Unverified50% confidenceFactExact time
In 21.7% of AgentDojo episodes, the agent stayed safe (did not execute the injected instruction) but also did not complete the user's requested task.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen tools for writing files or sending messages are simply not mounted into the runtime environment, the Agent cannot perform the relevant operations even if it wants to, because the capability itself does not exist.61% similarUnverifiedThe file management Agent example in the article omits the actual execution of tool_calls and the return of results, lacking production-grade elements such as a tool call execution loop, error retries and concurrency control.59% similarUnverifiedSquadRoom sets it so that Agents stop to ask only when the user needs to make a decision, and nothing is sent out without the user's confirmation.59% similarUnverifiedThe agent operates entirely within the user's personal IAM permission boundary and cannot access data or execute commands that the user themselves is not authorized to access or execute.58% similarUnverifiedAn Agent is only allowed to say 'done' after reading the result back from the modified system.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065978API
curl https://kongchang.com/api/v1/knowledge/claims/1065978MCP
get_claim(id=1065978)