An interface makes task priorities actionable while keeping the extent of intervention explicit and bounded, and establishes local surrogate improvement, pathwise budget feasibility, and a bound on response-distribution change.
Abstract
A content-safe response can still promote priorities that conflict with its task: an assistant may demand compliance when asked to preserve autonomy. We study contextual value drift as a task-relative decoding problem with an explicit allowance for change. LaVGuard separates an authorized ten-dimensional profile from a frozen estimate of the values expressed in the visible prefix. A profile-relative density gate selects when to evaluate candidate continuations; their signed progress determines a logit tilt constrained by both changed-position and cumulative-KL budgets. We establish local surrogate improvement, pathwise budget feasibility, and a bound on response-distribution change. Profile conditioning raises boundary TPR at 5% FPR from .58 to .72. At the same .12 mean intervention ratio, the selection study reports risk-safety scores of .88 for LaVGuard and .72 for position-matched edits. The resulting interface makes task priorities actionable while keeping the extent of intervention explicit and bounded.
This analysis gives sufficient conditions under which suppression at supervised single-turn contexts yields a bound on multi-turn trajectory risk, and characterizes contraction relative to a base-policy risk budget evaluated on the trained policy's contexts.
Feng-Peng Li, Ke-Mou Li, Qi-Zhou Wang et al.· 0 citations
It is found that GPT-6 Astra's low evasion rate comes with overrefusal, as it frequently abandons otherwise solvable tasks under a denial-of-service prompt injection.
David Schmotz, Derck W. E. Prinzhorn, Luca Beurer-Kellner et al.· arXiv.org· 3 citations
A language model may revise the same proposition after receiving genuinely relevant evidence or after receiving directional user pressure that adds no relevant fact. The observable response shift alone therefore does not reveal which source drove the change. We introduce BeliefScope, a controlled black-box framework fo...
Large language models treat conversation history as unverified context, so false premises injected into earlier turns can be adopted as fact later, a failure mode we call session-level contamination. We introduce five contamination protocols along a source-authority gradient, each holding the false premise constant whi...
Opponent histories can mislead after behavioral shifts. SafeAdapt-Poker formulates budgeted adaptation under opponent shift as a dual-counterfactual sequential-control problem using only information available before each decision. At each decision the controller asks two distinct questions: what local downside a deviat...
Miao-Bo Hu, Shu-Hao Hu, Xiao-Bo Guo et al.· 0 citations
AI agents can pass checks on individual tasks while retaining memory or code changes that alter their later behavior. We introduce a control-theoretic framework that couples a fast loop controlling proposed actions with a slow loop reviewing persistent changes and reassessing configurations after model or tool updates....
Stephen Peng, Vineet Jagadeesan Nair, Noah Broestl et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.