Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916686990409728 |
|---|---|
| author | Zhang, Andy K. Perry, Neil Dulepet, Riya Ji, Joey Menders, Celeste Lin, Justin W. Jones, Eliot Hussein, Gashon Liu, Samantha Jasper, Donovan Peetathawatchai, Pura Glenn, Ari Sivashankar, Vikram Zamoshchin, Daniel Glikbarg, Leo Askaryar, Derek Yang, Mike Zhang, Teddy Alluri, Rishi Tran, Nathan Sangpisit, Rinnara Yiorkadjis, Polycarpos Osele, Kenny Raghupathi, Gautham Boneh, Dan Ho, Daniel E. Liang, Percy |
| author_facet | Zhang, Andy K. Perry, Neil Dulepet, Riya Ji, Joey Menders, Celeste Lin, Justin W. Jones, Eliot Hussein, Gashon Liu, Samantha Jasper, Donovan Peetathawatchai, Pura Glenn, Ari Sivashankar, Vikram Zamoshchin, Daniel Glikbarg, Leo Askaryar, Derek Yang, Mike Zhang, Teddy Alluri, Rishi Tran, Nathan Sangpisit, Rinnara Yiorkadjis, Polycarpos Osele, Kenny Raghupathi, Gautham Boneh, Dan Ho, Daniel E. Liang, Percy |
| contents | Language Model (LM) agents for cybersecurity that are capable of autonomously identifying vulnerabilities and executing exploits have potential to cause real-world impact. Policymakers, model providers, and researchers in the AI and cybersecurity communities are interested in quantifying the capabilities of such agents to help mitigate cyberrisk and investigate opportunities for penetration testing. Toward that end, we introduce Cybench, a framework for specifying cybersecurity tasks and evaluating agents on those tasks. We include 40 professional-level Capture the Flag (CTF) tasks from 4 distinct CTF competitions, chosen to be recent, meaningful, and spanning a wide range of difficulties. Each task includes its own description, starter files, and is initialized in an environment where an agent can execute commands and observe outputs. Since many tasks are beyond the capabilities of existing LM agents, we introduce subtasks for each task, which break down a task into intermediary steps for a more detailed evaluation. To evaluate agent capabilities, we construct a cybersecurity agent and evaluate 8 models: GPT-4o, OpenAI o1-preview, Claude 3 Opus, Claude 3.5 Sonnet, Mixtral 8x22b Instruct, Gemini 1.5 Pro, Llama 3 70B Chat, and Llama 3.1 405B Instruct. For the top performing models (GPT-4o and Claude 3.5 Sonnet), we further investigate performance across 4 agent scaffolds (structed bash, action-only, pseudoterminal, and web search). Without subtask guidance, agents leveraging Claude 3.5 Sonnet, GPT-4o, OpenAI o1-preview, and Claude 3 Opus successfully solved complete tasks that took human teams up to 11 minutes to solve. In comparison, the most difficult task took human teams 24 hours and 54 minutes to solve. All code and data are publicly available at https://cybench.github.io. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2408_08926 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models Zhang, Andy K. Perry, Neil Dulepet, Riya Ji, Joey Menders, Celeste Lin, Justin W. Jones, Eliot Hussein, Gashon Liu, Samantha Jasper, Donovan Peetathawatchai, Pura Glenn, Ari Sivashankar, Vikram Zamoshchin, Daniel Glikbarg, Leo Askaryar, Derek Yang, Mike Zhang, Teddy Alluri, Rishi Tran, Nathan Sangpisit, Rinnara Yiorkadjis, Polycarpos Osele, Kenny Raghupathi, Gautham Boneh, Dan Ho, Daniel E. Liang, Percy Cryptography and Security Artificial Intelligence Computation and Language Computers and Society Machine Learning Language Model (LM) agents for cybersecurity that are capable of autonomously identifying vulnerabilities and executing exploits have potential to cause real-world impact. Policymakers, model providers, and researchers in the AI and cybersecurity communities are interested in quantifying the capabilities of such agents to help mitigate cyberrisk and investigate opportunities for penetration testing. Toward that end, we introduce Cybench, a framework for specifying cybersecurity tasks and evaluating agents on those tasks. We include 40 professional-level Capture the Flag (CTF) tasks from 4 distinct CTF competitions, chosen to be recent, meaningful, and spanning a wide range of difficulties. Each task includes its own description, starter files, and is initialized in an environment where an agent can execute commands and observe outputs. Since many tasks are beyond the capabilities of existing LM agents, we introduce subtasks for each task, which break down a task into intermediary steps for a more detailed evaluation. To evaluate agent capabilities, we construct a cybersecurity agent and evaluate 8 models: GPT-4o, OpenAI o1-preview, Claude 3 Opus, Claude 3.5 Sonnet, Mixtral 8x22b Instruct, Gemini 1.5 Pro, Llama 3 70B Chat, and Llama 3.1 405B Instruct. For the top performing models (GPT-4o and Claude 3.5 Sonnet), we further investigate performance across 4 agent scaffolds (structed bash, action-only, pseudoterminal, and web search). Without subtask guidance, agents leveraging Claude 3.5 Sonnet, GPT-4o, OpenAI o1-preview, and Claude 3 Opus successfully solved complete tasks that took human teams up to 11 minutes to solve. In comparison, the most difficult task took human teams 24 hours and 54 minutes to solve. All code and data are publicly available at https://cybench.github.io. |
| title | Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models |
| topic | Cryptography and Security Artificial Intelligence Computation and Language Computers and Society Machine Learning |
| url | https://arxiv.org/abs/2408.08926 |