Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-03-17 · source-backed
UK AI Safety Institute found model capability at 10M tokens jumped from 1.7 to 9.8 steps completed on a 32-step corporate network attack. Each 10x compute increase yields 59% more steps. No plateau found. Best run: 22/32 steps — 6 hours of a 14-hour human expert workload. Source
Each link below shares sources, entities, or timing with this story.
Shared entity: Best / Same source domain / Shared topic / What happened next / Tension
Both cover Best; reported by the same outlet (arxiv.org); overlapping topics (agent, best, human).
Shared entity: Best / Same source domain / Shared topic / What happened next
Both cover Best; reported by the same outlet (arxiv.org); overlapping topics (agent, best, expert, frontier).
Shared entity: Best / Same source domain / Shared topic / What happened next / Tension
Both cover Best; reported by the same outlet (arxiv.org); overlapping topics (agent, best).
Shared entity: Each / Same source domain / Shared topic / What happened next / Tension
Both cover Each; reported by the same outlet (arxiv.org); overlapping topics (agent, each).
Shared entity: Best / Same source domain / Shared topic / What happened next
Both cover Best; reported by the same outlet (arxiv.org); overlapping topics (agent, hour, human).
Shared entity: Best / Same source domain / Shared topic / Earlier coverage / Downstream implication
Both cover Best; reported by the same outlet (arxiv.org); overlapping topics (agent, best).
Shared entity: Each / Same source domain / Shared topic / Earlier coverage
Both cover Each; reported by the same outlet (arxiv.org); overlapping topics (agent, each, step).
Shared entity: Best / Same source domain / Shared topic / What happened next
Both cover Best; reported by the same outlet (arxiv.org); overlapping topics (agent, attack).