Washington’s DeepSeek order doesn’t stop at the app. It follows the model into other vendors’ products, which makes it a supply-chain problem more than a ban. Elsewhere today, a chatbot’s share button looks like a leak, and a single power event shows how much sits in one cloud location.
1. Meta’s Muse: a ‘share’ button that apparently means ‘findable’
Ars Technica’s summary line is short and bad: a single query is all it takes to find thousands of Muse conversations, listed by title. Ars’s full article was not available to us, so what follows rests on that line and the headline.
The headline says users’ personal and business secrets are exposed when they share conversations. That points to a familiar product failure. People hear “share” as “send this to one person.” The system apparently treats it as “make this retrievable by anyone who asks the right way.” Titles alone can give a lot away, and users often paste far more sensitive material into the chat itself.
I can’t say from the excerpt how the conversations are found, what they contain, or what Meta plans to do. Those details decide how serious this is. But the pattern is clear enough to act on: if an assistant’s share feature creates a link, treat that link as published.
2. The DeepSeek order covers anything built on DeepSeek, too
The order is broader than “no DeepSeek on government laptops.” According to TechCrunch, citing DHS, federal civilian agencies must stop buying DeepSeek’s products and services and remove them from IT systems. That also covers third-party products built using DeepSeek technology, though agencies can request exceptions.
DHS’s stated worry is data: sensitive information from US government users could be accessible to Chinese authorities under Chinese law. A December 2025 law had already targeted DeepSeek in intelligence national security systems. This extends the restriction to civilian agencies, for new purchases and existing systems alike.
My read is that the derivative clause is where the cost lands. DeepSeek is known for low-cost open-weight models, so it can end up inside other vendors’ software. Agencies, and the vendors selling to them, now need to know what is under the hood. The sources don’t say how that will be checked, and The Register’s coverage was not available to us beyond its headline. The data-access rationale also fits a hosted DeepSeek service more neatly than a product that merely uses its technology. The exception process may end up doing much of the real work.
The order lands alongside a September US report accusing DeepSeek and five other Chinese AI companies of distillation, and Beijing calls it an overstretched notion of national security.
3. UK regulator wants AI near-misses on the record
The Register’s headline says the UK finance watchdog wants firms to report AI failures and close calls. Only the headline was available to us, so details on scope, deadlines and whether this is binding are unknown.
Still, “close calls” is the notable phrase. Aviation and safety-critical industries collect near-misses because failures are rare and warning signs are cheap. Applying that logic to AI in finance suggests regulators want to see problems before customers do.
If so, the practical consequence is for firms that can’t currently tell a near-miss from normal noise. Reporting requires logs, monitoring and someone accountable for deciding what counts. I’d treat that as inference until the proposal’s details are clear.
4. A power event in London shows how much sits in one place
Google Cloud says a “power event” caused today’s outage, according to The Register’s headline, and a second Register headline says London zones were disrupted along with multiple services. The article text wasn’t available to us, so I won’t name affected companies or zones.
The cause is mundane, and that’s the point. Resilience plans often cover software bugs and bad deploys but depend on a physical building staying powered. A zone is supposed to be a failure boundary. Customers who put everything in one are betting on it.
The test for any team touched by this is whether failover has ever been exercised, or only drawn on a diagram. Google’s postmortem, if one appears, should show how far the disruption spread.
5. Bullhorn goes dark, and its customers wait for answers
The Register reports that a cyberattack knocked recruitment platform Bullhorn offline. The available text was only the headline, so we can’t say whether data was stolen, how long the outage lasted or who is behind it.
What the headline does show is the dependency. A platform that holds a staffing firm’s candidates, clients and workflow can stop that firm when it goes down, whether or not any data leaks.
The unanswered question is whether candidate or client data was taken. Until Bullhorn says, customers have to plan for both an outage and a possible breach.
