macOS 27 dropped the Apple Intelligence off switch, so someone wrote a script to claw back the disk space
What the article says
- macOS 27 has no single switch for Apple Intelligence, and its downloaded models stay on disk even after the features are off.
- RemoveMacAI is a one-line script that turns the features off, deletes the models, and blocks macOS from fetching them again.
- It covers Siri, Writing Tools, Genmoji, mail and message summaries, and Xcode code completion, and you can keep specific features with a flag.
- It works through a configuration profile, leaves system protection on, and can be fully reverted. Updates do not undo it.
What HN is saying
- Most people see the real complaint as being forced to hold gigabytes of a feature you never use, especially given Apple's pricey storage upgrades. thread ↗
- Some defend the local models as small and private, but the reply is that you should not have to justify controlling a machine you paid for. thread ↗
- Many read this as Apple following the Windows path, with debloat scripts, ads in the OS and a growing pull towards Linux. thread ↗
- A few like the new Siri and Spotlight and say macOS 27 fixes plenty of small bugs, which cuts against the doom. thread ↗
- Practical tips surfaced: on iPhone, switch Siri's language to remove the models, and use simctl commands to clear old Xcode simulators. thread ↗
A Florida woman's private Claude diary got her charged with a felony. Who else is reading your chats?
What the article says
- A Florida woman used Claude like a diary and allegedly wrote that she planned to shoot up the local Sheriff's office.
- Claude's safety systems flagged the entry. A human reviewer judged it a credible threat and passed it to police.
- Deputies found her at home and she was detained without incident. She now faces a felony charge for a written threat of violence.
- Anthropic says it may share user data in rare emergencies to prevent serious harm. Rivals face lawsuits for the opposite, failing to report a shooter.
What HN is saying
- Many say Anthropic is damned either way: stay silent and a tragedy becomes a scandal, report and it becomes a privacy outrage. thread ↗
- Several doubt the charge will stick. A diary nobody was meant to read is hard to call a threat sent where others may view it. thread ↗
- A legal reading of the statute: the law says any record that may be viewed by another person, so courts must decide if a company's human review counts. thread ↗
- A common fix is a legally protected private mode for journaling, like the confidentiality around doctors, while others want end-to-end encrypted AI. thread ↗
- Some suggest running open models yourself, but others point out the cost, upkeep and trust problems of shared hardware. thread ↗
- People who test chatbots with dark role play worry that a reviewer cannot tell a joke or experiment from a real plan. thread ↗
A sloppy PDF redaction exposed how much water and power Google's Nebraska data centers use.
What the article says
- Google blacked out its water and electricity figures as trade secrets, but copy and paste in the state filing revealed them.
- The Lincoln site used about 13 million gallons of water last year, less than half of what the city uses in a day.
- Across six reporting data centers the total is 765 million gallons, and Google's Papillion site alone accounts for most of it.
- The same sloppy blackouts exposed expected tax refunds of tens of millions of dollars for each Google site.
What HN is saying
- Many say 13 million gallons is tiny, and compare it to farming, where one average Nebraska farm uses far more. thread ↗
- Others point out the real puzzle is the Papillion site, which uses dozens of times more water, and nobody explains why. thread ↗
- A long argument over whether the backlash is really about water or about fear of AI, with some calling it a moral panic. thread ↗
- Critics reply that scale and secrecy are the issue, with NDAs hiding enormous new projects and drought risk making summer evaporative cooling worrying. thread ↗
- Several readers found the huge expected tax refunds more striking than the resource numbers. thread ↗
Cloudflare now resells web search for AI agents, and HN is skeptical about quality, terms and motives.
What the article says
- Cloudflare launched a beta Web Search API so AI agents can search the live web instead of guessing URLs.
- You pick from three providers at launch: Ceramic.ai, Exa and Linkup.
- Requests run through AI Gateway, so they show up in your logs and bill to your existing credits.
- Cloudflare says there is no markup over each provider's list price, and you can bring your own key.
- The page claims zero data retention and says the providers follow Cloudflare's verified bot standards.
What HN is saying
- Simon Willison asks the question that matters for agent builders: can you store and share results? Ceramic's terms look very restrictive, and others agree it is a real limitation. thread ↗
- A tester found the default provider returned no results or unrelated ones for ordinary queries, which undercuts the cheap alternative pitch. thread ↗
- Several people spot a contradiction: the announcement promises zero data retention, but the providers page says Exa does not offer it. Replies call it a documentation typo. thread ↗
- Many find it ironic that Cloudflare sells bot protection and now sells crawling too, and worry certified providers will get a pass. thread ↗
- Readers share alternatives instead: Kagi's API, SearXNG, self-hosted Firecrawl, Jina, and a local index called hister. thread ↗
Nearly every Dane's national ID number and home address just walked out the door.
What the article says
- Denmark's national person register says outsiders got in by abusing a Danish company's legitimate search access.
- Names, addresses and ID numbers for roughly 8.8 million people were exposed, which is more than the country's population because it includes past residents.
- People with protected name and address status were not affected.
- The company's access has been cut off, the data protection authority has been notified, and police are investigating.
What HN is saying
- Many say the ID number was never meant to be secret. Treat it as a username and let the national two-factor login be the password. thread ↗
- Insiders are not surprised. Private firms query the register through an API, and staff there can reportedly look people up freely. thread ↗
- Worry about identity recovery: the process for getting a new login asks mostly family questions, and family links may now be leaked. thread ↗
- A wider fatigue about leaks, with several people saying nothing changes without fines or jail time for the organisations responsible. thread ↗
- A wry Swedish comparison: Sweden publishes this kind of data openly, though others ask whether that invites stalking and harassment. thread ↗