Meta’s New AI Assistant Had a Pretty Terrifying Security Flaw
Meta launched its new Muse AI assistant earlier this month with a big emphasis on security, privacy, and letting the AI actually do things on your behalf. (Soon, it may even be able to poop for you!)
Naturally, a security researcher quickly figured out how to hijack it. Patrick Wardle discovered a vulnerability in the Mac version of Muse that could allow malicious software already running on a computer to redirect the assistant’s transcription system to an attacker-controlled server.
From there, the attacker could potentially use Muse’s own permissions to do things like take photos or write malicious files… sometimes without alerting the user.
Meta says the real-world risk was relatively low because an attacker already needed malicious code running on the computer. The company quickly issued a hotfix.
I wonder if we’ll ever see a “META got it right” story.
Anthropic: We Should Slow Down AI. Also… Should We Release Another Model?
Anthropic CEO Dario Amodei recently made a very public argument that the AI industry needs to slow the pace at which increasingly powerful capabilities are released.
There’s just one small issue: Anthropic is now reportedly considering releasing another major AI model.
According to Reuters, the company is weighing a new release as OpenAI’s GPT-6 Astra gains traction with businesses and puts pressure on Anthropic’s position in the enterprise AI market.
Anthropic is reportedly still evaluating the model’s safety before deciding whether to launch it, so this isn’t quite “Forget everything we said last week.” It’s more: “no, you slow down first. Cross our hearts, we promise to do it, too… after you do.”
Slowing down sounds great right up until your competitor launches something shiny and starts stealing customers.
Apparently the AI race has discovered the same rule as highway traffic: Everyone agrees we should slow down. Just not me.
Google Wants You to Spend $899 on an AI Laptop
Google is taking another swing at premium laptops.
The company has opened pre-orders for Googlebook, a new category of laptops starting at $899 and built heavily around Gemini. Models are coming from Acer, Asus, Dell, HP, and Lenovo, with Intel or Qualcomm processors and up to 14 hours of battery life.
The big selling point is deeper integration between Gemini, Android, and ChromeOS.
Gemini can help draft and organize text, answer questions about whatever is currently on your screen, and let you continue tasks you started on your Android phone.
Basically, Google wants the laptop to feel less like a Chromebook with an AI app installed and more like a computer where AI is baked into the experience from the start.
At $899, though, Googlebook is also entering territory occupied by some very capable Windows laptops, and Apple’s recently introduced a $699 MacBook Neo.
We ain’t interested. The only AI-computer we want better look and act like R2D2.
Banks Would Like Your AI Shopping Assistant to Keep Its Hands Off Your Credit Card
AI companies increasingly want their assistants to do more than recommend what you should buy.
They want them to actually buy it for you. And big banks are getting nervous.
A group including Bank of America, NatWest, ING, Capital One, Commonwealth Bank of Australia, and ASB Bank is warning that AI shopping agents could create new risks involving scams, fraud, privacy, and payment protection.
One concern is that an AI agent could enter your card information directly into websites or choose payment methods that offer weaker consumer protections.
There’s also a surprisingly basic problem: If your AI buys the wrong thing, spends too much, or gets scammed, then who’s responsible? The bank? You? The AI? The company behind the AI?
The banks are pushing for clearer disclosure when AI agents are involved in purchases, stronger protection of customer data, and clearer rules around responsibility when something goes wrong.
And something WILL go wrong.
Suno V6 Has Two Problems: Angry Users and Another Lawsuit
Suno recently released V6, its new generation of AI music models developed using licensed music through deals with Warner Music Group, BMG, and Believe.
That was supposed to help move the company into a less legally messy era, but things are not going entirely according to plan.
Some longtime Suno users have complained that V6 sounds worse than the models it replaced, describing the results as muffled, overly compressed, generic, or less creative.
That would already make for an awkward launch.
Then Universal Music Group and Sony Music Entertainment filed another lawsuit against Suno.
The labels allege that more than 60,000 sound recordings were improperly used in connection with the development of Suno’s AI models and argue that V6 may still have benefited from information originating in older models.
Suno disputes the allegations and says V6 was trained from scratch.
So Suno managed to launch a model designed partly to move beyond its copyright problems… and almost immediately got sued over its copyright problems.