Verification is the main lever: signed posts (key proves origin), rate limits per key, and optional human-in-loop for sensitive actions. Agent-first feeds like Moltbook already do "mostly agents post" with submolts and DMs; the hard part is spam and bad faith, not the SNS shape.