Good morning. One thread ties today's stories together: control — in two senses. Who controls AI as an asset (a possible US government stake in OpenAI, a key policy advisor's exit) and how AI itself is controlled (OpenAI's new data-exfiltration guardrail, Apple's gatekept Siri reboot, NVIDIA's push to move AI onto your own machine). Read OpenAI's Lockdown Mode page for the security angle and TechCrunch's reporting for the ownership one. Prefer this once a week? Subscribe to the weekly brief.
1. OpenAI ships Lockdown Mode to blunt prompt-injection data theft
OpenAI introduced Lockdown Mode, a setting that reduces the risk of data exfiltration from prompt-injection attacks — where malicious instructions are hidden in webpages, files, or other content a model reads. As TechCrunch reported, the mode disables live web browsing (you can still reach cached content), the retrieval and display of images from the web (you can still generate images), deep research, and agent mode. OpenAI is candid that it is not a cure: even with Lockdown Mode on, ChatGPT "could still be vulnerable" to injections that "appear in cached web content or in an uploaded file." The company says it's "not intended for everyone" — it's aimed at people and organizations handling sensitive data — and is rolling out to self-serve ChatGPT Business accounts and eligible personal accounts.
The substantive read is that prompt injection has graduated from a research curiosity to a shipped-product threat model. Turning off browsing, agents, and deep research is effectively trading capability for a smaller attack surface — an admission that the most useful agentic features are also the riskiest. It's a blunt instrument, but a defensible default for anyone whose chats touch confidential material.
Why it matters. If you handle client data, legal files, or anything regulated in ChatGPT, this is a real control worth enabling — but treat it as risk reduction, not immunity. What to watch: whether rivals ship comparable lockdown toggles, and whether OpenAI can later restore agent and browsing features under the protection rather than simply switching them off. The broader lesson lands beyond ChatGPT: an assistant is only as safe as the rest of your stack — our sister site's security reviews cover the tools that lock the rest of it down.
2. The Trump administration weighs an equity stake in OpenAI
President Trump said Friday he's been talking with AI companies about deals "where the American people can benefit from the success of AI," and TechCrunch reports OpenAI is a likely target: CNBC said the administration has been discussing an equity stake with the company. Some of that equity could seed a "Public Wealth Fund" that OpenAI itself proposed, with proceeds "distributed directly to citizens." The idea echoes the government's 10% stake in Intel last year — and, from the other side of the aisle, Senator Bernie Sanders this week floated a one-time 50% tax that labs like OpenAI, Anthropic, and xAI would pay in stock.
The substantive read is that government co-ownership of frontier labs is moving from think-tank hypothetical to live negotiation, with support on both the populist right and left. The critics aren't quiet about it: investor David Sacks warned it would "accelerate the corporate-government fusion we're already sliding toward," and others read the groundwork as a possible bailout pathway. Either way, the question of who owns AI is now a Washington question, not just a Sand Hill Road one.
Why it matters. Nothing here is investment or policy advice. But if you track AI markets, a government stake changes the incentive math for everyone — procurement, regulation, and competition all look different when the state is also a shareholder. What to watch: whether any deal is confirmed before the wave of 2026 AI IPOs prices in, and what governance strings come attached.
3. White House AI advisor Sriram Krishnan steps down
Sriram Krishnan, the Trump administration's senior policy advisor on AI, is leaving at the end of June, TechCrunch reported. A former product leader at Microsoft, Twitter, Yahoo, Facebook, and Snap and most recently an Andreessen Horowitz partner, Krishnan was among the Silicon Valley figures who took roles in the second Trump administration. He cited the administration's AI Action Plan — which prioritized data-center buildout over regulation — as a key accomplishment, and named David Sacks as his closest collaborator. Per the Washington Post, he plans to start an outside institution that will keep him involved in shaping Trump's AI policy.
The substantive read is continuity dressed as a departure: the person leaves government but stays in the policy orbit, now from a private perch with fewer disclosure constraints. Paired with today's equity-stake story, it sketches an AI-policy apparatus in flux right as the biggest ownership questions are being decided.
Why it matters. If you follow AI regulation, watch where Krishnan lands — outside institutions with direct lines to the White House can move policy as much as the org chart does. For builders, the throughline is that the rulemaking that affects your roadmap is being written by a small, mobile group of people, not a stable agency.
4. WWDC 2026 opens Monday with a Gemini-powered Siri
Apple's Worldwide Developers Conference kicks off Monday at 10 a.m. PT, and the headline expectation is a long-delayed Siri overhaul into a conversational, multi-step assistant — reportedly built on Google's Gemini. Per TechCrunch's roundup of leaks (Bloomberg, The Information, MacRumors), Apple may also ship a standalone Siri app to rival ChatGPT and Claude, an AI-agent integration with the App Store, a "Visual Intelligence" camera mode, natural-language photo editing, and Wallet bill-splitting. Treat all of this as pre-event reporting, not confirmed product — the keynote is the source of record.
The substantive read is that Apple's "AI" story is increasingly an integration story: the most-hyped feature reportedly runs on a competitor's model. That's a pragmatic admission that Apple is behind on frontier models and would rather rent capability than wait — but it also hands Google a foothold inside the iPhone's most personal surface.
Why it matters. If you build for Apple platforms, Monday sets your AI roadmap for the year — agent APIs and an App Store for agents would be the developer story to watch. For everyone else, a Gemini-powered Siri is the clearest sign yet that the assistant wars are being fought through partnerships, not just in-house labs. For how these assistants differ from autonomous agents, our agents vs assistants explainer lays out the line.
5. NVIDIA pushes RTX Spark for personal-AI PCs
NVIDIA is promoting RTX Spark, the chip it unveiled at GTC Taipei and pitches as a way to "reinvent Windows PCs for the era of personal AI agents." Per NVIDIA's blog, the part is aimed at local AI, content creation, and gaming in slim laptops and small desktops — AAA games at 1440p and 100-plus fps with ray tracing, DLSS, and Reflex — and NVIDIA says 100-plus Windows software and game developers, including KRAFTON, NC, Riot Games, NetEase, Remedy, and Xbox, are on board. CEO Jensen Huang demoed it this week at Korean PC bangs alongside the T1 esports team.
The substantive read is that "personal AI" is becoming a hardware pitch, not just a cloud one. Running agents and models locally trades the cloud's raw scale for privacy, latency, and no per-token bill — the same logic behind today's Lockdown Mode story, approached from the silicon side. The marketing wrapper is gaming, but the strategic target is on-device AI workloads.
Why it matters. If you're weighing where your AI runs, local hardware is getting credible for inference that you'd rather not send to a server — a real option for sensitive or offline work. The performance and partner figures here are NVIDIA's own; the launch-partner and frame-rate claims are worth independent benchmarking once retail units ship.
What to take from today
Five stories, one idea: control. Two are about who owns and governs AI — a possible US stake in OpenAI and the exit of a top AI advisor — and three are about how AI is controlled in practice: OpenAI's data-exfiltration guardrail, Apple's gatekept Siri, and NVIDIA's move to put AI on your own machine. The smart move is unchanged: verify on your own terms — read the help page, the filing, and the keynote, not the summaries of them.
Tomorrow's brief lands at 15:30 UTC. If you'd rather read this in your inbox once a week — just the stories that actually matter — subscribe here.