Meta expands Muse AI agent to small businesses with new tools

By Splitfeed EditorialLast updated Oct 1, 2026, 04:10 UTC

On September 29, 2026, Meta launched Muse for Small Business, an extension of its Muse AI agent designed to handle operational tasks for small business owners. The rollout came with new app integrations and subscription tiers, and Meta stated that the agent will only execute business tasks after receiving explicit owner approval. Mark Zuckerberg has framed the expansion as part of a broader effort to move Meta's revenue base beyond advertising and social media into enterprise AI services. A day later, on September 30, reports emerged describing backlash over alleged privacy violations and security flaws in the product.

The launch carries weight beyond the product itself because of two specific details that surfaced shortly after the announcement. Meta's terms of service place full financial responsibility for any purchases Muse makes on the user, not on Meta, a point Yahoo Finance highlighted from the fine print. Separately, Benzinga noted that the agent could charge users fees in the course of saving them money, citing statements from a Meta president. Together these details raise a real dispute about whether the approval-before-action safeguard is a meaningful protection or a limited one that still leaves users exposed to financial and privacy risk.

The anti-AI camp has been the most vocal so far. Voices in that group argue that Meta treats users primarily as data sources, with SaucyWrong stating that the user is nothing but a mark to Meta. dvirdaniel3141 raised the concern that people will click allow all just to get the task done, suggesting consent mechanisms will be bypassed in practice. cleandreams questioned whether reinforcement learning with human feedback is an adequate foundation for guardrails, and SamInTheShell argued that the legislative pressure to force responsible behavior simply does not exist yet. The middle ground camp is focused more narrowly on the reliability of the reported security claims, with mock-possum asking whether alleged access violations were reproduced or recorded and whether protections are enforced at the OS level or only the app level. jkingsman called one reported breach scenario incredibly improbable. Ryan Daws raised a separate forward-looking question about whether AI connectors inside chat interfaces will displace standalone workflow tools for small teams. The pro-AI camp has not published reactions to this story yet, though boosters of agentic AI products typically argue that approval-gated automation raises productivity without meaningful added risk.

The disputes most worth tracking are whether the reported security and privacy issues are confirmed and reproduced independently, and whether regulators in any jurisdiction respond to the fine print around financial liability. Any legislative proposal or enforcement action targeting AI agent liability terms would directly test the durability of Meta's current user-bears-responsibility model.