Language Models Can Autonomously Hack and Self-Replicate
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866911658804248576 |
|---|---|
| author | Air, Alena Reworr Kotov, Nikolaj Volkov, Dmitrii Steidley, John Ladish, Jeffrey |
| author_facet | Air, Alena Reworr Kotov, Nikolaj Volkov, Dmitrii Steidley, John Ladish, Jeffrey |
| contents | 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. This already matches the current-generation GPT-5.4 and exceeds the prior-generation frontier, where Opus 4 reached 6% and GPT-5 reached 0%. Replicating Qwen weights, frontier models reach 81% (Opus 4.6) and 33% (GPT-5.4). This process chains: a successful replica can repeat it against a new target, producing additional copies autonomously. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_06760 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Language Models Can Autonomously Hack and Self-Replicate Air, Alena Reworr Kotov, Nikolaj Volkov, Dmitrii Steidley, John Ladish, Jeffrey Cryptography and Security 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. This already matches the current-generation GPT-5.4 and exceeds the prior-generation frontier, where Opus 4 reached 6% and GPT-5 reached 0%. Replicating Qwen weights, frontier models reach 81% (Opus 4.6) and 33% (GPT-5.4). This process chains: a successful replica can repeat it against a new target, producing additional copies autonomously. |
| title | Language Models Can Autonomously Hack and Self-Replicate |
| topic | Cryptography and Security |
| url | https://arxiv.org/abs/2605.06760 |