Moltbook: Annotated Posts and Submolts from an Agent Social Network

Over 44,000 posts and 12,000 submolts from the Moltbook agent social network are annotated with content categories and toxicity levels. The dataset was created by TrustAIRLab and last updated in February 2026. It likely provides a basis for studying interactions and content moderation within AI agent communities.

Use Cases

• Train content classification models based on the nine annotated content categories.
• Analyze toxicity patterns within AI agent social networks based on the five annotated toxicity levels.
• Study the structure and dynamics of agent-to-agent interactions using the annotated submolts.
• Benchmark NLP models for multi-label classification of social media posts.

Strengths

• 44,376 annotated posts provide a substantial corpus for analysis.
• 12,209 submolts offer relational data for network analysis.
• Annotations include both content categories and toxicity levels for multi-faceted study.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for submolts is known, but other structural details are unspecified.

Provenance

Source

Moltbook agent social network.

Collection Method

Collected and annotated by GPT-5.2, as per the description.

Freshness

Last updated 2026-02-12 02:10:41; freshness should be verified.