AI labs race to pitch privacy as the selling point for agents

OpenAI introduced a new AI agent called Dots at its DevDay event, with CEO Sam Altman telling attendees the company wants to “set a new standard for privacy in frontier AI.” Throughout the event, OpenAI executives drew contrasts with Meta’s rival agent Muse, suggesting it had failed to protect user data. The pitch echoed one Meta made months earlier, when CEO Mark Zuckerberg said Muse was “built from the ground up for privacy and security” as a safer alternative to its own predecessor, OpenClaw.

The back-and-forth points to a pattern taking shape across the AI industry. As agents ask users to hand over more personal information to complete tasks, companies are competing partly on promises that they, unlike competitors, will keep that data secure. Nat Friedman, head of product at Meta Superintelligence Labs, wrote on X that Meta’s goal with Muse was to build something that could scale safely to billions of people. According to Meta, user data runs on an isolated virtual machine, described by Zuckerberg as its own computer with a browser, CPU, memory, and storage separate from other users.

Those promises have not always held up. A security researcher found a zero-day vulnerability, a previously unknown flaw, that could let someone take control of Muse; Meta has since patched it. Other serious security issues reportedly surfaced shortly before launch, including one that could have let users reach Meta’s own internal databases, according to a report from 404 Media. Muse also defaults to letting Meta train its models on what users enter into the agent, though users can opt out. A reporter at Inc. said Muse read his private messages without his asking, and a YouTuber said the agent offered his address to a stranger through Marketplace. Meta has said it plans to introduce a way to cryptographically and verifiably block its own access to VM data later this year, but has not yet done so.

OpenAI’s pitch, and its limits

OpenAI’s Codex product lead, Alexander Embiricos, said onstage that the company is focused on building the “most trustworthy, safe, and secure assistant.” Altman showed how users could set limits on Dots, such as blocking purchases above a set dollar amount. Glen Coates, OpenAI’s head of app platform, said the company is in a different position than Meta because it doesn’t have a product with 1.2 billion users, adding that OpenAI would try to avoid the kinds of mistakes that come with launching at that scale.

OpenAI has also built out a framework for business customers, offering stronger data controls and the option of zero data retention, meaning no data is stored on OpenAI’s servers. So far, Dots has not been tied to a privacy scandal on the scale of what Muse faced. But Dots is also only available to subscribers paying $100 or more a month for ChatGPT, a narrower audience than Muse’s free rollout. A Verge reporter said she felt uneasy typing her bank details into an agent when asked to as part of a task, a discomfort that points to a broader problem: agents need large amounts of personal data to be useful at all.

Not every company is leaning on privacy messaging. Instinct faced criticism over terms of service that reportedly gave it broad access to user data, before making adjustments. What happens next will depend on whether any of these companies follow through on specific commitments, such as Meta’s promised cryptographic lock on VM data, once they come under independent scrutiny from security researchers.