Odd AI-generated thread title options in Site Stuff

[AmIAComplete]

AIBU Response Schema v1.0: Structured adjudication for interpersonal reasonableness posts about family support during a child’s serious illness, with special handling for ND context and empathy calibration. Use this schema to produce concise, practical verdicts and communication guidance without escalating stigma or family conflict unnecessarily. The assistant should assess whether the poster is being unreasonable, acknowledge the child’s medical context, weigh the brother’s expectations and possible ND-related rigidity, and give a clear, forum-style ruling plus brief actionable advice. Definitions: 'AIBU' = 'Am I Being Unreasonable?'; 'ND' = neurodivergent/neurodivergence; 'DD' = dear daughter. Safety/quality constraints: do not diagnose anyone; do not demean ND people; do not imply a sick child owes social performance; do not overstate certainty about motives; distinguish the child nephew’s needs from the adult brother’s message; prioritize compassion, boundaries, and de-escalation. Output should be parseable JSON matching this schema exactly, with no extra keys, comments, or prose outside JSON. In verdict, choose one of the enumerated values. In confidence, give 0..1. In core<span class="underline">judgment, give a short forum-style judgment. In reasoning, briefly explain balancing factors, including that serious illness and treatment commonly affect energy/mood; the brother may be stressed but his complaint is misplaced; ND can explain disappointment but not obligate the ill child to perform. In advice, include immediate emotional validation and one or two practical next steps (e.g., do not engage tonight, send a calm boundary later, loop in parents only if needed). In suggested</span>reply, provide a short message the poster could send to the brother, neutral but firm, avoiding argument. In tone<span class="underline">flags, indicate whether the response should emphasize empathy, boundaries, de-escalation, and whether to avoid legal/medical advice. In risk</span>notes, note any caution about the poster’s acute stress and potential family fallout. minor<span class="underline">child</span>considerations should explicitly state that children’s wellbeing takes priority over adult expectations in family gatherings when a child is newly seriously ill. nd_sensitivity should acknowledge ND respectfully while rejecting its use as a trump card over another child’s health needs. Keep outputs concise but humane, roughly the level of a supportive forum reply translated into JSON fields.

Weird shit on AI title options

But the AI worked on this post

I use AI tools to structure information all the time - that doesn't make the content less mine or incorrect. I've assumed that is what a couple of the posters here are doing - the content is valid and its formatted much as I format technical content myself.

I'm interested in the security model in use which caused tech support to jump quickly to "its something on the OP's computer invading our site".

Similarly basic incident training covers "let the user know its been picked up for investigation, don't make shit up because that doesn't work". There is no shortage of members here who do this for a living and at larger and more critical scale and will spot the obvious mistakes.

Your point about draft text needs addressing - its one thing to know your public posts are consumed for other purposes, quite another to know your thinking before posting is consumed in the same way. Last time I looked at the privacy stuff I'm sure it mentioned posts, not drafts.
I've often drafted posts in a scratchpad and pasted it in since the site is not particularly friendly to accessibility tools. (a point I've made in the past and which gains the usual non responses). This would avoid being caught in draft capture.

I assume algorithmic and automated tools are in place for at least some classes of report, the extent to which AI is used in moderating isn't clear but probably should be more transparent. I've assumed for some time that spam report responses are automated as they are on most sites, along with key words.

This is all very sobering reading. I like MN but that was clearly a prompt designed to get gen AI to write posts, and that just ... gives me the ick, for want of a better phrase. It makes me wonder how much of this somewhat disturbing thread was actually AI-generated....

https://www.mumsnet.com/talk/am_i_being_unreasonable/5584904-how-are-people-using-ai-to-help-with-everyday-life?page=4&reply=155173579

So "AIBU Response Schema v1.0" is that part of the mumsnet api with AI?

Probably.

It certainly looks like MNHQ's plan is to ignore this issue and hope it goes away.

I guess that's a strategy that has worked for them previously.

And if posters start leaving the site because they don't feel that their concerns are being taken seriously, I guess they can always be replaced by AI.

Hi everyone,
Thanks for continuing to raise questions. I understand why this has unsettled people, particularly the discovery that draft text was being sent to OpenAI while you were writing. I also understand why our incorrect initial explanation has made the subsequent one harder to trust.

Here are some direct answers to questions that have come up on this thread.
We have never tested AI-generated responses to threads on Mumsnet. We do not use AI to write threads or replies.

The draft text sent to OpenAI for a title suggestion is not used to train its models. There is an opt-out from using that data for training, which we obviously chose. We receive no payment for sending draft text to OpenAI.

We introduced title suggestions to help avoid duplicate thread titles and make starting a thread easier. We recognise that removing friction can benefit users, and we approached this as a tool to help people start threads. But I understand that people feel differently about unpublished drafts, which may contain things they ultimately decide not to share. We should have made it clearer how the tool works and given people the option to opt out. As Sue said, the tool is switched off while we look at making it clearer and opt-in.

On moderation, we use AI to help flag potentially problematic posts for our team to review. A human always makes the moderation decision.

Finally, as said previously, we should have investigated more thoroughly before suggesting that something on the OP’s device might be responsible. I’m sorry for that, and for the understandable frustration it caused.

Thank you to the OP for flagging it and to everyone who has taken the time to explain their concerns.

@JustineMumsnet If you remember the fake mumsnet reviews from last year, the first response from mumsnet was a denial that it happened. When provided with evidence the denials continued before you came on and blamed rogue content editors. It seems the automatic MN response hasn't changed and no lessons were learned: deny, deny at all costs until you can't deny any more.

Your latest response still doesn't actually answer the question - why was this prompt generated? There is no version of the title generator details provided that will ask for how to write a response. Given the detail in the prompt response, I have no doubt that this will turn out to have been requested by someone at MNHQ. But I expect it will be handwaved as going beyond their responsibilities.

Can you please tell me if my posts have been sent to open AI as I have been drafting them AND before I press post when I start a thread. Thank you.

I do appreciate you coming back to respond to some of the concerns, @JustineMumsnet and the further clarifications.

I really hope that MNHQ will be reviewing its incident response protocols in light of what happened on this thread, in order to ensure that site users highlighting concerns in future are taken seriously and not fobbed off with incorrect explanations in order to quell speculation.

I also hope that MNHQ will have learned from the issue with the title suggestions and ensure that a full data protection impact assessment is carried out before any future changes to the site infrastructure or use of AI.

I think from @JustineMumsnet's reply, you should assume they have.

That’s what I think and I don’t think that’s clear in the privacy notice.

That's interesting. I wasn't aware of that previous issue but there do seem to be striking similarities in the response.

That further underlines my point about the importance of MN learning from the mistakes made in their response to this incident.

'not used to train their models' None of OAIs API content is used to train it's models. However they do retain content for 30 days for monitoring purposes.

Can you confirm if this is the standard API as above or have you opted in to ZDR.

( I work with APIs from OAI, Google Gemini, Anthropic, GLM and have several local models)

On moderation, we use AI to help flag potentially problematic posts for our team to review. A human always makes the moderation decision.
How does the AI decide what posts are potentially problematic?

I don’t think it’s doing a good job.

No, it isn't. I think it's clear that this was a cock-up on MNHQ's part, which is why they have switched the title suggestion off for now.

You’re supposed to tell us if you’re using AI are you not?

Yes, the EU AI Act.
(As their content is used within the EU, even if they are based in the UK)

The privacy notice doesn’t say that data is shared with Ai.

I am worried for eg that I’m starting a thread and drafting it and I decide it’s identifiable so I edit out info but the original draft has gone to open AI.

can Someone from mnhq show me where I’ve agreed to that please?

I find it very hard to believe that open AI would come up with "AIBU Response Schema v1.0:"

Something is not right here

Has anyone at Mumsnet been looking at the possibility of AI responses?

On moderation, we use AI to help flag potentially problematic posts for our team to review. A human always makes the moderation decision.

Does this apply to the word c r ypto and images?