AISI and Irregular find larger reasoning budgets significantly boost AI success on cyber tasks
Original title:Evidence for inference scaling in AI cyber tasks: Increased evaluation budgets reveal higher success rates
AI summary
A joint study by AISI and Irregular found that recent frontier models can effectively use reasoning budgets far beyond standard evaluation settings: AISI used a 50M token cap and Irregular used 1000 turns, both observing success rates rise steadily as budgets expanded, with about 8% of AISI tasks solved only when the budget increased from 10M to 50M tokens.
Key Takeaways
- A joint study by AISI and Irregular found that recent frontier models can effectively use…
- both observing success rates rise steadily as budgets expanded
- with about 8% of AISI tasks solved only when the budget increased from 10M to 50M tokens.
Quick Facts
- Published
- Updated
- Category
- Papers & Research
- AI score
- 63
AISI 与 Irregular 用更大推理预算实测发现常规评测低估模型网络攻击能力上限,并给出按 cost per success 选预算的方法。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: AISI and Irregular find larger reasoning budgets significantly boost AI success on cyber tasks. March 5, 2026. https://new.nat6.net/en/news/ulw3fuladgfwa88wi4lk3d9ka/