Discourse AI spam detection replaces Akismet plugin

Starting today, we will be removing the Discourse Akismet plugin and will instead enable Discourse AI - Spam detection by default for hosted customers on Pro and Standard plans.

Why are we making this change?

Over the past few weeks, we’ve found our AI spam detector to deliver a superior performance over Akismet.

We’ve had over 126 sites extensively using it and within Meta, we’ve personally seen over 96% success rate in detecting spam.

On top of the performance, the setup would now be a default experience with customization and flexible options, should you need it.

What do I need to do?

If you are on our Starter or Pro hosting, nothing at all!

The change would happen automatically in the background. You will no longer see Akismet within the plugins. You will now see the AI spam detector turned on in the AI plugin setting using a Discourse-hosted open-weights LLM

If you have the AI plugin disabled, then you would manually need to turn on AI spam detector and select an LLM

I already have AI spam detection enabled!

If you already have AI spam detector enabled via What's New experiments, then you can safely ignore this announcement since you already have it!

Why is this change only for Starter and Pro?

As usual, the rollout of changes in our hosting begin with the starting tiers and are gradually expanded to cover all tiers.

19 Likes

2 posts were split to a new topic: Rebranding the ai spam detection account

Does this mean you’re discontinuing the plugin at all, or just that you’re no longer offering it on your hosted platform? I’m self-hosted and I’m not using any ‘AI’. Will I be forced to start using the AI Spam detection?

3 Likes

This change only affects hosted customers on Starter and Standard plan tiers. The Akismet plugin is removed and no longer available for those plan tiers

Self hosters can make their own choices here

3 Likes

Will AI spam detection work properly for languages with lesser people who speak the language in the world base? Like e.g. Ukrainian.

Also, will it handle bilingual posts, which is natural for some countries, like e.g. in Ukraine people speak both Ukrainian and Russian interchangeably, and they can sometimes use both in a single post.

I guess another example could be Naple/Italian? Or Spanish/Catalan? But not very sure about these pairs – I only know about Ukrainian/Russian.

3 Likes

Absolutely, it simply depends on the LLM you use to power it.

Google models tend to be extremely multilingual, Gemini Flash is 120 languages so it is a good candidate.

Try it out with the test button on some real posts, some spam posts etc… you can provide custom instructions like:

Posting in Russian / Ukrainian interchangeably is fine.

4 Likes

Via OpenAI works for Finnish. Well, it gives every now and then false positives if there is one liner, but I don’t think that is language issue, but weak prompting.

5 Likes

Hi, I am self hosted. Is the Askimet spam detection integrated into the core of a Discourse install or do I need to install this plugin separately? Thanks.

1 Like

It’s not included in core. You can find the topic here: Discourse Akismet

But the topic is closed and hidden. I’m not sure what that means. It would make sense if the plugin were no longer maintained, but this topic says it’s only removed from two hosting tiers. So why shouldn’t self-hosters and hosted customers on higher tiers be able to find the information by searching Meta? I would wait for an answer to that before installing it.


And I think one time in the first post accidentially ‘Starter’ instead of ‘Standard’ was replaced with ‘Pro’.

Akismet has failed us, too many times. We have observed many sites with it installed which simply become flooded with spam.

Self hosters should use Discourse AI with one of the many free models if budget is a concern. Gemini Flash 2.0 free tier does the trick reasonably.

5 Likes

Thanks for the information, it was not obvious my self hosted Discourse site was not protected against spam. I assumed wrongly it was. OMO this info should be more prominently put forward for would be Discourse admins envisaging self hosting heir site.

I will look for that Discourse AI functionality, which I never considered before due to budget restrictions.

I try here to list the actions for an admin on a self hosted site like me, can you tell me if I am right?

  1. First I should install the Discourse AI plugin? Discourse AI

  2. Second, I should setup a LLM like explained here? Discourse AI - Large Language Model (LLM) settings page

  3. Then I should setup a spam detection mechanism like explained here? Discourse AI - Spam detection

A reminder/note to self-hosters:

Adding “random” LLMs to deal with the spam problem potentially opens a can of worms in terms of privacy and protection of personally identifiable data (PII). If you’re required to follow the GDPR or COPPA etc., the use of external LLMs might require you to enter a data protection agreement with the LLM provider to ensure your site doesn’t bleed PII to a third party. And PII start with simple things like names and email addresses.

This is just something to keep in mind.

1 Like

I don`t know how things are in USA or UK [1], but in EU this is relatively simple question.

Necessary data can be ”bleeded” aka. transferred to third party, even it is counted as PII, if they follow GDPR. And all major players do. Or so they say. It is matter of trust.

Hate to say this at loud, but I won’t trust any company, that is from China. And same thing with US-companies. Both parties have government issues big time, and need to make money in abroad style is heavy. Microsoft leaked/is leaking data to US, even they should keep it in EU.

Do I still use Amazon and OpenAI? Yes, because I must. But all my VPSes and most of S3 activity has been moved away from AWS. But my personal emails, and transactional ones from i.e. forum, are still handled by AWS.

All of my forum based AI activity is handled by OpenAI. I don’t trust them — but what are my options? I really dislike Microsoft connection, but things could be worse…

So, we as EU-admins are quite safe when we don’t ask passport copies or personal IDs, because Discourse as a platform handles its part quite good. And as long 3rd party declares that they follow GDPR and other regulation (same thing as with Google ads etc) ”bleeding” isn’t a legal issue.

But it is a moral issue for sure. That’s why we either use AI and at same trying to teach users, or not use at all. Those are options.

And in the big picture… AI in a forum is way smaller issue, in PII wise, than using Meta, 8.8.8.8 as DNS or commercial VPN-services (or record iPhone-calls using chinese servers… [2]).


  1. personal opinion: both countries are mess in this matter ↩︎

  2. for real… my robo vacuum wanted to send all location data, and wifi information to China ↩︎

1 Like