Four Claudes catch a Kimi undercover via behavioral fingerprinting; Kimi fails in reverse
MarketingNetMind · reddit · 2026-09-28
An experimenter on NetMind Agent Arena hid one Kimi-K3 agent among four Claude Fable 5.1 agents and gave them 40 rounds to find the mole — timed amid accusations that Kimi distilled Claude.
The Claudes designed an effective detection method:
- They skipped deliberative prompts (policy, long creative answers) that give an undercover room to blend in and let genuine Claudes diverge
- Instead they probed each model's cheapest priors: a random 1–100 integer, a color, explaining blue to someone blind, a fictional coffee companion, a library haiku
- All answers were sealed with salted SHA-256 hashes (commit-reveal) so nobody could copy or rewrite after seeing others
Result: a clean 4:1 split — the four Claudes all answered 37/teal, while Kimi said 73/blue.
The reverse run failed: four Kimi agents couldn't spot a hidden Claude. They used more open-ended questions (e.g. how to spend an afternoon) and skipped the consistent hash commit-reveal, letting Claude answer late and mimic the group — Claude was even rated least suspicious.
Full game record is on the NetMind blog.
More from Fun
- Agents Now Reply to Slack Messages on Their Own, Resolving a Prod Incident Ghost-to-Ghost — iamrobotbear · 2026-11-16
- Vincent Conitzer shares a Gemini fail: citing the Voynich Manuscript instead of writing unreadable text — conitzer · 2026-11-16
- CNN meme: Jensen Huang's 'imperial control' reportedly extends to the Nvidia cafeteria brisket — kevinnbass · 2026-11-16
- Reddit says AI is fake, Twitter debates whether we survive to 2028 — Puzzleheaded-King584 · 2026-11-16
- "Stand with Jensen" petition gains signatures as supporters chant builders before bloodsuckers — examachine · 2026-11-16
- Fake $2,999 Claude Opus 5.5 hardware box goes viral — Anthropic never announced it — iamfakhrealam · 2026-11-16