An AI-assisted dispute turns back on its maker

Meta has promoted AI agents as tools that can take on routine work across people’s professional and personal lives. One South Carolina business owner is now testing what happens when that promise collides with the platform’s own automated defenses.

Joe Thomas, owner of Kong Koatings Epoxy Countertops, says Meta suspended Facebook pages used by his business after he granted access to a Meta Muse AI agent he calls “Baby Mia.” The suspension cut off a marketing channel Thomas says is central to his business, and he has filed a lawsuit in Oconee County Court seeking $75,000 in damages.

The case is notable less because an AI system is a party to the lawsuit—it is not—than because Thomas says he relied on Meta’s own agent to help organize and pursue his complaint against Meta. The episode illustrates a growing friction in agentic software: services are encouraging automation while their security and moderation systems remain designed to treat unfamiliar automated activity as potentially hostile.

A business built around Facebook marketing

Thomas told 404 Media that he uses AI throughout his business, including for emails, business operations, and creating videos for several Facebook pages. The videos include AI-generated versions of Thomas and his wife demonstrating epoxy-floor work.

According to Thomas, Meta’s content-moderation systems disabled the pages after he allowed Baby Mia to access them because the activity appeared to resemble that of a hacker. He said the accounts were unavailable for about a week.

For a business that depends on Facebook advertising for inbound leads, that interruption had immediate commercial consequences. Thomas claims he lost at least $25,000 worth of work while the page was unavailable. The complaint seeks $75,000.

Those figures are allegations from the lawsuit and Thomas’s account, rather than an independently established measure of lost revenue. Still, the dispute points to a practical issue for small firms experimenting with AI agents: the value of automating work can depend on access to platforms that may not yet reliably distinguish authorized agent activity from account compromise or prohibited automation.

The agent as an advocate

Thomas said Baby Mia helped send messages to elected officials, including Senator Tim Scott, and prepared the lawsuit-related materials. He also said the agent sent a fax to President Donald Trump’s office. The account gives an unusually concrete example of a user treating an AI agent not just as a writing assistant but as an organizer for communications, documentation, and escalation.

A Man and His Muse AI Agent Sue Meta
Joe Thomas and his wife in an AI-generated video posted to his Facebook page

That distinction matters. Generative AI has commonly been used to draft messages, summarize records, and suggest next steps. An agent with access to an account can potentially go further, carrying out actions across services. That greater autonomy can make a tool more useful, but it also creates harder questions about authorization, identity, audit trails, and responsibility when an action triggers a platform’s safeguards.

Meta’s systems appear, in Thomas’s telling, to have encountered precisely that ambiguity. The user says the agent was acting with permission; the platform’s filters allegedly interpreted its behavior as a threat. Neither proposition is inherently surprising. Security systems are intended to flag anomalous account behavior, while AI agents are designed to perform actions that can look anomalous when compared with conventional user activity.

A broader test for AI-agent platforms

The lawsuit arrives as major technology companies increasingly pitch AI agents as systems that can work continuously on behalf of users. Meta CEO Mark Zuckerberg described a vision earlier this year in which people would use agents to help with relationships, health, careers, finances, home management, and hobbies.

Thomas’s experience exposes the operational gap between that vision and the existing rules of online platforms. An agent that drafts content is relatively easy to contain. An agent that accesses business pages, posts material, manages communications, or attempts to navigate support processes must operate inside systems built to detect fraud, spam, takeover attempts, and scripted abuse.

For users, the immediate lesson is that granting an AI system access to a business account may introduce a new category of operational risk. A platform may restrict access even when the account owner believes the automation is legitimate. For companies offering agents, the episode raises pressure to make permissions, logs, and user control clearer enough that platforms can recognize authorized activity without weakening security.

The legal case will not by itself settle those larger questions. But it offers an early, vivid example of a contradiction that may become more common: a company invites users to delegate work to AI, then its own enforcement systems may penalize the resulting behavior.

What comes next

The lawsuit was filed Monday in Oconee County Court, according to court records obtained by 404 Media. Its outcome remains uncertain, and the supplied reporting does not include Meta’s response to the allegations.

Whatever the result, the case makes clear that AI agents are moving from demonstration tools into everyday business workflows. As they do, disputes may increasingly turn on a basic question: when software acts on a person’s behalf, how should platforms tell the difference between authorized assistance and suspicious automation?

This article is based on reporting by 404 Media. Read the original article.

Originally published on 404media.co