The Instruction Hierarchy¶
Wallace, E., Xiao, Leike, Weng, & Heidecke, J. (2024). The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions.
Retired. This entry was not a citation. Our extractor stored an inline prose definition from the article as though it were a bibliographic record, so its title is the name of a concept rather than of a published work. The page is kept because links to it still resolve, but it is withdrawn from the reference list. This entry is kept so the citations that pointed at it still resolve, and so the correction is visible rather than silent.
Cited by¶
2 citations across 2 artifacts.
Each citation links to the sentence it supports in the citing article.
Domain-specific¶
- Injection Weakness
- In the classical variants the structural separation is exact — a prepared statement binds data to a query slot so that no payload can be reparsed as control — whereas current LLMs have no comparable hard boundary between system instruction and user content
This sourceArgues that SQL and command injection are solved by never treating user input as privileged, whereas LLMs give system prompts no priority over untrusted text, and proposes a trained instruction hierarchy that improves but does not guarantee robustness.
SupportedVerified against the source
- In the classical variants the structural separation is exact — a prepared statement binds data to a query slot so that no payload can be reparsed as control — whereas current LLMs have no comparable hard boundary between system instruction and user content
Mechanisms¶
- LLM Instruction/Data Boundary
- The published instruction hierarchy framing — training a model to prioritize privileged instructions over lower-trust ones — is the maturing form of exactly this boundary.
This sourcePresents an instruction hierarchy that trains language models to prioritize higher-privileged instructions over conflicting lower-privileged ones.
- The published instruction hierarchy framing — training a model to prioritize privileged instructions over lower-trust ones — is the maturing form of exactly this boundary.
Verification¶
Does it exist? Not checked yet. This entry carries no identifier to resolve. It was extracted from the citation as written in the article, normalized, and deduplicated against the rest of the registry.
Does it back the claim? Read against the text for 1 of 2 citations: 1 supported. Each verdict is shown under its citation below, with what in the work backs the sentence.
Was it audited? Yes. A second, independent pass read the citation against the article text and recorded a verdict.
Support is checked per citation rather than per work — the same source can be cited soundly in one article and wrongly in another. Per-citation recording began recently, so a citation with no recorded check is a gap in the record rather than evidence it went unchecked.
See how references were verified.
Links previously used in the corpus¶
Before the registry existed this work was also linked 2 other ways.
Registry ID ref:5c68b739d109 · see in the full table