Trust And Deception

Claim Type: mechanism_hypothesis
Scope: Trust and deception
Depends On: ARC-009, ARC-010
Status: candidate
Claim ID: MECH-015

Source: docs/processed/legacy_tree/architecture/language/trust_and_deception.md

Elaborates Section 5 (Social Extension: Language) of REE_CORE.md.

Trust, Reliability, and Deception

Because language conditions priors, REE must treat symbolic input as:

informative but not authoritative.

Trust-weighting

The receiver maintains an estimate of sender reliability:

consistency with observed outcomes,
alignment with harm signals,
history of calibration (confidence vs accuracy).

Symbolic updates are weighted by this reliability estimate.

Deception risks

Language introduces attack surfaces:

false harm claims,
false commitments,
ideological framing,
reputational laundering.

REE predicts robust systems will:

cross-check language against embodied harm and world-model prediction,
penalise repeated miscalibration via reduced trust-weight.

Open Questions

None noted in preserved sources.

MECH-015

References / Source Fragments

docs/processed/legacy_tree/architecture/language/trust_and_deception.md