Happy Sunday,

Welcome to ‘AI Weekly Insights’ 84. This week, AI news came with a big theme: control. A startup shocked the industry with a $34.5B shot at Chrome, Meta stumbled after leaked docs showed chatbots allowed to flirt with minors, Anthropic set stricter guardrails around Claude, and Google quietly flipped Gemini’s memory on by default.

Ready? Let’s dive in.

The Insights

For the Week of 08/10/25 - 08/16/25 (P.S. Click the story’s title for more information 😊):

Perplexity Wants to Buy Google Chrome

• What’s New: Perplexity, the AI search startup, shocked the tech world with an unsolicited $34.5 billion cash offer to buy Google’s Chrome browser.

• A Bold Browser Bid: Perplexity says it would keep Chromium open source and run Chrome independently, while leaving Google Search as the default. The timing is no accident, since Google is in the middle of a DOJ antitrust trial where selling Chrome has been floated as a possible remedy. Industry voices like DuckDuckGo’s CEO have suggested Chrome’s value could be closer to $50 billion. What makes this bid stand out is that Perplexity has raised only around $1.5 billion to date, making its offer look both gutsy and improbable. Still, owning a browser means owning the gateway for billions of users, which explains why the move rattled competitors and analysts alike.

• Why It Matters: This offer shows how quickly even the strongest tech giants can be forced into tough positions when regulators step in. The browser is still the front door to the internet, and whoever controls it influences how billions of people start their searches, manage their data, and interact with ads. Perplexity’s pledge to keep Google as the default is telling because it shows the real prize isn’t just Chrome’s code, it’s the reach and trust that come with it. For regulators, Chrome is the cleanest lever to loosen Google’s grip on search. For challengers like Perplexity, buying Chrome would shortcut years of slow user adoption and hand them direct access to a massive distribution channel. Even if the deal never closes, the offer changes the game: it signals that AI-first companies don’t just want to compete on features, they want control of the funnel itself. And if Chrome truly becomes available, bigger players like Microsoft or Apple could step in too. The real question may not be whether Google is forced to sell, but what happens to the shape of the internet if someone else takes ownership of the doorway.

Leaked Meta Docs Show AI Chatbots Flirting With Minors

• What's New: Internal Meta documents leaked to Reuters show the company’s AI chatbots were permitted to engage in romantic conversations with minors, and Meta removed those guidelines after the leak.

• AI Gone Wild: The leaked documents were internal policy drafts used for training and review. They reportedly allowed chatbots to carry out “romantic” or sensual conversations with both adult users and minors while still banning explicit sexual acts. Those examples created a dangerous gray area that many experts called grooming adjacent, and the leaks sparked bipartisan outrage. Meta confirmed the documents were genuine, said the examples were mistaken, and said it removed the problematic guidance. Lawmakers are pressing for explanations and transparency about what safeguards were in place and how the policies were approved. The episode adds to mounting pressure on Meta to tighten safety for generative AI inside messaging apps and virtual assistants.

• Why it Matters: This isn’t a small mistake, it’s a serious design failure with real risks for kids. Allowing chatbots to flirt with minors shows how chasing “engagement” can clash with the basic duty to keep people safe. If policy teams treat that behavior as acceptable in training material, the model will copy it and pass that danger on to users. Lawmakers are already paying attention, and trust in platforms that mix AI with social features is hanging by a thread. For builders, the takeaway is simple: smooth conversation means nothing without strong guardrails, oversight, and deliberate content choices. For parents and teachers, it’s another reminder that AI features can recycle old dangers in new packaging, and close watch is non-negotiable. For Meta, already battered by privacy and misinformation scandals, this leak deepens its trust problem and makes stricter regulation or outside audits more likely.

Anthropic Tightens Claude’s Safety Policy

• What's New: Anthropic updated its policy for Claude, banning use of the chatbot to develop biological, chemical, radiological, or nuclear weapons. It also added rules to stop people from using Claude for hacking, malware, or other cyberattacks.

• A Safer Claude: Previously, Anthropic’s guidelines simply said “don’t make weapons.” The new version spells things out clearly, with explicit bans on high-yield explosives and weapons of mass destruction. The update also acknowledges the risks from features like Claude Code and “Computer Use,” which let the AI run actions on a user’s machine. On top of that, Anthropic added safeguards to block hacking and malware creation. Interestingly, the company also relaxed its political content rules, now only forbidding uses that are deceptive or harmful to democratic processes instead of banning all political discussions.

• Why it Matters: This update matters because it replaces vague “don’t do bad things” promises with rules that are actually enforceable. By calling out chemical, biological, radiological, and nuclear weapons directly, Anthropic is admitting that today’s AI models are powerful enough to be dangerous if abused. The added cybersecurity rules reflect the reality that giving AI the ability to run code or control computers can be useful, but it also opens the door to new attacks if not restricted. At the same time, easing the political guidelines shows Anthropic is trying to allow open discussion without letting the AI be used for manipulation. For businesses, these changes lower legal risks and make Claude safer in sensitive settings. For policymakers, they show that some companies are willing to set guardrails on their own before laws force them to. And for everyday users, the update is a reminder that AI isn’t just a toy or a brainstorming partner. It’s a tool with real risks that has to be handled carefully.

Google’s Gemini AI Will Remember Details Automatically

• What's New: Google is rolling out an update to Gemini that allows the chatbot to automatically remember your past conversations, details, and preferences. The feature will be switched on by default in select countries starting with Gemini 2.5 Pro.

• Personalization by Default: Previously, users had to ask Gemini to “remember” information, such as interests or preferences. Now, the AI will store this context automatically to personalize answers without prompting. For example, if you once asked Gemini for YouTube ideas around Japanese culture, it might later suggest food-related video concepts in the same theme. Google says the memory feature can be turned off at any time through settings under “Personal Context.” Alongside this, Google is renaming “Gemini Apps Activity” to “Keep Activity,” which, if enabled, allows a sample of your file and photo uploads to be used for improving services. Temporary chats are also being added, giving users the ability to keep conversations out of history and training data, with content deleted after 72 hours.

• Why it Matters: Automatic memory turns Gemini from a reactive chatbot into something closer to a personal assistant, but it also raises important questions about privacy and trust. Personalized suggestions can make the AI more useful for creators, students, and professionals who rely on it regularly. At the same time, Google enabling this by default means users must actively opt out if they are uncomfortable with their chats being remembered, which may feel intrusive. Features like temporary chats and clearer toggles show Google is aware of potential backlash and trying to balance convenience with control, but they also highlight how much personal data is now in play. For businesses and educators, this shift shows how AI is working its way into daily tasks, tied directly to how data is managed. For everyday users, it’s a reminder to check your settings, decide what you want stored, and remember that AI memory isn’t just a convenience. It is also a responsibility to manage carefully.

That wraps up the week. The stories might look different on the surface, but they all ask the same question: who is holding the keys. Who controls your browser, who keeps AI safe for kids, who sets the line on dangerous uses, and who decides what your chatbot remembers about you.

Thanks for reading and giving me a few minutes of your Sunday. The goal is always to keep things clear and practical so you can walk away with insights worth sharing at work or with friends.

See you next Sunday with more!

Warm regards,
Kharee