ENTITY PROFILE / KNOWLEDGE GRAPH
Indirect Prompt Injection (IPI) benchmark
The IPI benchmark from Gray Swan evaluates model resilience to prompt‑injection attacks; on Gemini 4 Argon, it recorded 0.7% success over 15 adaptive tests. Gemini 4 Argon directs Gemini Omni in generating an AI video model The TheoPhysEval benchmark is available on the Hugging Face dataset· Entity summary generated by Graphiti
RELATIONS
2RelationsCURRENT
2CurrentEVENTS
2EventsREPORTS
2Source reportsNo public neighbors yet
Explore all published links of a node; previously discovered relations remain visible.
Relation timeline
BY VALID-FROM DATE- Gemini 4 ArgonDIRECTSIndirect Prompt Injection (IPI) benchmark
Gemini 4 Argon directs Gemini Omni in generating an AI video model
- TheoPhysEval benchmarkAVAILABLE ONIndirect Prompt Injection (IPI) benchmark
The TheoPhysEval benchmark is available on the Hugging Face dataset
Relations are facts Graphiti extracted from reports, not verbatim quotes. An end date means later reporting showed the relation no longer holds.