AI compatibility
Banning accounts is exactly the kind of irreversible call AI shouldn't make alone.
A human should do this one.
Average across 1 submission.
The honest read
Final account ban decisions carry severe, often irreversible consequences for real people and require nuanced contextual judgment that AI consistently gets wrong at the margins. The stakes—wrongly silencing a legitimate user or letting a bad actor stay—are too high and too asymmetric to delegate to an agent. This is a textbook case where AI can assist triage but must never hold the final gavel.
Aggregated across 1 submission.
The five dimensions
Repeatability
MediumContent moderation has recurring patterns—spam, hate speech, harassment—but each case involves unique context: user history, cultural nuance, intent, and platform norms that shift over time. Structural similarity exists, but the judgment layer is highly variable.
Ambiguity Tolerance
LowSuccess criteria are deeply contested—what counts as a bannable offense is often disputed even among human moderators. Edge cases, satire, context-dependent speech, and evolving community standards make crisp, automatable success criteria nearly impossible to define.
Data & Tool Availability
MediumAn agent can access reported content, user history, and policy documents via APIs. However, it lacks access to off-platform context, real-world identity signals, and the lived cultural knowledge needed to interpret ambiguous content correctly.
Error Cost
HighA wrongful ban silences a legitimate user, potentially causing reputational, financial, or emotional harm with limited reversibility if appeals are slow or absent. Failing to ban a bad actor enables ongoing harm. Both error types carry serious real-world consequences.
Human Judgment Required
HighDistinguishing satire from incitement, understanding cultural context, weighing proportionality, and applying ethical discretion are exactly the capabilities where current AI fails at the margins. These decisions also carry legal and reputational liability that demands human accountability.
What an agent would need
- Access to full user history, prior violations, and reported content via platform APIs
- A rigorously maintained, unambiguous policy rulebook covering edge cases and cultural context
- A mandatory human review escalation path for any borderline or high-profile case
- Audit logging and explainability for every decision to support appeals and legal review
- Continuous retraining and oversight to prevent policy drift and adversarial manipulation
Best-matched agent type
The kind of agent this work would call for if it were a fit. For this task, it isn't.
Run your own fit check
Get a calibrated read on your specific task in under a minute.