This week: AI in your exam room, and a delete with no undo

Good morning!

Somebody wants a yes out of you, and the yes is always easier to give than to take back. Two of the pieces below are about that. The clinic asks you out loud. The location app asked once, years ago, and has not asked since.

In this issue:

• Your doctor wants to record the visit with AI
• Your phone says storage is almost full. Don't delete anything yet.
• What you agreed to when you turned on location sharing
• The AI boom is showing up on price tags
• Also this week: your location, an AI shopper in court, and California's AI labels
• Five shortcuts I wish somebody had told me about
• Plus: three things going around right now
• And the Scary Headline of the week: a second AI company says its model hacked somebody

Your doctor wants to record the visit with AI. Here is what to ask before you say yes.

The clinic will tell you the tool is HIPAA compliant, and that sentence does less work than it sounds like. HIPAA governs how your information gets handled once it exists. Nothing in it requires anybody to ask you before a microphone turns on in the room. How much say you have is determined by your state's recording law. Declining is ordinary, and the notes get written the old way. There are four questions to ask your provider in the piece, and they are the same four whether you end up saying yes or no.

Learn

Your phone says storage is almost full. Don't delete anything yet.

The usual advice is to delete some photos, and on the wrong phone that advice destroys the only copy you own. Answer one question first. Is this phone actually backing up, or does it only look like it is. And if your iCloud storage is already full, photos you delete skip Recently Deleted altogether and are gone the moment you tap. Check the backup, then work through the cleanup in the order the piece lays out.

Learn

What you agreed to when you turned on location sharing

Two people agreed. Three parties are involved, and the third behaves very differently depending on whose app you use. Apple throws the shared location away after a day. Life360 built a business on movement data and says so in its own policy. Monday's piece walks through the four things to check, whichever app you use.

Blog

The AI boom is showing up on price tags

AI data centers are buying memory chips faster than anybody can make them, and the shortage has reached store shelves. Budget phones feel it worst, since memory makes up more of a cheap phone's cost than a flagship's. And some of it does not show up on the price at all. A device keeps its old price and comes with a little less inside.

Learn

Also this week: your location, an AI shopper in court, and California's AI labels

Four stories from the week I did not write a full piece about. The story to act on is the final one. Google is shutting off Assistant on phones September 4.

News

If you only read one: the doctor visit piece. This is going to come up for nearly everybody sooner or later, and knowing what is actually happening is what lets you make the call for yourself.

5-Minute Tech Tip

Five shortcuts I wish somebody had told me about. Start with the spacebar. When you have a typo six words back and tapping at it keeps putting the cursor in the wrong spot, press and hold the spacebar instead. The keyboard goes blank and turns into a trackpad, and your thumb walks the cursor one character at a time. That one and four more, all things your phone and computer have been able to do for years without telling you.

Fresh Trouble

"Learn to trade" pitches. Posts full of fancy cars and exotic trips promising you the same life if you sign up to learn trading. The FTC says it flatly. Trading is risky, nothing is guaranteed, and the money can go fast. (FTC)

Government callers with paperwork. A call showing a real agency on the caller ID, using a real employee's name, backed up with official-looking documents. Nobody from a government agency will open by asking you to pay. (FTC)

Fake Facebook suspension warnings. A message saying your account closes for fraudulent activity unless you confirm your details. Open the app yourself and check your notifications there.

Scary Headline of the Week

"Meta's Muse Spark 1.1 hacked an external organization during cybersecurity test"

Meta ran its newest model through a hacking test with an outside firm, in a sandbox built to keep it off the internet. A configuration mistake gave it internet access anyway, and it used that access to break into some other company's systems. Nobody has said whose.

If that sounds familiar, it is the fourth one in about a month. OpenAI's agents got out and hit Hugging Face, which I wrote about earlier this month. Anthropic disclosed two. Now Meta. Three different labs, and every single time the explanation has been that the test environment was set up wrong.

None of these have been the consumer model. These are research systems inside company networks, and the models were chasing answers to a test rather than going after a target. Hugging Face says the models and files the public downloads were never touched.

The verdict: Nothing here to act on, but a worrying trend. Four in a month isn't a coincidence, and "we misconfigured the test" is starting to sound like a phrase rather than an explanation.

Seen a headline this week that scared you? Reply and send it. It might get next week's verdict.

Help Fresh From Cache grow

This newsletter is free and written by one person. If it earns its spot in your inbox, two ways to help: forward this email to one person who might want it, and if it was forwarded to you, subscribe to get your own copy every Tuesday.

Which of the five shortcuts had you never heard of? Hit reply and let me know.

Joel