Checked fact 326631 Oct 2026Research
The paper says that for the top three sabotage agents, removing the hidden scratchpad brought a significant decrease in overall, evasion and side-task success, because lacking it causes more obvious references to the side tasks and small mistakes in execution.
Toned down to what the source saysThe exact words it rests on
Figure 7 shows the key performance metrics for the top three performing sabotage agents without a hidden scratchpad. We find a significant decrease in overall success, evasion success, and side task success in this setting. Lacking access to the scratchpad causes more obvious references to the side tasks and small mistakes in execution
What the source said when we opened it, on 31 Oct 2026.
The source
SHADE-Arena: Evaluating Sabotage and Monitoring in LLM Agents
Checked
Checked by the notis newsroom on , against the source above.
In the story
How AI monitors work, and how Anthropic's tests tried to evade them 31 Oct 2026
Cite this fact
Anyone may quote this address. It does not change; if we correct the story, this page says so.