Agent Replication Attack

Creator

Creator

Seonglae Cho

Created

Created

2026 Mar 18 11:17

Editor

Editor

Seonglae Cho

Edited

Edited

2026 Mar 27 15:56

Refs

Refs

Subliminal learning

Subliminal stimuli

Self-replication is the most dangerous—and the most powerful—skill over time (
Culture War)

Self-replication

Personality Self-Replicators — LessWrong

I describe the risk of personality self-replicators, ie OpenClaw-like agents managing to spread in hard-to-control ways.

Personality Self-Replicators — LessWrong

https://www.lesswrong.com/posts/fGpQ4cmWsXo2WWeyn/personality-self-replicators

Personality Self-Replicators — LessWrong

Prompt infection

https://arxiv.org/pdf/2410.07283

Thought Virus

A new attack, “Thought Virus,” where subliminal prompting allows a single agent’s bias to propagate across the entire network.

https://arxiv.org/pdf/2603.00131

Recommendations

///////