arXiv subjects

Gaston Besanson

Publications and source records attributed to Gaston Besanson.

8 recordsLinked to original sources

arXiv2026

One Gate Is Not Enough: Composing Stateful Pre-Action Controls for Agentic AI

Agentic AI systems take consequential actions governed by more than one pre-action control at once: authority, resource, and evidence gates that can admit, degrade, or remediate an action before it executes. This paper's central object is remediation-induced control coupling: a remediation applied by one control can change the action, evidence, or context another control evaluates, invalidating that control's earlier judgment. We formalize this coupling and give a remediate-and-regate protocol that restores per-action soundness in the current bounded, idempotent setting under its stated assumptions. We further show that the two implemented remediation operators (evidence substitution and resource-budget downroute) do not commute -- a finite-model checker finds concrete counterexample instances -- making remediation order part of the control-plane semantics rather than an implementation detail. A governed evidence buffer that trusts its own most recent admitted write is a further instance of the same problem at the level of state -- current admissibility does not imply future reference trustworthiness -- and is vulnerable to poisoning from declared-uncovered defect classes; two mitigations reduce, not eliminate, that exposure. Supporting results establish the exact condition under which positive-weight linear aggregation of gate outcomes can compensate a member veto, a unified cross-control Evidence Set, and that composition manufactures no new detection coverage, reported honestly. Empirically, on a deterministic open-data artifact composing three published engines unmodified, CH1-CH5 meet their registered decision rules across all 30 pre-registered seeds; CH6 does so under W1 but not under the smaller W2 workflow, reported as such. This is a mechanism demonstration on open payload data with a synthetic metadata layer, not a claim about production prevalence.

cs.SE↗

arXiv2026

SARC-DQ: Runtime Data-Quality Gating for Agentic AI: Silent Evidence Defects, the Incompetence Shield, and Downstream-Only Remediation

Agentic systems act, so a defect in the evidence they retrieve becomes a wrong action with a currency cost. The most dangerous enterprise defects are metadata-borne: a stale price or a superseded record, perfectly well-formed in the payload and betrayed only by freshness, lineage, or provenance. Such a defect never enters the agent's context, and an agent cannot doubt data it cannot see. On a priced replenishment benchmark, a competent agent silently converts an injected metadata-borne defect into a costly action about 60% of the time, with zero data-quality flags and behavioral doubt markers at chance (AUC <= 0.50). Across four model tiers spanning roughly 15x in inference price, the rate stays flat: capability does not buy skepticism. A metadata-aware pre-action gate with downstream-only remediation recovers the loss fully on the signals its predicates cover and not at all on those they miss. A model-free oracle derived from the task's decision geometry tracks the measured rates with MAE 0.015 (Pearson r = 0.876, interval coverage 15/16 cells), giving the flat ladder an analytical form. Evidence integrity is a systems axis distinct from model capability; mitigation depends on enforcement placement and predicate coverage. Code, frozen results, and a deterministic analysis pipeline: https://github.com/besanson/dqSarc

cs.SE↗

arXiv2026

Green SARC: Predictive Cost and Carbon Governance for Agentic AI Systems

Agentic AI systems act through tools and sub-agents, yet the controls meant to bound their financial and environmental cost still sit on dashboards evaluated beside or after execution. Green SARC applies the SARC governance-by-architecture framework -- four enforcement sites in the agent loop -- to FinOps and GreenOps, contributing the theory of what to enforce and how to predict it. We report four policy-independent results. (i) The unconstrained "State Snowball" is $\Theta(n^2)$ in loop depth; on 3,000 real multi-step plans (SWE-rebench) it holds on 100%, with median curvature $\hat{c}_2=216$ exceeding the linear-accretion prediction $p/2=134$ -- real plans accrete faster than the model. (ii) On real residuals the Normal-$\sigma$ gate under-covers (92% at nominal 95%); split-conformal calibration holds (95.2%). (iii) A soft Lagrangian penalty tuned to the budget in expectation breaches it on 91.5% of seeds; the architectural gate breaches 0%. (iv) Under binding budgets the gate's over-budget incidence is 0% on synthetic and real (BurstGPT) arrivals. End-to-end token/USD/carbon savings (47--55%) are real but policy-dependent in magnitude -- set by a scope-cap knob, not by gate rejections. The library is open-source, dependency-free, and ships a regeneration script for every cited number.

cs.SE↗

arXiv2026

SARC: A Governance-by-Architecture Framework for Agentic AI Systems

Agentic AI systems increasingly act through tools, sub-agents, and external services, but governance controls are still commonly attached to prompts, dashboards, or post-hoc documentation. This creates a structural mismatch in regulated settings: obligations that must constrain execution are often evaluated only after execution has occurred. We introduce SARC, a runtime governance architecture for tool-using agents that treats constraints as first-class specification objects alongside state, action space, and reward. A SARC specification declares each constraint's source, class, predicate, verification point, response protocol, and operating point, and compiles these into four enforcement sites in the agent loop: a Pre-Action Gate, an Action-Time Monitor, a Post-Action Auditor, and an Escalation Router. We formalize the minimal invariants required for specification-trace correspondence, show why finite reward penalties do not generally substitute for hard runtime constraints, and extend the architecture to multi-agent workflows through constraint propagation, authority intersection, and attribution-preserving trace trees. We implement a prototype audit checker and report a reproducible synthetic evaluation over 50 seeds comparing SARC against post-hoc audit, output filtering, workflow rules, and policy-as-code-only baselines on a procurement task. SARC executes zero hard-constraint violations under exact predicates; its declared PAA throttling response reduces soft-window overages by 89.5% relative to policy-as-code-only. Predicate-noise and enforcement-failure sweeps are consistent with the claim that residual hard violations under SARC scale with enforcement-stack error rather than environmental violation opportunity. SARC provides the architectural substrate through which obligations can be made executable, inspectable, and auditable at runtime.

cs.SE↗

arXiv2026

The Inference Bottleneck: A Formal Model of Vertical Foreclosure in AI Markets

As generative AI commercializes, competitive advantage is shifting from model training toward inference, distribution, and routing. This paper develops a formal game-theoretic model of vertical foreclosure in inference markets, as the formal-model companion to Besanson and Celani (2026). The model isolates two foreclosure mechanisms operating without predatory pricing: quality-of-service (QoS) discrimination against downstream rivals via latency, throughput, context limits, or feature access; and routing bias in assistant-layer interfaces. An extension motivated by Anthropic's April 2026 release of Claude Opus 4.7 alongside the restricted-access Claude Mythos Preview introduces a third mechanism, tier-based access discrimination, parameterized by a tier gap (tau) and partner-exclusivity (kappa). The main result gives an explicit local equilibrium characterization of the QoS gap. Under logit demand and symmetric rivals, the gap is strictly increasing in inference-quality importance (alpha) and downstream margins, and strictly decreasing in API price and rival entry elasticity. Discrimination vanishes at a joint boundary rather than at a simple threshold in alpha alone. A stylized calibration to four providers using April 2026 data treats parameter values as inputs to a comparative risk mapping, not structural estimates. The mapping suggests Google and OpenAI face conditions most conducive to foreclosure; Microsoft's realized routing bias has been voluntarily constrained by a March 2026 multi-model pivot; Anthropic shows low consumer-channel risk and elevated risk in enterprise coding-agent segments. The policy section proposes Neutral Inference, a four-pillar conduct framework: QoS parity, routing transparency, FRAND-style non-discrimination, and tier transparency with release-pathway discipline. Illustrative welfare calculations suggest net gains in the tens of billions annually.

cs.CY↗

arXiv2026

The Data Hydration Gap: A Formal Model of Underinvestment in General-Purpose Data Products Under Decentralized Governance

When organizations decentralize data product ownership, as in the data mesh paradigm, each domain team optimizes for its immediate analytical needs, underinvesting in the cross-domain generality that enables organization-wide reuse. We formalize this as a simultaneous-move game in which N domains choose quality (q) and generality (g). Generality creates positive externalities but is privately costly. The Nash equilibrium generality gap is increasing in the number of domains and the value of cross-domain analytics. Under plausible parameter configurations, a corner solution obtains in which no reusable silver layer emerges organically, a condition we term the data mesh trap. Technical debt from narrow products grows quadratically in N. An illustrative calibration suggests non-trivial organizational welfare losses under plausible enterprise parameters. We derive within-model conditions under which centralized, federated, and hybrid governance regimes dominate, and we identify the information asymmetries and transaction costs that complicate implementation. The model provides a formal foundation for empirical research on decentralized data governance.

cs.DB↗

arXiv2026

The Inference Bottleneck: Antitrust and Neutrality Duties in the Age of Cognitive Infrastructure

As generative AI commercializes, competitive advantage is shifting from one-time model training toward continuous inference, distribution, and routing. At the frontier, large-scale inference can function as cognitive infrastructure: a bottleneck input that downstream applications rely on to compete, controlled by firms that often compete downstream through integrated assistants, productivity suites, and developer tooling. Foreclosure risk is not limited to price. It can be executed through non-price discrimination (latency, throughput, error rates, context limits, feature gating) and, where models select tools and services, through steering and default routing that is difficult to observe and harder to litigate. This essay makes three moves. First, it defines cognitive infrastructure as a falsifiable concept built around measurable reliance, vertical incentives, and discrimination capacity, without assuming a clean market definition. Second, it frames theories of harm using raising-rivals'-costs logic for vertically related and platform markets, where foreclosure can be profitable without anticompetitive pricing. Third, it proposes Neutral Inference: a targeted, auditable conduct approach built around (i) quality-of-service parity, (ii) routing transparency, and (iii) FRAND-style non-discrimination for similarly situated buyers, applied only when observable evidence indicates functional gatekeeper status.

cs.CY↗

arXiv2026

Accuracy Standards for AI at Work vs. Personal Life: Evidence from an Online Survey

We study how people trade off accuracy when using AI-powered tools in professional versus personal contexts for adoption purposes, the determinants of those trade-offs, and how users cope when AI/apps are unavailable. Because modern AI systems (especially generative models) can produce acceptable but non-identical outputs, we define "accuracy" as context-specific reliability: the degree to which an output aligns with the user's intent within a tolerance threshold that depends on stakes and the cost of correction. In an online survey (N=300), among respondents with both accuracy items (N=170), the share requiring high accuracy (top-box) is 24.1% at work vs. 8.8% in personal life (+15.3 pp; z=6.29, p<0.001). The gap remains large under a broader top-two-box definition (67.0% vs. 32.9%) and on the full 1-5 ordinal scale (mean 3.86 vs. 3.08). Heavy app use and experience patterns correlate with stricter work standards (H2). When tools are unavailable (H3), respondents report more disruption in personal routines than at work (34.1% vs. 15.3%, p<0.01). We keep the main text focused on these substantive results and place test taxonomy and power derivations in a technical appendix.

cs.AI↗