Summary

  • An October exchange on AFRINIC's Resource Policy Discussion list moved from confusing email replies to proposals for detecting AI-assisted posts and restricting contributors.
  • The cited posts record disagreement, not an adopted AI rule. AFRINIC's published Code of Conduct gives the PDWG co-chairs responsibility for enforcement on the list.

The defence of the digest arrived under a digest subject line. On 14 October, Zamaphathwa Phathwa posted an appeal to distinguish useful writing assistance from automated participation. Its heading in AFRINIC's archive was still RPD Digest, Vol 225, Issue 36.

Two replies brought the discussion straight back to email etiquette. Ben Roberts asked for a subject line matching the topic being answered. Hendrik Visage also pointed to the reply identifiers in message headers. For readers trying to follow one argument, a reply addressed to a whole bundle of messages was precisely the problem.

The awkward juxtaposition explains why this dispute was difficult to settle. One side was talking about keeping a conversation readable. Another was asking what happens when suspicion about the writing tool becomes suspicion about the person. Both questions deserved an answer; neither answered the other.

One inbox setting, two arguments

RPD is where participants discuss Internet number-resource allocation policy in Africa. Its subscription page offers a daily digest: several messages delivered together instead of separately. That can suit someone who reads the discussion in batches. It becomes troublesome when a reply no longer identifies the particular message it addresses.

The exchange had begun on 8 October, when Benson Muite asked whether a different discussion tool might make participation easier. By 12 October, the argument included claims that people were feeding digests into AI systems. Those claims appeared in participants' emails; the posts supplied no test establishing the authorship of a particular contribution.

Sylvain BAYA offered a more elaborate possible procedure: hold digest replies, check them with a script, send posting guidance and potentially remove posting rights after repeated behaviour. He also defended legitimate uses of digests. This was a proposal inside an argument, not an AFRINIC enforcement announcement.

Phathwa's response emphasised human responsibility for submissions and warned against relying solely on automated detection. A person might use a tool for language or accessibility without handing it control of the discussion. The immediate replies, meanwhile, showed that a thoughtful argument about participation could still leave its readers asking which thread it belonged in.

That is a useful distinction to retain. A confusing subject line is visible in the record. It can be corrected without deciding whether software drafted the paragraphs underneath it.

A rule needs more than a suspicion

AFRINIC already publishes a Code of Conduct. On RPD, enforcement belongs to the Policy Development Working Group's co-chairs. The warning process escalates from a first warning to moderated posting for no more than three months, then removal for three months. A fourth breach may lead to permanent removal. Warnings remain valid for no more than twelve months.

Those are consequential powers: they determine whether a contribution reaches the discussion at all. The published page does not set out an AI-specific authorship test. Nor do the cited October messages announce that the proposed script was deployed or that an AI rule was adopted.

The question for anyone seeking action is therefore concrete: which act breached which rule? A misplaced reply, harassment and impersonation are not interchangeable complaints. If the problem is repeated disruption, the record needs to show that disruption. Calling a post machine-written does not by itself supply the missing connection.

There is a practical way to improve the discussion without resolving every argument about AI. Give digest readers usable reply instructions. Keep topics identifiable. Ask contributors to answer objections to what they have submitted. Where a conduct complaint follows, name the rule and the evidence, and give the person an opportunity to respond. These are recommendations, not a claim that a new procedure has already been adopted.

An email subject line will not settle who wrote the email. It can, however, help everyone find the argument worth answering. AFRINIC's participants can fix that small problem without turning it into a verdict on the larger one.

Correction, 24 October 2026: the earlier version described the reply guidance as coming from a list administrator. The cited messages identify Ben Roberts and Hendrik Visage; the header guidance appears in Visage's reply. We have corrected the attribution and limited statements about the absence of an AI decision to the documents examined.