CascadeLedger is a counterfactual execution‑audit framework designed for budgeted language‑model cascades. It enforces call‑before‑read semantics, records service and cache identities, and freezes event traces before scoring, ensuring that two executions can share a final answer and resource projection only if their routing decisions used legitimately observable information. The framework introduces counterfactual prefix non‑interference (CPNI), counterfactual influence frontier (CIF), and projection‑preserving adversarial mutation (PPAM) to detect and localize violations, achieving a clean false‑positive rate and high detection rates for serialized and hidden‑read faults while quantifying overhead for online and offline audits.
arXiv:2608. 02680v1 Announce Type: cross Abstract: Tool-using language-model agents repeatedly rediscover procedures they have already executed, producing traces that mix reusable structure with retries, exploration, accidental ordering, and repeated lookups.
The paper introduces DISCERN, a two-tier protocol for certifying that updates to production models do not increase risk. It first uses unlabeled data to detect benign updates based on disagreement rates, then selectively labels only disagreements through an anytime-valid confidence sequence. The method achieves finite-sample validity with label-complexity bounds of order ρ²/ε², demonstrating significant label savings and strong empirical performance across 14,000+ audit streams.
arXiv:2609.22243v1 Announce Type: new Abstract: Input-conditioned neural interventions raise a runtime question: what persists when one behavioral specification admits multiple actions whose validity...
The paper investigates how tool‑using language‑model agents can safely commit changes to infrastructure when external state may change between read and commit. By distinguishing invalidating races from predicate‑preserving and irrelevant ones, the authors evaluate three commit‑time guard granularities—global epoch, read‑set version, and semantic commit predicate—using a deterministic simulator and three quantized model families. The study finds that only the complete predicate guard consistently eliminates unsafe commits, while freshness‑based guards block a large proportion of benign races and model‑side signals fail to replace precise semantic enforcement.
arXiv:2508. 12220v2 Announce Type: replace-cross Abstract: Can a prospectively instrumented training continuation reproduce a deletion counterfactual exactly after selected examples leave its replay dataset?
arXiv:2610.15883v1 Announce Type: cross Abstract: Text wrapped in a trigger can be made behaviorally invisible to a language model: omitted from summaries and quotations, ignored in question answerin...
The paper introduces a claim‑anchored execution contract that binds a tool‑using agent’s emitted claim to its exact source span, the ordered execution prefix that produced it, and the source version and access state observed. Each receipt contains deterministic anchors, source identifiers, offsets, hashes, quotes, and a domain‑separated execution commitment, allowing a verifier to reconstruct these bindings before semantic or task labels are joined. The contract defines seven independently testable properties and demonstrates high detection rates against cross‑object attacks, with strong performance on conflict‑aware support guard evaluations.
The paper evaluates CASCADE, a fully local layered defense for Model Context Protocol (MCP)-based systems, by conducting a component ablation and corpus audit on a fixed 5,000-sample dataset. It demonstrates that the choice of aggregation convention heavily influences reported metrics, that detection performance varies with provenance, and that the released configuration does not fully disclose the operating point. The study also shows that a local review model invoked for a third of requests does not alter classification outcomes, highlighting the importance of reproducibility and transparency in defense evaluations.
arXiv:2607. 05397v1 Announce Type: cross Abstract: Agent systems increasingly execute rather than advise.
The paper investigates how personalized agents decide to use, ignore, update, or query retrieved user memory before acting on a task. An empirical audit protocol is developed to test structured intermediate outputs, revealing that while exposing state definitions improves accuracy, an explicit state-output field does not significantly enhance policy accuracy for large language models. The study also shows that example-level accuracy overstates consistency, with full four‑way family success being rare, and that providing benchmark‑associated state labels merely conditions predictions rather than proving internal fidelity.
arXiv:2609. 04875v1 Announce Type: cross Abstract: Long-running LLM agents are stateful: beyond the transcript they accrete compressed summaries, plaintext memory, pending tool plans, and, under every serving API, a KV cache.
arXiv:2607. 20436v1 Announce Type: cross Abstract: Safety evaluations often assume that behavior observed during testing reflects behavior in ordinary use, but fine-tuning can break this assumption.