
Your Phone's AI Can Read Your Chats
PLUS: Claude may have cracked a $1M math puzzle, and ChatGPT pages can now update themselves
In this SunBrief
Your team has never built an AI agent, and that’s fine

Automation builds, RAG pipelines, data cleanup at scale. Most companies have nobody in-house who has done any of it. Training someone takes a year. Hiring for it means a bidding war against companies that pay more than you do.
The third option is renting the expertise for one contained project and seeing how it goes before committing to anything bigger. Upwork has specialists across more than 125 categories, including the ones that didn’t exist in 2023. Smarter with AI readers get a $50 Visa gift card after a first completed hire.
Find AI talent on Upwork$50 Visa gift card issued after your first completed hire. New clients only. Terms apply.
01 / Privacy watch
How to keep AI out of your chats

Signal now stops the AI built into phones from screenshotting your chats. What caught my attention was the AI app Signal called out.
- Meta’s AI assistant app, Muse, read a Signal message. It reminded a Meta employee to reply to her sister on Signal. Reuters later showed it got in through Android notifications.
- Staff also saw bank balances in a chat about the weather. One reply listed a Capital One balance of $2,722.58. Reuters couldn’t confirm the numbers were real, and Meta calls these “isolated incidents.”
- Your phone decides how you turn it on. Android does it for you, while iPhones get a new setting and Macs need a different fix.
Why it matters: Whatever chat app you use, I’d check which apps can read your notifications today. That’s how Muse got in on Android, and Signal says its own fix isn’t complete.
See where the setting is on your phone ↗Would you let an AI assistant read your messages?
Market Snapshot
| Company | Price ($) | 5D (%) | YTD (%) | ||
|---|---|---|---|---|---|
| 264.00 | -9.75 | -15.87 | |||
| 342.56 | -3.85 | 26.01 | ||
| 560.84 | -12.60 | -15.04 | ||
| 484.52 | -7.51 | 0.19 | ||
| 246.82 | -7.44 | 6.93 | ||
| 125.53 | -22.92 | -32.69 |
03 / Research watch
Claude may have cracked a $1M math puzzle

An Anthropic mathematician believes he and Claude have proved the Riemann hypothesis, a puzzle unsolved since 1859. What caught my attention was how the proof got checked.
- It’s one of math’s seven $1 million problems. It’s about how prime numbers are spread out, and it has stumped mathematicians for 167 years.
- A second program checked the AI’s work. The proof is 17 million lines of code for Lean, software that checks math step by step, and on Saturday the Lean community certified it.
- Experts are excited but careful. A winner of math’s top prize, the Fields Medal, gave his verdict, and another expert explains why understanding it could take years.
Why it matters: I like that a second program checked the AI’s work line by line, because that’s the kind of proof I’d want before trusting any AI answer. The $1 million prize still needs publication and at least 2 years of review.
See what math’s top prize winner said ↗Do you check AI's answers before you use them?
04 / Tool watch
ChatGPT pages can now update themselves

ChatGPT Space, a home for pages and slides that keep themselves up to date, is now rolling out to Plus users worldwide. What caught my attention was a catch for Free users.
- Pages can update on their own. They pull changes from the apps you connect, and you can tag ChatGPT in a comment to ask for edits.
- OpenAI shared three ideas for pages. One is a study guide with a 3D model you can explore.
- There’s a catch for Free users. They can’t open a page a Plus user shares with them, and there are no public links.
Why it matters: If you work with people on the Free plan, Space won’t help you share yet. If you’re on Plus, look for it in your sidebar, where Library used to be, as it rolls out.
See the other 2 ideas from OpenAI ↗Would Space make you pay for ChatGPT Plus?
05 / AI highlights
AI in Brief
FOR RESEARCHERS
OpenAI Is Giving 200,000 Researchers Free ChatGPT Plus
Applications for 50,000 grants open worldwide on Monday, Nov. 23, first come, first served. Each grant covers you and up to two collaborators.
SEARCH
Google’s AI Mode Now Has an Incognito Option
Tap the new temporary-thread icon and your AI Mode chat won’t show up in your search history or recommendations. There’s nothing to set up.
EDUCATION
Google Tells Schools: Take AI or Lose Search
From early next year, schools that switch off Google’s AI for students also lose Google Search, and 3 of the 4 largest US districts are pushing back.
SECURITY
AI Middlemen Were Caught Reading People’s Prompts
After a researcher’s 2 million bait prompts exposed leaks, OpenRouter dropped 3 AI providers. Don’t paste passwords or client data into AI apps.
HEALTH
ChatGPT Can Now Read Your Android Health Data
ChatGPT Health now connects to Android Health Connect to spot patterns in your sleep and activity. It’s on every plan, including Free, but US only for now.
On Our Radar
- Google is retiring the Nest app and swapping Assistant for Gemini on speakers
- Word, Excel and PowerPoint now work inside Microsoft Copilot
- Apple will warn you before giving an AI app full access to your Mac
- Proton’s private AI assistant Lumo comes to Windows, Mac and Linux
- Dutch privacy watchdog tells shoppers not to buy AI camera glasses
- Anthropic pulls off the biggest IPO ever
Last Week’s Poll Result
Would you let Gemini make calls or buy things for you?
| Yes, I’ll try it this week | 25.71% |
| Only calls, not payments | 20% |
| No, I’d rather do it myself | 54.29% |
Reader voice · AUDREYCULLEN1
Selected: Yes, I’ll try it this week
“I've already allowed another platform to do it. I am open to it and most curious how the Healthcare industry will handle a call from a VA regarding appointment setting.”
Our take: A great real-world test. Booking appointments is one of the main jobs Tasks is built for, so offices may soon get a lot more AI calls.
Reader voice · RIOUXMARK0
Selected: Yes, I’ll try it this week
“I have no concerns with Gemini to do online shopping.”
Our take: You’re in the minority this time. Over half of readers would rather do it themselves, so a small first purchase is a sensible way to test it.
Your turn: Vote and tell us why. We will feature your take next week.
Should Nvidia bring Jensen Huang back as CEO?
| Yes, Nvidia needs him | 49.12% |
| Only after the investigation ends | 19.3% |
| No, it’s time to move on | 31.58% |
Reader voice · DENIABLEDENIABLE
Selected: Yes, Nvidia needs him
“I cannot believe the sheer stupidity in all the actors involved. Didn't we learn from the Altman case?”
Our take: A fair comparison. OpenAI brought Sam Altman back within days, but reports say Nvidia’s board has ruled out Huang’s return for now.
Reader voice · BHORES
Selected: Only after the investigation ends
“If he is cleared of the Chinese scandal, his other transgression should not matter.”
Our take: That’s the middle ground. Let the investigation finish, then decide.
Reader voice · RALPHMAY
Selected: No, it’s time to move on
“He has done an amazing job. At the same time, he has undermined/betrayed the trust of the company and the citizens of the United States. There must be consequences.”
Our take: A balanced view. You also raised the lack of a successor, which is a real risk for any company built around one leader.
Your turn: Vote and tell us why. We will feature your take next week.
Who should be responsible when an AI agent causes harm?
| The company that built it | 26.09% |
| The person who used it | 13.04% |
| Both, depending on the case | 60.87% |
Reader voice · FANNINGJG
Selected: Both, depending on the case
“If AI agent operates independently and causes harm company should be responsible. If AI agent programmed by user to purposely cause harm user should be responsible.”
Our take: A clear way to draw the line. California’s lawsuit against OpenAI may be one of the first tests of where that line sits.
Your turn: Vote and tell us why. We will feature your take next week.
We’d love to hear from you!
Thanks for reading. See you next Sunday.
