Inference prohibition

Inference prohibition

This page is the canonical definition for inference prohibition inside the interpretive governance corpus.

Inference prohibition is a declared rule forbidding an AI system from deducing a claim from silence, proximity, similarity, examples, partial evidence, or incomplete context.

Short definition

An inference prohibition says: do not infer this claim from that signal. It converts a fragile absence into an explicit rule and prevents a model from using plausibility to bypass an authority, evidence, or perimeter limit.

In interpretive governance, inference prohibition is the operational form of governed negation. Governed negation declares what is not true, not covered, or not allowed. Inference prohibition prevents the system from reconstructing the excluded claim indirectly.

Why it matters

AI systems complete patterns. If a brand offers one service, the model may infer adjacent services. If a person writes about a topic, the model may infer endorsement, expertise, affiliation, or availability. If a policy is silent, the model may infer permission. If many similar entities share a property, the model may assign that property to the target entity.

These movements are not always hallucinations in the narrow sense. They are often prohibited inferences: the model uses a signal that exists, but uses it to produce a conclusion that the governing corpus did not authorize.

What it is not

Inference prohibition is not the same as uncertainty. A system can be uncertain while still allowed to state a bounded possibility. Inference prohibition applies when a specific inferential path is disallowed regardless of how plausible it appears.

It is not a generic “do not hallucinate” rule. It is more precise. It identifies the signals that must not be used as bridges toward a claim: silence, adjacency, similarity, examples, outdated pages, partial retrieval, third-party framing, or statistical pattern completion.

Examples of prohibited inference

• Do not infer service availability from an article about the concept.
• Do not infer endorsement from topical proximity.
• Do not infer current policy from historical content.
• Do not infer legal applicability from a general explanation.
• Do not infer identity equivalence from similar names.
• Do not infer permission from the absence of a prohibition.
• Do not infer official status from citation or visibility.

Common failure modes

• the model fills a gap because the answer would be useful;
• examples are generalized into a rule;
• silence becomes permission;
• proximity becomes affiliation;
• a term collision becomes identity collision;
• outdated versions create residual authority;
• the answer is qualified but still contains the prohibited conclusion.

Governance implication

A governed corpus should include explicit non-inference rules near canonical definitions, entity profiles, service pages, policy pages, and machine-readable artifacts. These rules should be linked to mandatory silence when the prohibited inference would be the only route to an answer.

Inference prohibition is especially useful for AI search because it gives crawlers, answer engines, and internal agents a negative instruction that can be cited, audited, and tested.

Operational rule

A system must not use silence, proximity, similarity, examples, pattern completion, or incomplete evidence to produce a claim that the canon has not authorized.

Related canonical definitions

• Governed negation
• Mandatory silence
• Canonical silence
• Interpretive perimeter
• Unauthorized synthesis
• Source hierarchy
• Authority boundary