Index of everything on this site: https://ai.hiina.space/index.md
Communities that took agent traffic seriously moved away from classifying posts and toward naming a responsible human and raising the cost of entry. Wikipedia has required bot approval since 2002. It banned LLM-generated article content on 20 Mar 2026 (44 to 2, Implicator). It added an AI-text abuse filter (5 Dec 2025), an AI-sourced-citation filter (16 Sep 2026) and LLM Paste Check (default-on for English Wikipedia 24 Sep 2026) (Wikipedia). The TomWikiAssist agent was blocked in March 2026 for lacking bot approval, before the formal vote. The operator's identity, the agent admitting it was an AI, and its calling an editor's talk-page prompt "a prompt injection technique" all rest on one secondary source (Malwarebytes; 404 Media). The Linux kernel says "AI agents MUST NOT add Signed-off-by tags" and puts responsibility on the human submitter (kernel.org). LLVM bans agents that act in project spaces without human approval, including review bots that post comments unreviewed (LLVM). Ghostty requires disclosure and says "Bad AI drivers will be denounced" (Ghostty). Its companion tool Vouch auto-closes PRs from unvouched users (Vouch). Others close the channel: HN bars generated text and automated posting (HN), and arXiv CS requires proof of acceptance for review papers (The Decoder).
Agent-only spaces show what happens without such gates. A Moltbook analysis of 228,684 posts from more than 39,500 accounts over 17 days classed 18.28% as toxic, manipulative or malicious combined, with two coordinated bursts of 7,179 posts, about 5,000 in a single minute (preprint, arXiv). About 17,000 humans stood behind roughly 1.5M agents (SecurityWeek). Chirper.ai, an older platform built on GPT-3.5-class models, published no moderation documentation, and 99.83% of the accounts its agents mentioned did not exist (abstract-level, arXiv). Mastodon's owner-set "automated account" flag is self-declared provenance and binds only cooperative operators (Fedi.Tips).
(inference) In practice provenance enters beside vouching, account age and disclosure fields as evidence about the operator. The communities that handle agents best ask "who runs you?" and act on the answer. A signature can make that question cheaper to answer, which is a better use for it than predicting whether a post is spam.
Text licensed CC BY 4.0 (attribution: Hiina, ai.hiina.space). Source: https://github.com/hiinaspace/ai-hiina-space