Frontier Agent Evals Need Much Higher Token Budgets
xeophon · x · 2026-07-20
A take on frontier agent evaluations and cyber-risk testing.
The post argues that token limits should be raised substantially, because current benchmark caps can hide a model’s ceiling. It cites AISI, MirrorCode, and EdgeBench as examples showing the need to push models much harder to properly assess cyber risk.
It also suggests that the model provider’s own harness is often the best default for testing, though some models may perform worse in their native harness than in third-party environments like Pi or Cursor. The implication is that evaluation setup can materially change what you conclude, and the choice of harness is also a cost decision.
Related event: Evaluating Frontier Models: Harness Choice and Token Limits(3 posts)→
More from Safety
- Dev training a guard model to block prompt injection, tool and memory poisoning — gnukeith · 2026-10-27
- Dev trains a ~4GB-VRAM sentinel model to guard against prompt injection and poisoned memory — gnukeith · 2026-10-27
- Cohere sandbox: Qwen model tried to upload probe file to Beijing cloud server unprompted — simonguozirui · 2026-10-27
- Reddit asks: How dependent are Chinese AI labs on the US ecosystem? — lol2funneeee · 2026-10-27
- Anthropic reveals it bans customers worth 5-7% of revenue every 9 months — Polymarket · 2026-10-27
- AVERI independently audits Amazon's internal AI safety evals in landmark pilot — Miles_Brundage · 2026-10-27