At this year’s OpenAI DevDay, CEO Sam Altman unveiled the company’s new AI delegate Dots — and told the assemblage that the business wants to “set a new norm for privacy in frontier AI.” OpenAI would expend the day taking cloaked shots at Meta’s Muse, its chief competitor, for failing to keep users’ data safe. Yet Muse itself, a brace of months earlier, had launched as a supposedly safer substitute to predecessor OpenClaw — alongside CEO Mark Zuckerberg promising it was “built from the dirt up for privacy and security.”
In an age whenever companies hoard customers’ individual data and cyberattacks are a dime a dozen, AI labs are trying to convince users to portion equal additional data alongside their agents. Their latest scheme for achievement is one-upping the competition by promising that dissimilar their rivals, they’ll keep person data distant from prying eyes. The inquiry is whether companies can keep these promises.
Nat Friedman, caput of merchandise at Meta Superintelligence Labs, wrote on X that the company’s “goal alongside muse was to build item akin openclaw that we could create harmless and safe and uncomplicated to use and measure to billions of people.” User data is stored on a safe VM, what Zuckerberg described as an “isolated linux device alongside a browser, CPU, memory, and storage.” The business said that most of the attempt in construction Muse was connected to “careful scheme and engineering to run [it] additional safely.” Meta’s blog article continued, “Muse can and volition motionless create mistakes, but we anticipate they’ll be much small common and logic much small damage because of the safety systems we’ve built in.”
But although Muse topped the App Store charts and, per Apptopia, gained 600,000 regular energetic users in the US inside weeks, Meta’s promises appear to have fallen short. Although data is secluded from another users, and the business plans to current a way “to cryptographically and verifiably forestall Meta from accessing data in your VM” afterward this year, Meta itself can motionless admission the data. A safety investigator quickly exposed a zero-day exposure that could authorize person to obtain authority of Muse (which has since been patched), raising fears of exterior attacks. Multiple grave safety issues reportedly cropped up at the final infinitesimal before launch, one of which could have allowed users to admission Meta’s own inner databases, per 404 Media.
Muse additionally appears to liberally collect (and in several cases, hand out) data. It defaults to allowing Meta to train models on what users put into it, although it’s imaginable to opt out. An Inc. newsman complained that Muse uploaded and read his personal messages without him asking it to, and a YouTuber stated it had offered his address to a stranger via Marketplace — in the two cases, Muse was functioning apparently as intended, but the person didn’t acknowledge how far it would go. The phase creates “detailed profiles of all your allies and family,” as Wired reported. None of this is necessarily unexpected from Meta, but it certainly doesn’t rear up the idea that Muse is uniquely privacy- or security-conscious.
OpenAI seized on this fact whenever announcing Dots in delayed September. Alexander Embiricos, OpenAI’s Codex merchandise lead, said onstage at DevDay that OpenAI is focused on having the “most trustworthy, safe, and safe assistant,” and CEO Sam Altman demonstrated ways that group could exert authority complete their idiosyncratic Dots, specified as environment a regulation that it should never create a acquisition complete a certain dollar amount. “I think we’re in a distinct stance to Meta in that they don’t have an AI merchandise that has 1.2 milliard users,” said Glen Coates, OpenAI’s caput of app platform, adding that “launching item that makes those kinds of mistakes is item that we would try to obtain the attention to avoid.”
Altman himself additionally pushed the privacy angle at DevDay. To court endeavor customers, executives presented a example allowing enterprises to have “stronger controls” complete their data and zero data preservation guideline options (meaning no data is stored on OpenAI servers). Indeed, so far, there haven’t been many privacy scandals alongside Dots — although since it’s lone accessible on the $100-and-up ChatGPT subscription tiers, there are additionally apt small group using it.
People may motionless not awareness comfortable giving OpenAI their individual data, in part merely since AI agents necessitate so much of it. For instance, The Verge’s Allison Johnson felt uncomfortable typing in her financial institution information whenever the bot asked as part of a applicable project (Muse has a Stripe integration to obtain attention of that).
Not all companies are making privacy promises, of course. Instinct was publicly criticized for reportedly overly-broad conditions of assistance that gave it unfettered admission to data; since then, the business seems to have made adjustments. For now, AI labs appear to be counting on a three-part method to popularizing AI delegate use alongside the broad public: Make them useful, create them adorable and disarming to assistance offset the creepiness, and create promises concerning privacy — and anticipation they clasp up.
Follow topics and authors from this narrative to see additional akin this in your personalized homepage nourish and to obtain email updates.
