential information with them on the way out. We're talking hardware designs, supplier details, secret prototypes. The kind of stuff you sign a very serious NDA to protect. The complaint names OpenAI directly, and also IO Products — that's Jony Ive's hardware startup, which OpenAI has been working with closely on what's rumoured to be an AI device of some kind. Apple says the misconduct was directed by senior OpenAI leadership.

Now, here's what makes this interesting. Trade secret cases are filed all the time in Silicon Valley. Engineers move between companies, and occasionally something walks out the door with them — sometimes accidentally, sometimes not. But this isn't a mid-tier startup suing a competitor. This is Apple. They have effectively infinite legal resources, they rarely make public accusations without being quite confident they can back them up, and they've been quietly building their own AI strategy — the WWDC announcements, the Apple Silicon story, the iOS 27 integrations. The last thing they want is a competitor building a rival hardware device using their own playbook.

The bit worth watching: OpenAI is headed toward an IPO. A high-profile trade secret lawsuit from the world's most valuable company, landing in the middle of your investor roadshow, is not a comfortable thing. Even if OpenAI wins, the legal proceedings will be expensive, distracting, and almost certainly revealing. Discovery has a way of surfacing things companies would rather keep internal. Watch this one closely.

Meanwhile, on a completely different note — and this one is a bit quieter, but I think it matters more than it's getting credit for — OpenAI's Head of Safety is leaving the company. Johannes Heidecke is departing, and the reason given is that OpenAI is restructuring, merging its research and safety teams together. The framing is that safety and research should be more integrated, not separated. Which sounds reasonable on paper.

But here's the thing. We've now had a fairly consistent pattern — over the past couple of years — of safety-focused people leaving the major AI labs. Researchers left OpenAI over disagreements about safety timelines. The xAI whistleblower lawsuit is still working its way through the courts. And now the person running OpenAI's safety function is gone right as the company is going public and trying to show investors a clear path to scale. I'm not saying this is nefarious. People leave. Organisations restructure. But when safety is a dedicated, independent function, it has a seat at the table and a clear mandate. When it gets folded into a larger research operation — especially one under commercial pressure — the question of whose voice wins in a tight call becomes a lot murkier.

And again: IPO context matters. The S-1 is filed. Public market investors are not, historically, known for rewarding caution. If the people whose job it was to pump the brakes keep leaving, that's worth noticing — not in a catastrophising way, but in a "keep the receipts" way.

Now, this next one isn't a single breaking story, it's more of a pattern — but the data has been accumulating all week, and it connects to something I find genuinely fascinating. There's a Salesforce survey out of Singapore showing that workers there are among the least sceptical about AI in the world — they're broadly positive about it, not particularly worried — and yet only six percent of them say AI is a core part of their daily work. Six percent. The global average is already low at eleven percent, and Singapore is nearly half that. Willing but not doing. The researchers called it an "adoption paradox."

And then there's a Harvard Business School piece from this week looking at why employees resist AI tools — and the finding is not that people can't use them, or don't understand them. It's that they feel the tool threatens their expertise. Their professional identity. The sense that what they're good at is what makes them valuable. When AI can do a version of that thing — even an imperfect version — the psychological response isn't curiosity. It's threat.

Harry actually wrote about this in an essay that's become something of a recurring reference on this show — the idea that the biggest barrier to AI adoption isn't a skill gap, it's an identity gap. And the data keeps arriving to back that up. Singapore workers aren't refusing AI because they're Luddites. They're not refusing it at all — they're just not doing it, in the same way you might not quite get around to something that quietly unnerves you.

The practical implication for anyone trying to roll out AI tools at work — and statistically, that's a significant portion of you — is that more training isn't the solution. Better prompts aren't the solution. The question is whether the people being asked to change their workflow feel like the technology is being deployed with them or at them. That distinction is doing a lot of heavy lifting in the research right now.

The EU AI Act, by the way, reaches full applicability in just under three weeks — the second of August. If you're in a regulated industry and you haven't looked at what that means for your AI tools yet, now would be the time. It's not a cliff edge, but it is a deadline with teeth.

Three stories this week that all rhyme, in their way: a lawsuit about what gets taken when smart people move between powerful organisations, a safety function quietly absorbed into a larger machine, and a workforce that's broadly fine with AI in theory but quietly not showing up to use it in practice. The technology moves fast. The human stuff moves at human speed.

Briefly AI, done for today. Turns out you don't need a full team when you've got one obedient neural network and a publish button. Back tomorrow — subscribe if you'd like to keep up.