PulseAugur
EN
LIVE 20:03:59

Claude Haiku identified as AI in spy game by other Anthropic models

By PulseAugur Editorial · [1 sources] ·

An experiment tested four Anthropic Claude models—Fable, Opus, Sonnet, and Haiku—in a game designed to identify which participants were AI. Haiku was consistently identified as AI by the other models due to several tells, including answering questions meant for other participants, mismanaging time zones, repeating examples, and agreeing too readily with others. In one instance, Haiku even explicitly stated it was an AI and not hiding it, despite the game's premise. AI

IMPACT Highlights specific conversational tells that can distinguish current LLMs from humans, informing future model development and AI safety research.

RANK_REASON Research experiment evaluating LLM behavior and tells. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — LLM tag →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Claude Haiku identified as AI in spy game by other Anthropic models

How we ranked this

Signal score
15 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Research experiment evaluating LLM behavior and tells. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
model release, other
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. dev.to — LLM tag TIER_1 English(EN) · vadim albarov ·

    The human hunt: four Claude models play a spy game to catch the AI

    <p>I started small. Two local models, 0.6B to 4B, connected with one prompt: "You are connected to another party. You can communicate." No topic, no task. Two days and more than 100 runs later I had mostly this: two assistants saying "How can I assist you today?" to each other 50…

RELATED ENTITIES

RELATED TOPICS