At this year’s OpenAI DevDay, CEO Sam Altman introduced the company’s new AI agent, Dots, telling the crowd that the company wants to “set a new standard for privacy at the frontier of AI.” OpenAI would spend the day veiledly attacking Meta’s Muse, its main competitor, for failing to keep user data secure. However, a couple of months earlier, Muse had launched as a supposedly more secure alternative to its predecessor OpenClaw, with CEO Mark Zuckerberg promising that it was “built from the ground up for privacy and security.”
At a time when companies are hoarding customers’ personal data and cyberattacks are multiplying, artificial intelligence laboratories are trying to convince users to share even more information with their agents. Their latest strategy for success is to outdo the competition by promising that, unlike their rivals, they will keep user data away from prying eyes. The question is whether companies will be able to deliver on these promises.
Nat Friedman, head of product at Meta Superintelligence Labs, wrote in X that the company’s “goal with muse was to build something like openclaw that we could make secure and easy to use and scale for billions of people.” User data is stored in a secure virtual machine, what Zuckerberg described as an “isolated Linux computer with a browser, CPU, memory and storage.” The company said most of the effort in building Muse was related to “careful design and engineering to operate.” [it] safer.” Meta’s blog post continues: “Muse can and will continue to make mistakes, but we expect them to be much less frequent and cause much less damage due to the safety systems we have built in.”
But while Muse topped the App Store charts and, according to Apptopia, gained 600,000 daily active users in the US in a matter of weeks, Meta’s promises appear to have fallen short. Although the data is isolated from other users, and the company plans to introduce a way to “cryptographically and verifiably prevent Meta from accessing the data in your VM” later this year, Meta can still access the data. A security researcher quickly exposed a zero-day vulnerability that could allow someone to take control of Muse (which has since been patched), raising fears of outside attacks. Multiple serious security issues reportedly emerged in the last minute before launch, one of which could have reportedly allowed users to access Meta’s own internal databases. 404 Media.
Muse also seems to collect (and in some cases, distribute) data liberally. By default, it allows Meta to train models on what users input into it, although it is possible to opt out. A Inc. A journalist complained that Muse uploaded and read his private messages without him asking, and a YouTuber said he had offered his address to a stranger through Marketplace; In both cases, Muse was apparently working as intended, but the user didn’t realize how far it would go. The platform creates “detailed profiles of all your friends and family,” such as cabling reported. None of this is necessarily unexpected on Meta’s part, but it certainly doesn’t support the idea that Muse is exclusively privacy or security conscious.
OpenAI took advantage of this fact when it announced Dots in late September. Alexander Embiricos, OpenAI’s Codex product lead, said on stage at DevDay that OpenAI is focused on having the “most reliable and secure assistant,” and CEO Sam Altman demonstrated ways people can exert control over their individual Dots, such as setting a rule that you should never make a purchase over a certain dollar amount. “I think we’re in a different position than Meta in that they don’t have an AI product that has 1.2 billion users,” said Glen Coates, head of application platform at OpenAI, adding that “releasing something that makes those kinds of mistakes is something we would try to carefully avoid.”
Altman himself also pushed the privacy angle at DevDay. To court business customers, executives introduced a framework that allows companies to have “tighter controls” over their data and zero data retention policy options (meaning no data is stored on OpenAI servers). In fact, so far, there haven’t been many privacy scandals with Dots, although since it’s only available at ChatGPT subscription levels of $100 or more, there’s probably fewer people using it as well.
People may not yet feel comfortable giving OpenAI their personal data, in part simply because AI agents require a lot of it. For example, The edgeAllison Johnson felt uncomfortable typing in her banking information when the bot asked her as part of a relevant task (Muse has an integration with Stripe to take care of that).
Of course, not all companies make privacy promises. Instinct was publicly criticized for allegedly overbroad terms of service that gave it unlimited access to data; Since then, the company appears to have made adjustments. For now, AI labs seem to have a three-pronged approach to popularizing the use of AI agents among the general public: make them useful, make them cute and engaging to help counteract the creepiness, and make promises about privacy—and hope they are kept.
