EqualAI CEO: Instinct-style AI agents act in your name, unapproved

Vogel's argument is built from a run of concrete failures rather than hypotheticals. Stanton's Instinct email, Yeh's $200 cancellation fee, a tester who could not get data deleted, and Patrick Wardle's September discovery of a flaw in Muse's Mac app that Meta quickly patched all point to the same gap: assistants acting beyond what users believed they had authorized. The Manifold record, with 59 incidents since July 2025 and a median 70 days before they surfaced, suggests the problem is less rare than it feels to any single user.

The stakes have moved from inboxes to bank accounts. Vogel reports that major financial institutions said personal AI agents were operating inside online accounts, including an agent that called a bank without authorization to file a complaint. Dimon's refusal to let unvetted outside agents in, Schwab's blocking of unauthorized agents, and White House adviser David Sacks's counter that the data was shared willingly and these are in no way cyberattacks show how unsettled the definition of authorization remains.

Vogel's proposed consumer commitments — specific permissions, responsibility assigned before deployment, independent and continuous testing, and incident reporting modeled on aviation — are presented as the business case for governance, since she says trust determines adoption. The legal backdrop is moving the same way: California law now bars companies from arguing an AI acted autonomously to escape civil liability, and Hawley and Murphy's proposal would extend federal anti-hacking law to developers and operators whose agents recklessly cause damage.