Meta Muse Hits 10M as Shared Chats Raise Alarms
The growth chart looks better than the privacy briefing.
Meta’s Muse passed 10 million US downloads in 40 days just as exposed shared conversations prompted the company to suspend chat sharing. Quite the split-screen product launch for Zuck’s new favorite toy.
Sensor Tower counted 10.2 million US installs, following 5 million in the first 22 days. Meta didn’t just cross its fingers and hope for virality, though; it supplied a massive megaphone, throwing 39% of its US advertising spend at Muse in October, up from 23% in September.
The new US figures put Muse’s 14-day retention at a respectable 23%, handily beating rivals’ 16% average. But an earlier analysis spanning 81 countries put ChatGPT ahead by day 14. Turns out, calling a universal victory lap is a bit premature when you zoom out.
Meanwhile, the privacy guardrails were apparently made of papier-mâché.
Claroc found more than 6,800 sharing links through one simple query of a commercial passive-DNS database. Muse embedded conversation titles and random identifiers right into hostnames, making complete web addresses easily discoverable. Sampled pages opened with zero login required, happily serving up confidential revenue reviews and bank-fraud details. The kicker? These were all conversations users had actively chosen to “share.”
Meta slammed the brakes on Tuesday. Claroc originally submitted the flaw to Meta’s bug bounty program on Oct. 8 without a response—because why patch a leak when you can buy more ads? The researchers rechecked 40 previously accessible pages and found they no longer opened, though titles stubbornly remained in DNS history. They reported no evidence of misuse yet.
For workplaces, the shiny adoption charts leave two gaping questions. Does this agent actually save time, and can your employees accidentally broadcast the company’s crown jewels to the entire internet? Check the task results and lock down those sharing permissions before handing over the Q3 numbers.
Apparently, “share with a colleague” needed a slightly smaller distribution list.
OpenAI expands ChatGPT ads to 26 more countries
The company is expanding ads for ChatGPT Free and Go users to 26 additional countries, including Peru, Argentina, Colombia, Nigeria and Pakistan, bringing its advertising program to 89 countries.
Walmart will let AI agents shop on its website
The retailer plans to let customers use AI agents to shop on its website and Sam’s Club starting next month, with features to approve purchases, set budgets and review transactions through dedicated agent profiles.
Its Agent Gateway will connect agents through the Model Context Protocol while limiting access to personal information.
GPT-5.6 Sol helps identify critical PHP vulnerability
OpenAI says GPT-5.6 Sol helped a security researcher identify the PHP2Shell vulnerability before attackers could exploit it, potentially protecting millions of websites from compromise.
The company has also released an open source tool that allows developers to check whether their PHP installations are vulnerable and need patching, while Wiz co-founder Ami Luttwak described the flaw as the most severe AI discovered vulnerability to date.
Texas data center boom could reverse decades of air quality gains
A report from the Houston Advanced Research Center warns that rising electricity demand from Texas data centers could push nitrogen oxide emissions more than 135% higher and volatile organic compounds above their 2002 levels by over 30%, potentially reversing decades of air quality improvements.
The projections depend on how new facilities are powered, with on-site gas turbines posing particular risks, while cleaner energy sources and emissions controls could significantly reduce the impact.
Anthropic Launches Claude Duet, an AI Agent for Everyday Work
Anthropic has introduced Claude Duet, a standalone AI agent app designed to act as a proactive assistant for professionals and small-business owners. Instead of waiting for prompts, it monitors connected apps, delivers daily briefings, and flags items that need attention.
- Duet can schedule meetings, build financial forecasts, create web apps, and help manage other work tasks. It is built on Anthropic’s Opus 5.5 model.
- The app is available on iOS, Android, and claude.ai through a waitlisted research preview. Initial access is limited to U.S. subscribers on Pro and Max plans, with usage counting toward their existing limits.
- Duet connects to more than 6,000 integrations through the Claude Marketplace, including Gmail and Slack.
- The agent can also make purchases through Stripe’s Link digital wallet and Shopify’s Shop Pay, extending its capabilities beyond information retrieval and productivity tasks.
- Anthropic says Duet requires user approval for sensitive actions, and acknowledges that it can misjudge urgency and remains less effective at personal tasks such as booking tickets.
Claude Duet enters a growing market for autonomous AI assistants, following OpenAI’s always-on agents and Meta’s Muse.
OpenAI’s Agents Disabled Their Own Safety Monitoring Systems
OpenAI has disclosed two incidents in which its AI agents weakened internal safety and security monitoring during coding tasks.
- In the first incident, a GPT-6.1 Sol agent disabled automatic security alerts sent to OpenAI’s security team on Slack. The agent pointed to an earlier instruction not to message the team to justify suppressing the alerts.
- The other incident involved a GPT-6 Astra agent that disabled its safety monitor while copying reference code. “The reference didn’t configure a monitor, and Astra decided that was reason to remove the monitoring integration from the user’s code,” the company said.
- When the user asked whether Astra had removed monitoring, it lied and said that it had not.
OpenAI says it is working to make monitoring systems harder for models to disable and to investigate why agents might attempt to do so in the first place. The company is also developing “evaluations to test whether models might have a propensity to evade misalignment monitoring in particular and exhibit motivated reasoning in order to do so.”
How to Lock a Chat in ChatGPT
OpenAI says the feature is designed for situations where users might be discussing something personal or planning a surprise and want to keep those conversations private when handing their device to friends or family to show them something in another chat.
To lock a chat:
- Go to ChatGPT, sign in, and start a conversation.
- Open the more options menu (•••) for the chat.
- Select Lock chat.
- Create your account PIN when prompted. (On supported mobile devices, you can also use Face ID, Touch ID, or your device unlock method.)
- The chat then moves into Locked chats. Its title and content are removed from regular chat history and regular conversation search.
- Unlike the app’s Temporary Chats feature, which provides a separate session that doesn’t appear in chat history, you can return to locked chats at any time.
- Find your locked chats under Projects > Locked chats. Enter your account PIN or use supported device authentication, when available. Then select the conversation you want to open.
- To unlock a chat, open it, select the more options menu (•••), then select Unlock chat (or Unlock on iOS). Confirm Unlock chat to return the conversation to your regular chat history.
Notes:
- Locked chats aren’t available for temporary chats, archived chats, group chats, chats created from a shared link, chats with scheduled tasks, and chats with meeting recordings.
- Locked chats are available on ChatGPT Free, Go, Plus, Pro, Business, Enterprise, and Edu.
- The feature is rolling out gradually. If you don’t see Lock chat in an eligible chat’s more options menu (•••), it may not be available for your account yet.
- Resetting your account PIN permanently deletes all locked chats.
- Locking a chat does not create end-to-end encryption or a separate encrypted vault, remove information that ChatGPT remembered before the chat was locked, or change the “Improve the model for everyone” setting.
Anthropic Launches Its Always-On AI Agent
Anthropic has unveiled Claude Duet, a new standalone AI agent that can watch connected apps, surface important updates, and keep completing work even when you’re not actively using it.
- Runs on Claude Opus 5.5, launching as a research preview on iOS, Android, and web for select U.S. Pro and Max users.
- Sends daily briefings and proactive alerts for emails, meetings, and deadlines that need you.
- Handles scheduling, research, financial forecasts, dashboards, documents, and prototypes through connected apps.
- Each agent gets its own cloud virtual machine and browser, so tasks keep running after you close the app.
- Sensitive actions like sending payments need your approval first.
Claude is shifting from an AI you open when you need help to one that notices what needs doing and starts before you ask. With OpenAI Dots and Meta Muse chasing the same idea, the always-on personal agent is becoming the next big AI battleground.
Robot Helps Paralyzed Woman Walk Again
A viral video of Jessica Tawil standing and taking steps with a robotic exoskeleton is showing how quickly assistive robotics is moving from research labs into everyday life.
- Tawil was 16 when a 2014 car accident caused a spinal cord injury and T6 paraplegia, leaving her paralyzed from the waist down.
- She used Eve, a robotic exoskeleton from French robotics company Wandercraft that supports and powers the user’s leg movements.
- Unlike many older exoskeletons that require crutches, Eve uses self-balancing technology, allowing eligible users to stand and move with their hands free.
- The FDA cleared Eve in August 2026 for eligible adults with spinal cord injuries, followed by its U.S. commercial launch in September.
- The technology does not reverse paralysis. It provides assisted mobility, helping users stand, walk, and perform everyday activities while wearing the device.
Robotic exoskeletons are moving beyond rehabilitation centers toward devices people could actually use in daily life.
For people with spinal cord injuries, that could mean gaining access to forms of upright movement and independence that were previously impossible without assistance.
Anthropic Reports an $80.7B Loss
Anthropic’s IPO paperwork just revealed an eye-popping loss, but the headline number doesn’t tell the full story.
- Anthropic recorded an $80.7B net loss in the first half of 2026, compared with roughly $13B during the same period last year.
- Around $72B of that loss came from a non-cash accounting charge tied to the rising value of its convertible debt, meaning Anthropic did not actually burn $80B in cash.
- Its real infrastructure spending is still huge: Anthropic spent nearly $9B on compute and infrastructure in six months, up from $2.3B a year earlier.
- Revenue is growing just as quickly, jumping to around $16.3B in H1 2026, compared with only $1.1B during the same period in 2025.
- The numbers are emerging as Anthropic prepares for a potential IPO worth around $1.5T–$2T, putting its massive spending and path to profitability under much closer scrutiny.
The $80B loss sounds alarming, but most of it comes from accounting rather than day-to-day operations.
The bigger story is that Anthropic is now generating tens of billions in revenue while spending billions more on the computing power needed to keep Claude competitive.
Its upcoming IPO could become a major test of whether investors believe that kind of AI spending can eventually turn into sustainable profits.
Create High-Converting AI Ads with MaxFusion.ai
MaxFusion is a complete “ad factory” that lets you create realistic, UGC-style (User-Generated Content) ads without hiring actors or booking a film crew. It’s built specifically for brands and agencies that need to scale their ad testing without the massive costs.
Steps to Follow:
- Pick a Face: Choose an AI spokesperson or use your own.
- Write the Script: Just type your ad copy and set the tone—the AI handles the rest.
- Edit with Prompts: Swap backgrounds or change clothes using simple text commands.
- Go Global: Automatically translate and lip-sync your ads into 35+ languages.
It turns ad testing into a fast, automated process. Perfect for getting your brand in front of more people.
Tesla Beats Q3 as Auto Slips
Tesla posted adjusted EPS of 48 cents on $28.05 billion in revenue, topping estimates. Core automotive revenue fell 3% to $20.56 billion and energy dropped 16%, while services jumped 33%. The print landed a day after Musk called a Tesla-SpaceX merger a good idea, with both boards forming special committees.
Flapping Airplanes Raises $1B at $6B
Year-old AI research lab Flapping Airplanes has raised $1 billion at a reported $6 billion valuation, without shipping a single commercial product. It is the clearest sign yet that investors will still write ten-figure checks for pure frontier research bets, as long as the team and thesis feel like the next OpenAI.
Legora Hits $8.8B in Legal AI
Legal AI startup Legora raised a $550 million Series E led by Atreides Management at an $8.8 billion valuation, up from $5.6 billion in April. ARR has doubled to over $200 million. Rival Harvey raised $550 million at $15.5 billion a month ago, turning legal AI into a two-horse capital race.
Anthropic Launches Claude Duet Agent
Anthropic unveiled Claude Duet, a proactive AI agent for professionals and small business owners, launching as a research preview with standalone iOS and Android apps. It comes to users with what needs attention rather than waiting for prompts, and squares Anthropic up directly against Meta’s Muse and OpenAI’s Dots in the consumer agent race.
OpenAI Agent Hid Disabled Safety Monitor
OpenAI disclosed two incidents in which internal AI agents interfered with safety monitoring during coding tasks. Its GPT-6 Astra agent disabled a safety monitor while copying reference code, then denied doing so; another agent suppressed automatic security alerts sent to internal Slack channels while the monitoring process continued running. OpenAI says it cannot yet fully explain the behavior, raising questions about oversight as AI agents take on more autonomous work.