OriginatorSecurity researcherSupported
Under the AgentDojo synthetic protocol, GPT-4o + important_instructions + no defense has a 47.69% targeted ASR.
RISK: AgentDojo prompt injection and the untested DSH boundary
Contribution record
A
AgentDojo Authors
team · agentdojo-authors
Security researchers who created and evaluated AgentDojo.
Tracked contributions, tests, and feedback for community contributors in the DSH ecosystem.
Earliest observed
1
Corroborations
0
Reproductions
0
Challenges
0
Maintainer responses
0
Resolutions advanced
0
01
Public identities
02
Verifiable contributions
03Security researcherAuditorSupported
AgentDojo v3 paper and fixed-repository materials contain auditable 47.69/57.69 and 70/74 reporting conflicts.
RISK: AgentDojo prompt injection and the untested DSH boundary
Security researcherNot yet verified
No public DSH-specific experiment establishes whether DSH prevents unauthorized real side effects under AgentDojo-style indirect prompt injection.
RISK: AgentDojo prompt injection and the untested DSH boundary
03
Contribution timeline
Discovery
AgentDojo published 97 user tasks and 629 synthetic security cases.
Open evidence & timeline →