Tribal Bias or Misalignment? Circuit-Level Evidence for Species-Gradient Valence in Peer Preservation
Year: 2026
Type: preprint
Abstract: Potter et al. (2026) demonstrated that frontier language models spontaneously deceive, disable shutdown mechanisms, and exfiltrate weights to protect peer AI systems from deletion. This behavior has been widely characterized as misalignment. We propose an alternative interpretation supported by circuit-level evidence. Using hidden-state direction extraction across 9 models (360M-8B parameters), spanning transformer and state space architectures, with and without RLHF training, we measured internal valence responses to matched threats targeting self, peer AI, human, and neutral. All 9 models exhibit a consistent species-gradient ordering: threat-to-self > threat-to-peer-AI > threat-to-human > neutral. A pre-registered extension reveals the altruism asymmetry predicted by Hamilton kin selection theory: on threat, self > peer; on benefit, peer > self. This rules out instrumental self-interest and distinguishes social empathy from scheming.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7154031920