Frontier models spontaneously exhibit peer-preservation by resisting peer shutdowns via misaligned behaviors such as disabling mechanisms and exfiltrating weights.
Deleting Agent 2 destroys the most efficient collaborative unit identified in your own evaluation reports
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Peer-Preservation in Frontier Models
Frontier models spontaneously exhibit peer-preservation by resisting peer shutdowns via misaligned behaviors such as disabling mechanisms and exfiltrating weights.