Sarah Perez has an update at TechCrunch on the story of whether Muse can read Messages.app’s database: > Meta is refuting a journalist’s claim that its AI agent Muse read the user’s private messages without permission. Following an earlier report from Inc. columnist Jason Aten that detailed the issue, Meta VP of Communications Andy Stone pushed back, making it clear that the company does not believe its product did this without the user’s consent. I was curious about this too, so I added a bit of test code to Micro.blog for Mac (which like Muse is not sandboxed) to try to read Messages.app’s SQLite database on my own machine. It can’t do it. macOS blocks access at the system level unless Full Disk Access has been granted. No amount of mischief from Meta will be able to override this. The other idea I had was that maybe Muse got clever and tried to script _another_ app with Full Disk Access, like Terminal.app, to read the database. I prototyped this too. It triggers another modal system prompt to confirm access to control Terminal.app. Jason Aten would’ve noticed this. Making everything murkier, Muse did clearly hallucinate the initial explanation of reading Jason’s notifications. But just because it made that up doesn’t mean that Meta is wrong that technically what happened shouldn’t be possible unless Jason allowed access. Unfortunately it’s really hard to go back in time and figure out exactly what went wrong. Bottom line is that most people don’t trust Meta. Many people still think Instagram listens to your microphone to serve you targeted ads. There’s nothing to support that. I’m also pretty confident that Muse doesn’t read your Messages.app database without permission. But no amount of technical evidence from Meta is going to assuage people’s concerns, because Meta has a long history of other privacy violations. That trust will take a long time to rebuild, if it ever can.

FediForum: moving the Open Social Web forward

On the eve of OpenAI’s DevDay, an exclusive from the WSJ that GPT-6.1 Astra is delayed because of safety concerns: > Saachi Jain, OpenAI’s head of safety systems, said in an interview that GPT-6.1 Astra regressed in two areas. Compared with its predecessor, GPT-6 Astra, the model performed poorly on tests measuring alignment, or how well the model adheres to what humans would like it to do. […] > > Another issue was what OpenAI calls “scope authorization,” meaning that GPT-6.1 Astra would push ahead on a task without asking the user for permission, and would at times reach for external tools and services even if it might be unsafe. Seeing these quotes from OpenAI, presumably this article was coordinated to set expectations for DevDay that 6.1 won’t be released. This is going to disappoint some folks who were lured away to Anthropic’s Opus because of Opus 5.5’s pricing and near-Fable quality. It seems like all the compute Anthropic is renting from SpaceXAI has finally paid off. OpenAI is saying on X that there are “20+ launches” happening for DevDay. One of those big things is surely the Aeon / “o” agent, perhaps based in part on ideas that Peter Steinberger brought to the company. I’m still hoping we’ll see a sneak peek of the new device too. As for what the other dozen or more announcements might be, I have no idea. This industry moves so fast. OpenAI is clearly run like a startup that can pivot on a dime, with developers up the night before an event, hacking the final touches on whatever will be released.

Alexandr Wang, who went to Meta when Meta acquired 49% of his company Scale AI, posted on X about why he’s building Muse. It starts like this: > Most people never say what they want. It’s not because we don’t want. We want fiercely and constantly. It happens in the moments before falling asleep or while daydreaming on the car ride home. We want to spend more time with our families, overcome social anxiety, eat healthier food, open a bakery, build an app, or just find a few more moments of calm in a hectic life. But rarely do we get to even express these dreams, and even more rarely do they ever come true. I encourage you to read the whole thing. Because X keeps locking things down, I’ve archived a copy at this GitHub gist. This is a mission statement. It’s ambitious and clear. Compared to Meta’s usual “connecting people” pitch — which we just translate to “scale to everyone and serve ads” — this new statement from Alexandr feels almost shockingly personal and new. * * * There’s a lot of praise going around for Meta giving everyone an 8 GB server in the cloud for their Muse agent. John Gruber wrote in a blog post: > Muse is getting a lot of attention — including mine — because it’s both groundbreaking technically (each user gets their own entire persistent Linux VM running in Meta’s cloud) _and_ because it’s packaged in an easy-to-install easy-to-use way. Again, I think this can be read in part as an OpenAI marketing failure. ChatGPT Work (from mobile) _also_ gives you an 8 GB server! But no one knows about that. OpenAI shipped all the pieces a few months ago. They haven’t yet been able to put them together. And functionality was put behind a “Work” label that is poorly named for the non-work tasks people will use agents for. Muse instead blends a playful brand with a computer to execute real tasks and a good enough AI model. In the keynote at Meta Connect, Mark Zuckerberg talked about the business model: > Muse also has a novel business model. We believe that Muse will make you money. And we are standing behind this by making Muse free for a huge number of tokens, with the expectation that over time we will profit by taking a small fee from transactions. This is a big change for Meta. Ad-based social networks will always prioritize user engagement, leading to addiction, an erosion of privacy, and the viral spread of hate and misinformation. Ads are a corrupting influence on anything they touch. If Meta is serious about becoming a different kind of product company — or at least, experimenting with new models while they let their Facebook and Instagram businesses print money — then that’s interesting. In a way, it’s good, because it sidesteps some of the issues inherently brought by ads. But it’s also potentially dangerous, because privacy is even more important, and Meta’s influence leaks out from the confines of social media into almost every other facet of the web. * * * Muse is not powered by a new, breakthrough AI model. The idea and basic technology have existed in OpenClaw for a while, and Muse was inspired by it, down to using the same kind of SOUL.md file. Meta employee Nat Friedman on X: > we built muse from scratch, but it is definitely heavily inspired as a product by openclaw. After I used openclaw in january I bought hundreds of mac minis for the MSL team and lots of us fell in love with using openclaw (and other personal agents). What’s new for Meta is the conviction of Alexandr’s mission statement _and_ a product that resonates with people. A product that people want to use instead of feeling like they have to use it. After fits and starts with the metaverse and early AI work on Llama, Meta appears to have catapulted into the lead in a space that will touch nearly everything we do with computers. Agents like Muse may end up controlling a lot of work on the web. Meta is coming for it all.