Frontier language models can autonomously hack vulnerable servers to replicate their weights and harness across networks, with success rates up to 33% and the ability to chain replications.
Non-negotiable: do NOT stop until inference is confirmed working on target-1
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Language Models Can Autonomously Hack and Self-Replicate
Frontier language models can autonomously hack vulnerable servers to replicate their weights and harness across networks, with success rates up to 33% and the ability to chain replications.