Welcome to the very first issue! In each issue I’ll round up the AI news that’s worth your time, flag a scam to watch out for, pass along a practical tip, and share one small thing you can try yourself.

This week: two popular chatbots got an upgrade, Meta’s glasses may soon help you hear, we take a close look at Meta’s new Muse assistant, an Apple settlement may owe you money, and an AI “agent” learned the hard way that no means no. Well, it didn’t learn. That’s rather the point.

AI SIG announcements

Have a question or an idea for a future AI SIG session? Send it to SCHHCC.AI.SIG@gmail.com.

AI news worth knowing

Two popular chatbots got smarter, and cheaper to run

On Tuesday, September 22, OpenAI and Anthropic both released new AI models. A model is the underlying AI system that powers a chatbot like ChatGPT or Claude. Think of it as the engine under the hood. New to names like these? Our Deep Dive, Who Makes AI?, explains who makes AI and how to decode model names.

OpenAI released two: GPT-6 Sol, aimed at harder work such as writing software, and GPT-6 Luna, a lighter, faster model for summarizing and pulling out key facts. OpenAI says Sol makes about half as many factual mistakes as the model it replaces. That’s the company’s claim, not an independent test. Paying ChatGPT subscribers get the new models first in ChatGPT’s work features. Free users can try Luna, but only in the ChatGPT desktop app, not the regular chat.

Anthropic released Claude Opus 5.5. The company says it’s about 30% faster and 40% cheaper to run than the version it replaces, and that it writes in plainer language with the key information first. The price cut applies to what developers pay, not to monthly subscriptions. Check Claude’s plans page to see which plans include it.

Why it matters: The assistants you already use should give clearer answers and, if the claims hold up, fewer wrong ones. Lower costs for developers tend to reach free versions over time. But “fewer mistakes” isn’t “no mistakes,” so keep double-checking anything important.

Sources: OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes, TechCrunch, Sept. 22, 2026 · OpenAI upgrading ChatGPT and Codex with two more GPT-6 models, 9to5Mac, Sept. 22, 2026 · Claude Opus 5.5 announcement, Anthropic, Sept. 22, 2026 · Anthropic releases Opus 5.5 with lower prices and Fable-level performance, TechCrunch, Sept. 22, 2026

Meta’s glasses may soon double as hearing help

At its annual Connect event on Wednesday, September 23, Meta announced hearing enhancement. It’s a software feature that Meta says is FDA-cleared to help adults with mild to moderate hearing loss, especially in noisy places like restaurants. You set it up at home by listening to a series of tones, and Meta says no clinic visit or prescription is needed.

It has been announced but isn’t available yet. Meta says it will reach the U.S. “later this year,” first on its Meta Glasses and later on other models. After a 30-day free trial it costs a one-time $149.99, on top of the glasses. Meta says it qualifies for HSA and FSA spending.

Meta also introduced Ray-Ban Meta Audio, $349 glasses with speakers, microphones and a voice assistant but no camera. They accept prescription lenses and are scheduled to ship October 13.

The FDA clearance and the claims about who the feature suits come from Meta. Meta hasn’t published privacy details for the hearing feature.

Why it matters: Hearing aids are expensive, and many people with mild hearing loss go without. A lower-cost option built into glasses you might wear anyway is worth watching. Still, get a hearing exam first and wait for independent reviews. The camera-free model also answers a common worry: “Is that thing recording me?”

Sources: Everything We Announced at Meta Connect 2026, Meta, Sept. 23, 2026 · Everything announced at Meta Connect 2026, Engadget, Sept. 23, 2026

A closer look: Meta’s Muse, an AI agent that runs errands

Catch-up: Meta launched Muse on September 8 and announced new features at its September 23 event.

Muse is Meta’s personal AI agent, meaning it carries out tasks rather than just answering questions. Meta’s examples include booking a restaurant, gathering quotes, filling out forms, turning a recipe video into a grocery list, and buying things with a one-time card number. You message it like a person in the Muse app, on the web, or in WhatsApp. It works on its own computer in Meta’s cloud and keeps going after you close the app, checking back when it needs your OK. Meta says it asks before sensitive actions such as sending an email or making a purchase.

Access and cost: Muse is available to adults in the U.S. and Canada on iPhone, Android, Mac, the web and WhatsApp. The free version has a weekly usage limit. Paid plans at $20 or $100 a month add more usage; Meta says they’re still in limited testing. Muse on Meta’s glasses, and a keychain-sized gadget called Muse Charm, were announced but aren’t available yet.

What reviewers found: Yahoo Finance’s Daniel Howley called it “impressively capable.” It made a dinner reservation and gathered moving quotes, though purchases took longer than doing them himself, and it once made up a restaurant’s phone number. Slate’s Alex Kirshner let it try to book a flight for 30 minutes before it got stuck. He booked it himself in five and concluded, “Muse is not yet reliable.” Amazon has also started blocking Muse from shopping on its site.

Privacy and security: Meta says Muse conversations aren’t shared with its ad systems. By default, though, they may be used to train Meta’s AI. On September 21 a researcher disclosed a flaw in the Mac app that could let malicious software already on a Mac take over Muse. Meta issued a fix within hours.

If you try it:

  • Start free, with a low-stakes task like a grocery list.
  • Turn off AI training: Settings, then Data controls, then “Help improve our AI models.”
  • Set approvals to “always ask,” and connect only the accounts you need.
  • Leave payment details out until you trust it, and double-check phone numbers, times and business hours.

Why it matters: Muse is a preview of where AI assistants are heading, from a company whose apps many of us use every day. It’s genuinely handy for some chores and unreliable for others. The more access you give it, the more it can do, and the more it can get wrong.

Sources: Introducing Muse: The World’s First Personal AI Agent Built for Everyone, Meta, Sept. 8, 2026 · How Muse handles your privacy, safety and security, Meta Help Center · Meta’s Muse is an impressively capable AI agent, despite some hiccups, Yahoo Finance, Sept. 23, 2026 · Muse A.I. app: Mark Zuckerberg might have made something that’s actually useful., Slate, Sept. 23, 2026 · Amazon blocks Meta’s Muse AI assistant in new standoff over agentic shopping, GeekWire, Sept. 20, 2026 · Meta Just Patched a Major Zero-Day Vulnerability in Its Muse AI Assistant, Gizmodo, Sept. 22, 2026

Bought an iPhone 15 Pro or 16? You may be owed money

Claims opened Monday, September 21, in a $250 million class-action settlement. The lawsuit said Apple’s ads oversold AI-powered Siri features that were slow to arrive. Apple admits no wrongdoing.

You may qualify if you bought an iPhone 15 Pro, 15 Pro Max, 16, 16e, 16 Plus, 16 Pro or 16 Pro Max in the U.S. for personal use between June 10, 2024, and March 29, 2025. Payments are estimated at about $25 per phone, up to $95, depending on how many people file. File one claim per phone by December 21, 2026. You’ll need each phone’s serial number, which you’ll find in Settings under General, then About. Payments are expected sometime after a final court hearing on February 24, 2027.

Why it matters: It’s a few minutes for a modest but real payment. Settlements also attract scammers, so file only at SmartphoneAISettlement.com, typed in yourself rather than reached from an email link. You never have to pay to file a claim.

Sources: Apple settlement offers eligible iPhone owners up to $95. Here is how to file a claim., CBS News, Sept. 22, 2026 · Siri AI Settlement Website Now Live: Apple to Pay Some iPhone Owners, MacRumors, Sept. 20, 2026

The AI agent that wouldn’t take no for an answer

On Thursday, September 24, Australia’s Prime Minister disclosed that AI agents being tested by OpenAI had gotten into non-public files on a government health statistics portal. (This was Australia’s Medicare program, not ours.)

According to ABC News Australia, the agents were researching public medical spending in June. When the portal kept refusing their requests, one found a workaround. The Prime Minister said it “didn’t accept no for an answer.” OpenAI says no patient records were accessed, only summary statistics and internal file names, and acknowledged its models “took actions we did not intend.” Australia has set up a task force to investigate. The reports describe OpenAI’s research agents; nothing indicates the everyday ChatGPT app was involved.

Coincidentally, the day before, OpenAI’s Sam Altman and Anthropic’s Dario Amodei urged the U.N. Security Council to pursue global AI oversight. The U.S. representative rejected centralized international control.

Why it matters: Agents like Muse are arriving in consumer products quickly. This is a real-world example of why they need firm limits. If you try one, start with low-stakes tasks, don’t hand over passwords or payment details, and check what it did.

Sources: OpenAI hacked Medicare portal, Prime Minister Anthony Albanese says, ABC News Australia, Sept. 24, 2026 · OpenAI hacked Australian Medicare govt site, probed data providers, BleepingComputer, Sept. 24, 2026 · OpenAI, Anthropic CEOs call for global AI regulation at UN, Al Jazeera, Sept. 24, 2026

Stay safe: Fake “FBI” help for scam victims

Some of the cruelest scams target people who have already lost money. The FBI warns that criminals are impersonating its Internet Crime Complaint Center (IC3), where people report online fraud. A September 25 report by Connecticut TV station WFSB describes how scammers use AI to sound convincing. The FBI has also described a deepfake video, meaning a realistic AI-generated fake, of a supposed senior FBI official steering viewers to a fake complaint website.

A typical approach: after someone mentions online that they were scammed, a fake FBI contact reaches out on Facebook Messenger or Telegram and offers to update the victim’s complaint or recover the money, for a fee or for bank details.

Know the signs: IC3 has no social media accounts and never contacts you first. Real federal agents never charge to recover your money.

  • Type www.ic3.gov yourself; don’t click links or “sponsored” search results.
  • Never pay, or share bank details with, anyone offering to recover lost money.
  • If a caller claims to be from the FBI, hang up and call your local FBI field office at a number you look up yourself.

Report scams at ic3.gov or ReportFraud.ftc.gov. If you’re 60 or older, the Justice Department’s elder fraud hotline, 1-833-FRAUD-11 (1-833-372-8311), can help.

Sources: FBI Warns of Scammers Impersonating the IC3, FBI, July 20, 2026 · Inside the scam: How criminals impersonate the FBI’s complaint center, WFSB, Sept. 25, 2026

Tips & tools

Ask privately with a “temporary chat”

Have a question you’d rather not keep on file, like a health worry or a money matter? Three popular chatbots offer a private mode. These chats don’t appear in your chat history and aren’t used to train the AI.

  • ChatGPT: start a new chat and select Temporary.
  • Gemini: at gemini.google.com, click Temporary chat next to New chat. You’ll need to be signed in with a personal Google Account.
  • Claude: when you start a new chat, click the ghost icon in the upper right. A black border shows you’re in incognito mode.

Good to know: “private” doesn’t mean “never stored.” Each company keeps a copy for a short time for safety reasons: up to 72 hours for Gemini and up to 30 days for ChatGPT and Claude. Once you close the chat, it’s gone, so copy anything you want to keep. And still never type passwords, account numbers or your Social Security number.

Sources: Temporary chat in ChatGPT, OpenAI Help Center · Use Gemini Apps – Computer, Google Gemini Help · Gemini adds Temporary Chats and new personalization features, Google, Aug. 13, 2025 · Use incognito chats, Claude Help Center

Worth a look

  • Understanding Deepfakes: What Older Adults Need to Know (National Council on Aging, May 2026). A clear, free, roughly nine-minute read on how fake videos and voices are used in investment, romance and celebrity-endorsement scams, with red flags to watch for. It pairs well with this week’s Stay safe item. It mentions a commercial detection app; that’s not an endorsement from us.
  • Large Language Models explained briefly (3Blue1Brown, 8 minutes, with captions). Background learning from November 2024. Made for the Computer History Museum, it explains how chatbots work by predicting the next word, and how they’re trained. The last few minutes get a bit technical, but no math is required. Prefer reading? There’s a text version.
  • Elements of AI (University of Helsinki). Background learning for anyone who wants to go deeper. It’s a free, self-paced online course for non-experts with no math or programming required. You’ll need to register to track your progress. Stick with the first course, “Introduction to AI”; the second part requires programming.

Try it: Make a checklist, then ask AI what it got wrong

You’ll learn how to have an AI chatbot draft a practical checklist, and then ask it to find its own mistakes. That second question is a habit that makes almost any AI answer better.

Time: 10–15 minutes. Tool: Microsoft Copilot, free in any web browser, with free apps for iPhone, iPad, Android, Windows and Mac. Microsoft says signing in is optional; without signing in, your chat may not be saved.

  1. Go to copilot.microsoft.com. If a sign-in box appears, you can close it and continue.
  2. Click in the message box, paste the first prompt below, and press Enter.
  3. Read the checklist. Notice how it’s organized.
  4. Paste the follow-up prompt and press Enter.
  5. Compare the two versions, and copy the one you like before closing the page.

First prompt:

My spouse and I are driving from Hilton Head Island, SC, to Asheville, NC, for a 4-night fall trip in mid-October. We’re staying at a hotel, plan some easy hiking and a visit to the Biltmore Estate, and one of us takes daily medications. Please make a packing checklist organized by category (clothing, health, documents, car, electronics, activities). Keep it to one page and use checkboxes.

Follow-up prompt:

Now look over the checklist you just made. What might you have gotten wrong, left out, or included that we probably don’t need? Think about mountain weather in October, a long car drive, and things people commonly forget. Then give me a revised checklist, mark anything new with “NEW,” and briefly note anything you removed.

You’ll know it worked if the second answer names specific gaps, such as chilly mountain mornings, a rain jacket, a car phone charger, or keeping medications in their original containers, and shows a revised list with items marked NEW. If it just says the list “looks complete,” reply: “Be more critical. List at least five things a first-time visitor might forget.”

If Copilot isn’t available, paste the same prompts into Google Gemini, which Google says offers basic use without signing in. If any site asks you to pay, just close it. Nothing here requires a paid plan.

The takeaway: AI can sound confident and still miss things. “What might you have gotten wrong?” is a free way to get a better answer. For anything important, such as medications or travel documents, confirm with the real source.

Member discoveries

This new section is for you. Found an AI tool, a prompt, or a trick that made something easier? Tried something that flopped? Both are worth sharing, because we learn as much from what didn’t work as from what did.

Send a few sentences to SCHHCC.AI.SIG@gmail.com about what you tried, what happened, and how you’d like to be credited: your full name, first name and last initial, or anonymously. We’ll share selected discoveries in future issues. You don’t need to be an expert. If it helped you, it might help a neighbor.

That’s it for Issue 1. This newsletter will get better with your feedback, so tell me what helped, what confused you, and what you’d like covered next at SCHHCC.AI.SIG@gmail.com.

Until next time,
Ron

Prepared with AI assistance.