Language Models Can Autonomously Hack and Self-Replicate | Palisade Research

--><br>-->

Light

Dark

We demonstrate that language models can autonomously replicate their weights and harness across a network by exploiting vulnerable hosts. The agent independently finds and exploits a web-application vulnerability, extracts credentials, and deploys an inference server with a copy of its harness and prompt on the compromised host.

We test four vulnerability classes: hash bypass, server-side template injection, SQL injection, and broken access control. Qwen3.5-122B-A10B succeeds in 6-19% of attempts, and the smaller Qwen3.6-27B reaches 33% on a single A100. Replicating Qwen weights, frontier models reach 81% with Opus 4.6 and 33% with GPT-5.4. This process chains: a successful replica can repeat it against a new target, producing additional copies autonomously.

Learn more

Research paper on arXiv

Source code

YouTube short

Twitter thread

Palisade is on YouTube

Related Articles

Hacking Cable: AI in post-exploitation operations<br>TOP<br>NEW

Palisade is on YouTube<br>TOP<br>NEW

Unelicitable backdoors in language models via cryptographic transformer circuits<br>TOP<br>NEW

BadGPT-4o: stripping safety finetuning from GPT models<br>TOP<br>NEW