Apple Sues OpenAI for Allegedly Stealing Hardware Secrets
This one has some real teeth: Apple alleges a pattern of trade secret theft by former Apple engineers who went to OpenAI — and the complaint also drags in Jony Ive’s IO Products, which is building OpenAI’s hardware play. According to Apple, these weren’t accidental overshares — they allege senior OpenAI leadership directed employees to bring over confidential presentations, prototype specs, and supplier details. If the allegations hold up, this isn’t just a lawsuit; it’s a fundamental challenge to OpenAI’s entire hardware ambitions before they’ve shipped a single device.
OpenAI’s Head of Safety Is Leaving the Company
Johannes Heidecke is out as OpenAI’s head of safety, departing as the company moves to further integrate its research and safety teams — which, depending on your level of cynicism, either sounds like smart organizational design or like safety getting quietly absorbed into the product machine. This follows a string of high-profile safety-team departures over the past couple of years, and it lands in the same week OpenAI is getting sued for allegedly directing employees to steal IP. Not a great week for the “trust us, we care about doing this right” narrative.
Meta Turns Off the Instagram Feature That Let Users Make AI Deepfakes of Public Accounts
Meta announced a feature, the internet revolted, and Meta killed the feature — in what may be the fastest product lifecycle in recent social media memory. The AI Muse tool let anyone generate images based on any public Instagram account just by tagging it, with zero permission required from the account owner. “We’ve heard the feedback that this feature missed the mark,” Meta said, which is corporate for “we genuinely did not anticipate that people would not enjoy being turned into AI content without their consent.” Wild stuff.
Instagram’s Adam Mosseri: If You Don’t Like AI, ‘Then You Shouldn’t Have It in Your Feed’
Mosseri is not filtering AI content off Instagram, but he’s very graciously offering to let you filter it out yourself — which is a neat philosophical trick where the burden of curation shifts entirely to the user while the platform keeps its engagement numbers. “I don’t think we should filter out AI content” is a complete sentence with a complete meaning. Paired with the deepfake feature debacle above, it paints a pretty clear picture of where Meta’s priorities are parked.
SK Hynix Raises $26.5B in the Biggest Foreign IPO in US History
The AI chip boom just set a Wall Street record: SK Hynix — the memory giant powering Nvidia’s HBM stack — pulled off the largest foreign IPO in US history and is now being nudged to build domestic fabs. This is the AI infrastructure buildout made flesh in dollar form, and it underscores something important: the real money in AI right now isn’t chatbots, it’s the memory and silicon underneath them. When the chip suppliers are raising $26.5 billion in a single listing, you understand the scale of what’s being built.
Would You Host Part of an AI Data Center in Your Home?
Solar company Sunrun wants to turn your spare room into a micro data center — pay you for the privilege, run compute through nodes they install in your home, and call it “distributed AI infrastructure.” It’s either a genuinely clever answer to the land and power constraints strangling big data center buildouts, or it’s the beginning of a beautiful arrangement where your house is slowly outsourced to the cloud. I’ll reserve judgment until I see the terms of service on what happens when a node catches fire.
AI Fiction Is Easy to Detect Because It’s Stupid and Bad, Research Finds
New research finds that AI-generated fiction is highly detectable — not because of watermarks or metadata, but because ChatGPT leans heavily on dream sequences and Gemini can’t stop describing what characters look like. As a piece of literary criticism, “your prose is bad and you should feel bad” isn’t exactly a technical paper, but as a practical finding it’s useful: the tells are narrative and structural, not just stylistic. The models apparently haven’t learned that restraint is a virtue.
Claude Code Costs Up to $200 a Month. Goose Does the Same Thing for Free.
Block’s open-source Goose is emerging as a serious free alternative to Claude Code’s $20–$200/month AI coding agent — and the developer community is noticing. This is the open-source pressure valve doing exactly what it’s supposed to: when a premium tool gets expensive, someone builds and releases a comparable one that undercuts it to zero. Anthropic is betting that deep integration and polish justify the price; Goose is betting that developers are cheap and stubborn. History suggests the latter is not a bad bet.
Bottom Line
OpenAI is getting sued for stealing secrets, losing its safety chief, and watching its partners kill features under public pressure — being the most powerful AI company in the world apparently comes with a lot of fires to put out.