Apple just made romance a regulated App Store feature, and AI companions are the category most exposed. Elsewhere, memory costs are squeezing even Apple, while Harvard’s year-old breach and a 2028 AI-ransomware warning show how long security problems linger.

1. Harvard’s breach reached the people who never enrolled

The Register reports that ShinyHunters hit 850,000 alumni, donors, students and family members in last year’s raid on Harvard. The word that matters there is family. A university’s database is not only its student body, but a wide circle of people connected to it, most of whom never chose to hand over data.

The Register page we have is mostly a headline, so we can’t say what data was taken, when Harvard learned of it, or when it told anyone. Any claim about a disclosure delay would be guesswork. What the headline does support is the scale, and the fact that it is about last year’s incident.

For founders, the practical lesson is about retention. Every record you keep for a donor, alumnus or relative is a record that a breach can turn into someone else’s problem, possibly a year later.

The Register

2. AI ransomware has a date, and patching is the bottleneck

The Register’s headline says AI-powered ransomware attacks will arrive in Britain by 2028. The article text we have is only that headline, so we can’t say who made the forecast or on what evidence. Treat the date as a claim, not a measurement.

Cloudflare’s Zero Wire launch, which is context rather than confirmation, describes the mechanism people worry about. Cloudflare says AI can analyze large codebases at scale and surface deeply nested flaws that went unfound for years. It says that will put more pressure on security teams to assess each disclosure.

That suggests the scarce resource is not finding holes but fixing them. Cloudflare says so itself: a WAF rule is a compensating control, not a replacement for patching vulnerable software. If discovery speeds up while patching stays slow, defenders fall further behind however good the tooling is.

The Register · Cloudflare Blog

3. Apple turns romance into a regulated App Store feature

Starting November 16, any app offering dating or romantic relationship features must carry an 18+ rating, and developers must add new content descriptors so their ratings match what the app does. Dating apps mostly restrict themselves to adults already. TechCrunch says the interesting part is AI companions.

Character.AI, Replika and similar apps let users create friends, lovers and fantasy characters, and TechCrunch notes they are often aimed at younger people. Top companion apps across the App Store and Google Play had been downloaded 220 million times as of last summer. Because the rule covers apps where romantic features are only one of many, a general chatbot with a romance mode could be pulled into the 18+ bucket.

The cost falls on companion developers who relied on a young audience, and Apple controls the rating. The open question is how broadly Apple reads ‘romantic content’ once apps start filling in the new descriptors.

TechCrunch

4. Raspberry Pi raises prices again as memory costs bite

The Register reports that Raspberry Pi has raised prices again, barely a month after the last round. The text we have is only the headline, so we can’t confirm the amounts or the stated cause.

Apple’s earnings give a sense of the pressure on memory. CEO John Ternus called memory pricing ‘unprecedented’, said Apple expects even higher costs in the December quarter, and expects them to keep rising after that. Its inventory is up 154% year over year, and its manufacturing purchase obligations rose by $33 billion from June to September.

Apple is a different company with far more purchasing power than a single-board computer maker, so this is context, not proof of why Pi prices moved. Still, if Apple is stockpiling memory and warning of more increases, smaller buyers have less cushion. Anyone budgeting a hardware product should assume component prices can change between quotes.

The Register · TechCrunch

5. An AI editing another AI’s weights deserves a hard look

The Register’s headline says AI can now tweak another AI’s weights, with the pitch being that you can forget retraining. The source text we have is only that headline, so the method and its results remain unverified here.

The appeal for a small team is obvious: adjusting a model without a full retraining run could save time and compute. But it moves the risk rather than removing it. When a model’s weights are edited by another model, you need to know what changed and whether the improvement holds on your own task rather than on a benchmark.

Until those answers are available, treat it as a technique to test, not a replacement for evaluation.

The Register