What changed. Gemini (and Google's video/music tool, Flow) now has a "Media watermark" switch in settings — On or Off. With it off, images made with Gemini's image tool, videos made with its video tool, and music made with its music tool no longer carry the small visible AI mark in the corner. You'll find it in the web version of Gemini by clicking the gear icon in the bottom left, then "Media watermark." Google has said it plans to bring the same option to Search.

What did NOT change. Turning the switch off only removes the mark you can see. Every file still carries an invisible digital watermark (called SynthID) and metadata (called C2PA) embedded in the file itself, regardless of the setting. Tools built to check for that hidden marker — like Google Search's "About this image" feature — can still identify the content as AI-generated even with the visible mark switched off.

Read the full article on AI for Regulars

Group Chat becomes 1:1 only. If you've been using Copilot in a group conversation with other people, that group format disappears on August 18. Microsoft says existing group chats will be turned into individual chats, so your own messages stay visible in your history — but anything other participants shared or created inside the group, like their prompts or generated images, will no longer be accessible to you. If there's something from a group chat you want to keep, save it before the 18th.

Podcasts are shutting down completely. Copilot's feature for turning documents or web pages into an AI-generated audio "podcast" is being removed, not just changed. After August 18 you won't be able to create new ones or open ones you already made, and any links you shared to a Copilot podcast will stop working. You can still download individual podcasts from your podcast library before the deadline — look for the download option in the podcast's menu.

Read the full article on AI for Regulars

Who actually sees ads. Only logged-in adults on the Free and Go plans. If you pay for Plus, Pro, Business, Enterprise, or Edu, you won't see ads at all — that's one of the things your subscription covers. Ads also never appear in temporary (incognito-style) chats, regardless of your plan.

What the ads look like. OpenAI says ads are visually separated from ChatGPT's actual answer and clearly marked as sponsored, and that they don't influence what the AI tells you — the answer itself is generated the same way either way. You can tap an ad to see why it was shown to you, dismiss it, leave feedback, or delete your ad data with one tap.

Read the full article on AI for Regulars

What's actually changing. The rollout starts on the web this week and reaches phones and tablets on August 17. Once it's live, eligible students can open a Gemini tab inside Classroom and turn their own class materials — a reading, a set of notes, a study guide the teacher posted — into flashcards or practice quizzes tailored to that specific assignment. Google calls these "contextualized starter prompts": instead of a blank chat box, students pick a class and an assignment first, and Gemini builds the study aid around it.

Your school has to turn it on — it doesn't switch on by itself. This is the part worth knowing before you worry: students only get access if their school's IT administrator has already enabled Gemini in Classroom, Gemini, and Gemini Notebook for that account. If your school's admin hasn't turned those on, nothing changes for your child on August 10 or 17 — the tab simply won't appear. Schools can also set up a separate group for students under 18 and switch access off for just that group, independent of older students or staff.

Read the full article on AI for Regulars

What's actually changing: Free and Go accounts are moving to GPT-5.6 Luna as their default model this week. The daily rate limit on text chats is being removed in the same rollout wave — OpenAI says that part lands the following week, starting around August 10. Once it does, you'll be able to keep typing and asking questions all day without hitting a wall. Both changes are rolling out gradually, so they may show up for you a few days before or after they show up for someone else — that's normal for this kind of update, not a sign anything is broken.

What's still limited: This change covers text only. Uploading files, generating images, using voice mode, and other tools still have daily limits on Free and Go. If you're doing a lot of image generation or document uploads, you'll still run into caps there — Plus and Pro remain the way to raise those.

Read the full article on AI for Regulars

Here's what was supposed to start on August 2, 2026. Under the EU AI Act, companies and public bodies that use "high-risk" AI systems — software that helps decide who gets a job interview, a loan, a school place, or how police and border agencies use AI — were meant to have finished risk assessments, documentation, and human oversight by that date. That's the part that's been delayed.

The delay came through something called the "Digital Omnibus on AI," a formal amendment to the AI Act. The European Parliament and EU governments approved it in June 2026, it was published in the EU's Official Journal on July 24, and it became binding law on July 27 — a week before the original deadline. The new dates: December 2, 2027 for those standalone high-risk systems, and August 2, 2028 for AI built into regulated products like medical devices or cars. EU officials said the extra time is needed because the technical standards and testing tools companies need to comply weren't ready.

Read the full article on AI for Regulars

The rule applies to any generative AI system with more than one million monthly users in California — in practice, that covers ChatGPT (OpenAI), Gemini (Google), Meta AI, and other large platforms. Since companies rarely build a separate version of a product just for one state, the changes will likely show up for users everywhere, not only in California.

What's supposed to change

Read the full article on AI for Regulars

The numbers presented were stark. Americans reported losing $21 billion to scams in 2025, and seniors alone accounted for more than $7.7 billion of that. Consumer advocates testified that the real total is likely 7 to 25 times higher, since most fraud goes unreported.

One witness, Deborah Del Mastro of Martinez, California, described a call where she heard what sounded exactly like her daughter, panicking, saying she'd been kidnapped. She wired $5,400 before learning her daughter had never been in danger — the voice on the phone was an AI clone.

Read the full article on AI for Regulars

What it actually does. Gemini used to open in a floating window over the page. Now it lives in a side panel that stays open next to whatever tab you're viewing. Click the "Ask Gemini" icon and it can summarize the page you're on, answer questions about it, or — if you choose to share more than one tab — compare information across up to 10 open tabs at once, which is handy for things like comparing prices or reading several reviews side by side. It also connects to other Google apps: it can draft an email and hand it to Gmail as a ready-to-send compose window, pull up relevant results in Maps, check your Calendar, or pull key points from a YouTube video's transcript. People on a paid Google AI Pro or Ultra plan also get "Auto Browse," a feature that lets Gemini carry out a multi-step task on your behalf across tabs, such as researching several options and filling in details for you.

What it can see. When you share a tab with Gemini — the tab you're currently viewing is shared by default once you open the panel — it reads that page's content and web address, and does the same for any extra tabs you add. Google says that data is logged temporarily to your account, but doesn't show up in your regular Gemini Apps Activity, and isn't used to train its models without permission. You can stop sharing a specific tab at any time by clicking the "X" next to its name in the panel, and you're not required to share anything to keep using Chrome normally.

Read the full article on AI for Regulars

What's new. Announced Thursday, July 23, 2026, ChatGPT Health lets you connect your Apple Health data — steps, sleep, workouts — straight into your ChatGPT conversations. If your doctor's office uses certain electronic record systems (Epic and Oracle Health are the two OpenAI named), you can link those too, so ChatGPT can see things like recent lab results or visit summaries. Once connected, you get a dedicated Health tab that keeps its own conversation history, separate from your regular chats, where you can track how a symptom, medication, or sleep pattern changes over time and prepare questions before a doctor's appointment. It works on the web and in the ChatGPT app on iPhone — it hasn't rolled out to Android yet.

Why now. OpenAI says people already send it roughly 300 million health-related questions a week, up from about 230 million back in January. Rather than leaving that happening informally in ordinary chats, the company built a dedicated space for it — with, it says, the same underlying model but a setup designed specifically around tracking health information over time rather than answering a one-off question.

Read the full article on AI for Regulars

What Meta's own watchdog found. On June 5, 2026, the Oversight Board — the independent body Meta itself set up to review its content moderation — published its first ruling on an account disablement case. It upheld Meta's decision to permanently disable the specific Instagram account in question, which had posted threats and slurs. But the board used the case to call out much bigger problems: it found Meta's system for sorting violations into "severe" and "egregious" categories, which determines whether an account gets a warning or an instant permanent ban, to be "unclear and often contradictory." The board's most pointed recommendation was that people should be told whether and how AI was involved in reviewing their account and deciding any penalty against it — something Meta currently doesn't disclose.

Thousands say they were caught up in it. A petition from the Canadian nonprofit People Over Platforms, titled "Hold Meta Accountable for Wrongful Account Disabling," has drawn tens of thousands of signatures over 2026 from people who say Facebook, Instagram, or Threads accounts — including long-running small business pages — were disabled without warning, often citing "severe" categories like child exploitation, with no human reachable to explain why. Local news outlets including CBS affiliates in Baltimore and Philadelphia have covered similar cases. Meta disputes that AI is the problem: a company spokesperson told Boston 25 News that its current moderation tools make 13% fewer mistakes than human reviewers did and catch 10% more genuine violations — but didn't dispute that appeals can be slow or unclear.

Read the full article on AI for Regulars

The FBI's Internet Crime Complaint Center (IC3) issued a public warning this week about scammers who specifically go after people who already filed a fraud report or lost money in the past. The pitch: they can help you recover what you lost. It's a second scam layered on top of the first one.

The FBI describes two versions of the trick. In one, someone messages you on Facebook Messenger or by email claiming to have an "update" on your IC3 report, then moves the conversation to Telegram and sends a link that actually collects your financial information. In the other, an AI-generated deepfake video shows someone posing as a senior FBI official, pointing you to a fake website designed to look like the real ic3.gov — but built to steal your personal details through a simplified "report" form.

Read the full article on AI for Regulars

What Google is accused of. The case, In re Google Assistant Privacy Litigation (Case No. 5:19-cv-04286, U.S. District Court for the Northern District of California), alleges that Google Assistant sometimes activated and recorded audio without anyone actually saying "Hey Google" or "OK Google" — a mistake the industry calls a "false accept" — and that some of those recordings were reviewed by outside contractors to help improve the software. Google denies any wrongdoing and settled without admitting fault.

Who can file a claim. Two groups qualify, and you don't need to belong to both. You're likely eligible if, between May 18, 2016, and March 19, 2026, you or your minor child either bought a Google-made device — a Google Home, Home Mini, Home Max, Nest Hub, Nest Hub Max, or Pixel phone — or used any Google Assistant-enabled device (including third-party smart speakers with Assistant built in) and believe a conversation was picked up without the wake word being said. Both groups must also confirm they live in the US or a US territory.

Read the full article on AI for Regulars

How it works. A dedicated AI classifier scans chats with Meta AI for signs of self-harm or suicidal distress. Flagged conversations get a human review before any alert goes out, and Meta says that when the signal isn't clear-cut, it errs on the side of notifying a parent anyway.

What you'll see — and what you won't. If your teen's account is flagged, you'll get a notification letting you know there's cause for concern, along with crisis resources and tips for starting a conversation. You won't see the exact wording of what your teen wrote — Meta keeps that private.

Read the full article on AI for Regulars

What it actually does. Normally, you chat with ChatGPT back and forth: you ask something, it answers, you ask a follow-up. ChatGPT Work is different — you describe the outcome you want, and it works through the steps on its own, sometimes for a long stretch of time, and comes back with a finished result. That could mean turning a rough list of numbers into a formatted spreadsheet, building a slide deck from a set of notes, drafting a document, or pulling together a short webpage. It can keep working after you close the app, and it will check in with you before doing anything it considers a "consequential" action, like sending an email or making a purchase — you can also pause, redirect, or stop it at any point.

It works by connecting to your other apps. To do this, ChatGPT Work needs access to things like your email, calendar, or file storage (for example, Gmail, Google Drive, or Slack), depending on which connections you set up. That is the trade-off worth understanding before you turn it on: the more accounts you connect, the more the agent can see and touch on your behalf, so it's worth connecting only the accounts you actually want it working in, and reviewing what permissions you're granting.

Read the full article on AI for Regulars

How to use it. Look for a magnifying glass icon near the top of the sidebar, or press Ctrl+K on Windows or Cmd+K on a Mac, and start typing. The search looks through everything at once — your past chats, any Projects you've set up, and images or documents you've uploaded — and lets you narrow the results down by type instead of scrolling through one long mixed list.

It even finds archived chats. If you've ever archived a conversation to tidy up your sidebar, it normally disappears from view entirely. The new search still finds it — archiving hides a chat from the list, it doesn't remove it from search.

Read the full article on AI for Regulars

How to get it. Sign up for free through the Apple Beta Software Program, then install iOS 27 Public Beta from Settings > General > Software Update on your iPhone. No developer account and no fee required — just some time and a supported device.

Back up your phone first. A public beta is, by definition, not finished software. Apps can crash, battery life can suffer, and features can behave unpredictably. Before you install it, back up your iPhone to iCloud or your computer so you can restore it if something goes wrong.

Read the full article on AI for Regulars

How it works. Avatar creation starts inside the Gemini app: tap "Add Files" in the message box and choose Avatar, then scan the QR code that appears using your phone. From there, you'll be asked to look into the camera, turn your head a few different ways, and read some numbers aloud so the system can map your face and voice. The whole thing takes a few minutes. Afterward, you can bring your avatar into a conversation by typing "@me" or your name, and Gemini's video-generation model — called Omni — can produce short videos where you appear to be speaking, using your actual face and voice.

Who can use it. Avatar requires a paid Google AI plan (Plus, Pro, or Ultra) — it isn't available on the free tier. You also need to be 18 or older, and you have to be the one physically present during setup; you can't create an avatar of someone else from a photo.

Read the full article on AI for Regulars

Here's how it works: when someone calls you from a number you've saved as a contact, Android silently verifies through an end-to-end encrypted RCS handshake whether the call is really coming from that contact's device. If it can't confirm the match, your screen shows a warning telling you to be careful before you continue the conversation.

The feature launched June 2, 2026, and is rolling out to Android 12 and newer through the Phone by Google app, starting with Pixel devices.

Read the full article on AI for Regulars

This is part of the EU AI Act's transparency obligations, which are now coming into force for consumers. Some business-side requirements under the Act have been delayed, but the rules that directly affect what you see and hear as a regular user are moving ahead on schedule, according to coverage from InsidePrivacy (May 2026).

What you'll notice

Read the full article on AI for Regulars

What's actually new. When you turn on Incognito Chat before talking to Meta AI, those messages are processed in a Trusted Execution Environment (TEE) — a secure, isolated part of the server that Meta says even its own engineers can't access. The AI chats also disappear by default when the conversation ends, leaving no history stored on Meta's servers.

Why this matters. Normally, when you chat with Meta AI in WhatsApp, those messages don't have the same protections as your regular WhatsApp conversations. Your regular messages to friends and family have always been end-to-end encrypted — Incognito Chat doesn't change that. What it does is bring your Meta AI chats closer to that standard for the first time.

Read the full article on AI for Regulars

The biggest practical change: for the first time, ChatGPT can search the web in the middle of a spoken conversation. Before, voice mode could only answer from what it already knew. Now, if you ask something that needs a current answer — a price, an opening time, today's weather — it can look it up and speak the answer back to you, without you having to switch to typing.

You don't need to do anything to get it. Free accounts are automatically switched to GPT-Live-1 mini; Plus and Pro subscribers get the larger GPT-Live-1. If you don't see it yet, updating the ChatGPT app in the App Store or Google Play should bring it in.

Read the full article on AI for Regulars

Sol is the flagship. It's built for tasks that take serious thinking: complex research, writing long reports, coding projects, and multi-step problems where the AI has to plan ahead. If you've heard the phrase "deep reasoning," this is the model that does that. There's even a stronger version called Sol Ultra for the most demanding work. Most everyday users won't need this one for routine tasks.

Terra is the middle option. OpenAI describes it as balanced — roughly as capable as what ChatGPT used for last year, but at half the running cost. The most important thing for regular users: Terra is now the default model for Free and Go accounts. That means if you use ChatGPT without paying, you're getting a better AI starting today, automatically, with no action required on your part.

Read the full article on AI for Regulars

That can be handy when you want a quick answer. But it also means you may see fewer links to the original websites — so you might have to scroll more to find the source.

The AI answer can be wrong sometimes, so it's always smart to click through and check the original page, especially for health or money questions.

Read the full article on AI for Regulars

Here's the catch: many of the apps running these trends are small or unknown. When you upload your face, you often agree — buried in the fine print — to let the company store or even sell your photo.

Your face is unique biometric data. Once it's out there, you can't take it back.

Read the full article on AI for Regulars

The new options let you reduce sensitive content, turn off the ability to generate images, and receive a safety alert from OpenAI in the rare event that serious self-harm concerns are detected. These controls live inside your account settings on the ChatGPT website or app.

To turn them on, go to your account settings, find the "Parental controls" section, and follow the prompts to link your child's account to yours. Once linked, you can choose what's allowed and what isn't.

Read the full article on AI for Regulars

That voice is fake. AI tools can copy someone's voice from just a short audio clip — even a social media video — and use it to make a convincing phone call.

The trick works because you're scared and acting fast. Scammers count on that.

Read the full article on AI for Regulars