Topic starter
21/08/2026 4:54 pm
TL;DR: Trust and Safety teams should shift from banning behaviours to preventing harm, using transparent AI, human judgment, user education, and faster feedback loops, according to ActiveFence’s interview with Modulate CEO Mike Pappas. The practical lesson is that moderation policy, not just model performance, determines whether online communities stay healthy and resilient.
Editorial analysis by NHI Mgmt Group, based on content published by ActiveFence: “Expert Exchange: Encouraging Prosocial Behavior with Mike Pappas”.
Key questions
Q: How should teams design moderation policies around harm instead of banned words?
A: Start by defining the harmful outcome the platform wants to prevent, then map acceptable and unacceptable behaviour to that outcome.
Q: Why does fast feedback reduce repeat harmful behaviour online?
A: Fast feedback works because it links the action to the correction while the user still connects the two events.
Q: What do security teams get wrong about AI content moderation?
A: They often treat content moderation as a safety or policy issue instead of a control that protects identity, data, and workflow boundaries.
Practitioner guidance
- Define harm-based moderation policies Write policy boundaries around harmful outcomes, not just banned phrases or categories, and map each boundary to a response path for moderators and automated systems.
- Make AI decisions explainable to reviewers Require moderation models to surface the reason for a flag, the confidence level, and the rule triggered so human reviewers can correct edge cases consistently.
- Shorten feedback loops for repeat harm Move from delayed case handling to near-real-time correction so users receive a consequence while the behaviour is still cognitively connected to the action.
Bottom line: The article frames trust and safety as a behaviour-shaping discipline, where policy clarity matters as much as detection accuracy.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
21/08/2026 5:11 pm
Proactive moderation is a governance problem, not a classifier problem: the central decision is what kind of community the platform is trying to shape. Detection only becomes useful after the policy boundary is defined, and that boundary should be written around harm outcomes rather than abstract prohibited words. For practitioners, the lesson is that moderation effectiveness starts with governance clarity, not model tuning.
A few things that frame the scale:
- 92% of organisations expose NHIs to third parties, raising concerns about supply chain security, according to the Ultimate Guide to NHIs.
A question worth separating out:
Q: How can platforms balance user education with enforcement when harm repeats?
A: Use education for low-confidence, first-time, or ambiguous incidents, then escalate if the behaviour continues. Education helps when users have not understood the impact of their actions, while enforcement is still needed when the pattern persists. The balance is strongest when both paths are tied to the same harm-based policy.
👉 Read our full editorial: Proactive trust and safety controls can reduce repeat online harm