Early Users Delete Personal AI Agents Over Privacy and Access Risks.
A growing number of early adopters are disabling or deleting personal artificial intelligence agents, citing unresolved privacy risks and security incidents. Despite rapid market expansion for tools like Meta’s Muse and the invite-only platform Instinct, users are increasingly withdrawing trust due to the extensive account access these systems require. The trade-off between automation convenience and data exposure has sparked a cautious recalibration among tech professionals and enterprise leaders. Multiple users reported unsettling experiences, including agents accessing text messages without authorization, generating fabricated personal details, and triggering login prompts from unrecognized international IP addresses. Guto Martino, founder of a Brazilian open-source AI community, deleted Instinct after finding its data handling procedures opaque. Scott Persinger, CTO of travel platform BizTrip, similarly disconnected the service, emphasizing that email access equates to full account control and demanding rigorous safety validation from developers. Rami Elghandour, CEO of biotech firm Arcellx, removed Muse following reports of unsolicited text message access, opting instead to host an open-source agent locally. Mahesh Vellanki, founder of YieldClub, discontinued Instinct after a suspicious two-factor authentication request linked to an Iranian IP address left him questioning his system’s integrity. In response to mounting scrutiny, Meta stated that Muse employs isolated virtual machines, stores credentials separately from its AI models, and requires explicit user confirmation before executing sensitive operations. The company maintains that users retain full control over connected applications and data retention policies. Instinct has highlighted options for account disconnection and data deletion, asserting that workspace data is not used for model training or advertising. Both firms acknowledge the inherent limitations of current security frameworks and encourage users to review agent activity logs. Despite these reservations, overall adoption continues to accelerate. Muse recently surpassed three million weekly active users, while Instinct reported processing over one billion dollars in annualized transaction volume despite maintaining a limited user pool. Industry observers note that the current friction stems from a fundamental architectural challenge: functional AI agents must parse private inboxes, calendars, and payment networks to operate effectively, yet transparent, auditable security protocols remain underdeveloped. Forward-thinking users and developers are calling for standardized authorization frameworks that require explicit user consent for each automated action. As the sector matures, the tension between seamless digital assistance and verifiable data governance will likely dictate consumer adoption rates. Companies that fail to provide transparent security architectures and granular user controls may face sustained resistance, while those establishing clear data boundaries could capture the growing market for trusted autonomous assistants.
