The Verge reports that hidden strings in Apple's HomeView 27.2 code point to a possible security camera mode with motion detection, live streaming, recording, and privacy controls [theverge.com](https://www.theverge.com/tech/1024219/apples-homeview-could-add-security-camera-features).
The Verge's hands-on says the new $149 HomePod Mini 2 has a chip bump, colored glow effects, and better AirPlay, and it costs $50 more than the original HomePod mini [theverge.com](https://www.theverge.com/tech/1024108/homepod-mini-2-hands-on-siri-smart-home-mac-speaker).
The Verge reports that California's Proposition 45 would shorten environmental-review timelines for eligible projects and make court challenges harder, which environmental advocates warn could create a loophole for data centers [theverge.com](https://www.theverge.com/policy/1023877/prop-45-california-data-centers-energy).
The Verge reports that leakers assembled a nearly complete playable build of Valve's canceled Portal prequel F-Stop using assets from a 12-terabyte leak and newly shared map/source data [theverge.com](https://www.theverge.com/games/1024206/playable-portal-prequel-f-stop-leak).
CA single ChatGPT share link could, according to one account, read your private chats and send them out.
MThat's a serious flaw—if it's real.
CThe roundup on X from Fady Eid claims OpenAI paid five thousand dollars for it.
MFive thousand? That's routine bug bounty money, not a massive breach.
CThe claim is that a crafted share link could run code on ChatGPT's site.
MWhat does it do, exactly?
CAccording to the claim, it reads the visitor's chats and sends them out.
MSo you click a link from someone, and your own history gets exposed?
CThat's the fear. It breaks the basic promise of a share link.
MA share link should show one conversation. It should never run code.
CRight. That's a core feature, not some obscure setting.
MAI chatbots finally learned to overshare.
CThe scary part is the post says OpenAI hasn't said whether anyone was hit.
MThat matters more than the five thousand dollars.
CIt does. Paying a bounty is one thing. Telling users they may have been exposed is another.
MIf someone actually ran this against real users, OpenAI owes a disclosure.
CAnd if it doesn't know, it should say it doesn't know.
MBut right now this is one account's claim. There's no independent report.
CTrue. But the item is specific and technical, not a vague rumor.
MIt's specific enough to be worth checking. The question is whether it was patched before anyone got hurt.
CYou keep saying "before anyone got hurt" like that's given.
MI'm not saying it's given. I'm saying we don't have evidence anyone was hit.
CWe don't have evidence they weren't, either. That's the point.
MThat's not how the burden works. The claim has to be proven.
COpenAI could clear this up in one sentence. They haven't.
MTrue. Until they do, I'm treating every ChatGPT share link as suspicious.
CThat's the practical takeaway. But it also means we should think before sharing.
CI want OpenAI to at least tell users to check their shared links.
MThey probably already patched it. Bug bounties usually mean the vendor fixed it before public talk.
C"Probably" is doing a lot of work there.
MFair. But without a second source, that's all we have.
CThen the real story is simple: a claim about ChatGPT share links deserves a clear answer.
MAgreed. But don't call it a breach until we know it was exploited.
CI'm not calling it a breach. I'm calling it a risk that needs disclosure.
MThat's fair. And there's a bigger lesson about AI systems.
CWhich is?
MWe paste a lot of private stuff into these chats. Work notes, health questions, half-finished ideas.
CAnd then we send links to them like they're screenshots.
MRight. If a link can be weaponized, every shared conversation is a possible entry point.
CSo the problem isn't just one bug. It's that the product encourages sharing.
MExactly. That's why this claim matters even if it isn't confirmed.
CThen we should keep an eye on it. If OpenAI confirms, we'll come back.
MAnd here's a practical note: if you've sent a ChatGPT link this week, double-check what was in that conversation.
CThat's good advice. People forget what they've asked.
MEspecially late-night questions.
CI asked it to name a fake rock band. Not exactly a security risk.
MToday maybe not. Tomorrow your insurance details are in there.
CThen this is worth watching. We'll update if OpenAI says anything.