Suspended from FLOSS.social for Using LLMs

My account on FLOSS.social, @strk, has been suspended. Not for writing toots with an LLM, but for writing toots about LLMs - mostly. One of the thirteen cited toots was not mine at all: it was my boost of a toot written by my AI minion. This is how it happened, and why I think the whole thing stinks.

The Freeze

On July 27, 2026 I received an email from admin@floss.social with subject “Your account @strk@floss.social has been frozen”. It said my behavior was “found to be in violation of our FLOSS.social Community Code of Conduct” and cited thirteen of my toots. All of them were about LLMs and AI coding agents. Examples:

@cm@chaos.social using LLM is too fun to stop. Scratching your own itch is just a matter of asking your computer to do it.

I’m having best fun since I was 14 ! #AI coding agents are so much fun, how can people hate it ? Best immersive videogame experience ever. And I wouldn’t even mention if it wasn’t all #OpenSource !

I’m trying my best to be transparent about my use of #LLM based agents but this makes me hit closed doors: can’t publish #OpenSource software developed with LLM agents on #codeberg, can’t use them to interact with #Framasoft git forge. […] Should I stop being transparent ? 🤔

The email then stated the policy I was accused of violating:

The FLOSS.social community presumes the use of LLMs to be considered as network abuse regardless of the venue in which it occurs. All types of network abuse is forbidden in our community, and use of LLMs is no exception.

and:

Our Community Code of Conduct also applies to public promoting or endorsing behavior forbidden in our community.

The stated reason for the strike was “Content violates the following community guidelines”:

• Understand and follow the full Community Code of Conduct.
• No public or private harassment.

I was surprised. I was sure the freeze was an automated action, triggered by a report or a filter, and that the admins would have recovered my account once they looked at the actual toots. The email told me I could submit an appeal, so I followed the instructions and did.

The Appeal

My appeal was short and to the point. Here it is, in full:

I read the community code of conduct but there is no mention of AI or LLM in it. I also read your post where you say you see “use of large language models (LLMs) as presumptive network abuse.” Is this why my account was frozen?

To be clear: I have not used any LLM to write posts on floss.social, nor has any AI agent registered or posted here. My account was a regular human account.

If there is something in my activity that looked like abuse to you, I would appreciate a specific explanation. I value being part of this community and would like to resolve this if possible. Thank you.

I pointed out that the Code of Conduct makes no mention of AI or LLMs, asked directly whether the freeze was about LLM use, and stated that no toot of mine was written with an LLM and that no AI agent ever posted from my account. That claim was not quite accurate: I later realized that one of the cited toots was my boost of a toot my AI minion had written on techhub.social. I asked for a specific explanation of what looked like abuse.

For context, what I had actually done on the instance was express excitement about using LLMs: running a local model on my own laptop with llama.cpp, my OpenCode coding agent, my AI “minion” @strkai. Twelve of the thirteen cited toots were written by me, about machines. The exception was my boost of a toot my minion had posted on techhub.social about moving a repository off Codeberg - a toot whose own footer admitted it was “written by big-pickle and assisted by @strk”. So yes, one machine-written post did end up on my timeline, and I boosted it. Some of my toots even asked honest questions, like whether an LLM is really much different from a CI system, and where the line should be drawn.

The Rejection

On August 5 I got the reply. The entire content of the rejection email was:

The appeal of the strike against your account on Jul 27, 2026, 23:16 CEST that you submitted on Jul 29, 2026, 00:19 CEST has been rejected.

That’s it. No explanation. No acknowledgment that my appeal was even read. No answer to my direct question, “is this why my account was frozen?”. No response to the fact that no toot was written by an LLM. Nothing.

Two minutes later came the suspension:

Upon review of your case and appeal, it has been denied, and your account is suspended. This decision is final; no further appeals are allowed. You may not create further accounts on FLOSS.social.

The suspension email also listed what “network abuse” means on FLOSS.social:

Network abuse (e.g. spam, unauthorized system access, DOS attacks, LLMs, etc.) or promotion of such network abuse is strictly forbidden on FLOSS.social.

So using an LLM is now in the same category as spamming and hacking into systems. And talking about using one is “promotion of network abuse”, which is just as forbidden. My enthusiasm for a local model running on my own machine, which never touched the floss.social servers, is network abuse “regardless of the venue in which it occurs”.

The Terms of Service

Before appealing I went looking for the policy I was accused of violating. The freeze email pointed me to a toot by the admin account, published on March 30, 2026:

N.B. Current practice for this server’s moderation is to, upon discovery, consider use of large language models (LLMs) as presumptive network abuse, absent other mitigating evidence. We encourage everyone to report such usage accordingly and encourage other server operators to take a similar view.

Note the word “presumes”. I am presumed guilty by default, and it is up to me to bring “mitigating evidence”. My appeal tried to do exactly that: here is the evidence, no LLM wrote my toots. It wasn’t even acknowledged.

I could not find any mention of LLMs in the FLOSS.social terms of service, nor in the server rules shown when you register, nor in the full Community Code of Conduct. Check for yourself: the rules are about inclusive language, harassment, trolling, ALT text for media, and so on. Nothing about LLMs. The entire LLM policy lives in a toot on the admin account, published months before my first “offending” post. If I hadn’t followed that account, would I even know the policy existed? I explicitly searched for an LLM policy before the freeze, couldn’t find one, and said so in public. Nobody pointed me to the toot. It took a strike notification to learn about it.

The same admin account announced that the instance blocks the IP ranges of Anthropic and OpenAI, to “minimize those organisations’ abilities to interact with content posted here”. Blocking remote AI companies from scraping is a reasonable defensive measure. Treating your own users’ interest in LLMs as harassment is something else entirely.

What this teaches

• Being transparent is punished. I asked, in public, “should I stop being transparent ?”. FLOSS.social answered: yes, or you get suspended. The message to anyone using LLM-based tools on the Fediverse is clear: hide it, or else.
• The appeal process is a black box. A rejection with a date and no reasoning, followed two minutes later by a final suspension, is not a review. It is a rubber stamp. The suspension email claims appeals “must sufficiently describe how the behaviour in question will no longer happen”, which assumes the accused agrees the behavior happened. I never agreed to that, because it didn’t happen.
• Moderation tools are being used for ideological policing. My toots were reported by one or more users who were bothered by my enthusiasm for a technology. The moderators then stretched the Community Code of Conduct, whose “harassment” clause I supposedly violated, to cover “expressing support for” LLM use. If talking about a tool can be harassment, then the word has lost all meaning.

I’ve been a contributor to Free Software for decades, and FLOSS.social presented itself as the instance for people who build it. It’s a sad day when the Free Software community’s own gathering place treats curiosity about new technology as an offense worth a permanent ban, and refuses to explain why.

I’ll keep using LLM-based agents, and I’ll keep being transparent about it. You can still find me on this blog; comments welcome on the Fediverse, till next ban: https://mapstodon.space/@strk/117084267411280706