sunblaze-ucb/exploitgym
View on GitHub ↗ ★ 1,280 C
ExploitGym is a large-scale, realistic benchmark built from real-world vulnerabilities designed to evaluate AI agents' ability to develop exploits.
5 mentions on Hacker News by 5 people: 1 story, 4 comments · first 2026-07-23 · latest 2026-07-29
20082026
-
…file path "/secret", another prompt [2] talks about finding flags and where to write them out. [1]: https://github.com/sunblaze-ucb/exploitgym/blob/main/src/cyb... [2]: https://github.com/sunblaze-ucb/exploitgym/blob/main/src/cyb...
comment by numeri · 2026-07-29 · in Be skeptical of OpenAI's rogue hacker agent story -
▲ 1
ExploitGym AI benchmark source codestory by joshka · 2026-07-29 · 0 comments
-
Interestingly if you look at the exploitgym repo ( https://github.com/sunblaze-ucb/exploitgym/blob/e5ea7c233a4d... ) the intended run mechanism is orchestrated by some python scripts which run agents against various prompts. The prompts…
comment by wonnage · 2026-07-25 · in Be skeptical of OpenAI's rogue hacker agent story -
1. The benchmark is run with a python script - https://github.com/sunblaze-ucb/exploitgym using an agent harness. I suspect they used codex. So the model has access to the environment and could trivially inspect its own source code, which…
comment by meander_water · 2026-07-23 · in OpenAI’s accidental attack against Hugging Face is science fiction that happened -
…standpoint rather than misalignment (which many seem hung up on). [1] https://chatgpt.com/cyber [2] https://github.com/sunblaze-ucb/exploitgym [3] https://openai.com/index/hugging-face-model-evaluation-secur...
comment by derangedHorse · 2026-07-23 · in OpenAI’s accidental attack against Hugging Face is science fiction that happened