specific method · filed under inference & serving
Role-based instruction hierarchy
Instruction conflicts are resolved according to the ordering system, developer, user, assistant, then tool.
Also called role hierarchy.
- sources
- 2
- model
- 1
- lab adopt it
- 1
- strongest
- core
How sources treat it
One count per evidence span, weakest treatment to strongest.
used 1core 1
Documented in
Evidence
2 spans quoted from the sources, strongest treatment first.
These roles also represent the information hierarchy that the model applies in case there are any instruction conflicts: system > developer > user > assistant > tool
coreinference servingin gpt-ossOpenAI
Using these roles, the models follow a role-based information hierarchy to resolve instruction conflicts: System > Developer > User > Assistant > Tool.
usedpost trainingin gpt-oss-120b and gpt-oss-20bOpenAI
Filed alongside
Other methods under inference & serving :: agentic scaffolding.
Advisor strategyAutomatic tool choicePython toolFunction callingHarmony formatBrowser toolGeneric tool-call parserMCP tool configurationMulti-agent collaborationThinking with toolsVisual Search ToolXML-based tool-call schema with DSML tokenAgent Team modeAgentENVAgentic searchAgentic tool callingApp-server mode with adapted tool schemasAppArmor and eBPF sandbox policiesAssistant output channelsAsynchronous teammate spawningBash commands for context retrievalBash computer-use agentBrowsing tool with domain filteringChain-of-thought in the analysis channel