CTOs face fragmented GenAI policy, prompting a governance-broker market, arXiv argues

In a v1 arXiv preprint, Nathalie Baracaldo argues that GenAI policy enforcement has not kept pace with the rapid growth of chat and agent capabilities, and that policy means different things to different practitioners, producing siloed, non-portable controls that complicate audits and liability. The paper, From Alignment to Access Control: A Framework for GenAI Policy Enforcement, surveys the landscape of enforcement approaches and argues for a principled method to analyze and compare how policy is defined and applied across tools, data flows, and organizations.

The lede grounds the debate in an academic author rather than a corporate press release, signaling policy is now part of the research agenda.

Baracaldo's method emphasizes mapping policy intents to concrete enforcement points, distinguishing desirable outcomes from the enforcement surfaces used by products. The paper is described as a companion extension of USENIX Security 2026 Enigma talk and proposes a framework to dissect existing approaches, exposing where gaps permit policy drift and where controls remain portable across ecosystems.

It suggests a disciplined sequence: define policy objectives, catalog enforcement capabilities in each tool, and test portability across platforms under realistic risk scenarios. The conclusion is not a call for a universal standard but for a rigorous, auditable comparison across vendors.

A governance-broker economy emerges as a second-order effect

Baracaldo's framing implies a second-order economy: enterprises will increasingly rely on governance brokers who stitch policy across tools, data flows, and regulatory regimes. Rather than a single vendor delivering omnipotent controls, organizations will contract with intermediaries who inventory policy catalogs, assign risk scores, publish audit-ready reports, and orchestrate cross-platform enforcement.

The paper argues this is not a boutique service but a structural feature of GenAI adoption, particularly in regulated industries where auditors demand traceable decision trails and risk-adjusted access controls. In this sense, governance brokers reframe procurement as a policy-integration problem rather than a pure price play.

These brokers would operate across multiple layers: policy catalogs that map local rules to model actions, risk-scoring engines that quantify compliance exposure, and enforcement orchestrators that align tools from different vendors. The advantage, the preprint suggests, is interoperability: if a broker can translate a jurisdictional requirement into modular predicates that any tool can honor, enterprises avoid lock-in while preserving auditable controls.

But the price is evident in increased complexity and liability questions about who bears responsibility for misinterpretations when a policy is interpreted differently by a broker and a vendor.

Procurement, liability, and the audit trail From a procurement perspective, the fragmentation described by Baracaldo shifts budgets toward governance capabilities and away from a single platform's raw capabilities. Enterprises will buy cross-vendor policy enforcement as a service, layering it on top of tool stacks with SLAs that prioritize auditable enforcement, transparent risk scoring, and cross-organization reporting. Liability questions shift to the orchestrator layer: if a broker configures a rule that a tool misapplies, who is accountable—the vendor, the operator, or the broker? Regulators, the paper implies, will increasingly scrutinize governance interfaces and require verifiable audit trails as part of compliance programs.

That shift also invites a new class of contracts and technical terms. Interoperability becomes a procurement criterion: the ability to migrate policy logic across platforms without losing history or context becomes a key risk metric.

The paper contends that the hard part is not writing the rules but ensuring they survive platform migrations and data-flow reconfigurations. In practice, this means procurement will tilt toward modular policy interfaces, standardized metadata exchange, and explicit liability clauses tied to governance outcomes rather than to model performance alone.

Cross-industry pilots will stress-test the policy economy

Baracaldo points to pilots in regulated sectors—finance, healthcare, and critical infrastructure—as the testing ground for cross-platform policy enforcement. In those environments, auditors demand end-to-end traceability from policy intent to model action, and risk scoring must reflect both bias and governance risk across data channels.

A broker-led approach could provide the unifying layer that makes disparate vendor tools interoperable enough to pass regulatory scrutiny, though it would also heighten the need for transparency about how rules are translated and updated as regulations evolve.

Yet pilots will also reveal the limits of the broker model. If policy enforcement remains inherently local—shaped by data location, data-sharing agreements, and jurisdictional nuance—brokers may become expensive adapters rather than value creators.

The preprint warns that fragmentation is not an obstacle to standardization but a persistent condition of diverse risk postures and regulatory regimes. The outcome will hinge on whether governance interfaces can deliver auditable, platform-agnostic controls that regulators can trust.

Signals to monitor: 12-month tests for the broker thesis Executives should watch for a spread of cross-vendor enforcement catalogs and standardized audit trails, which would indicate the broker model gaining traction. A proliferation of procurement cases that favor modular, interoperable policy controls over monolithic toolchains would also signal a shift in how compliance budgets are allocated. Finally, regulators and auditors may begin treating governance brokers as part of the compliance stack, rewarding demonstrably auditable policy execution rather than vendor-specific feature sets. These signals would not prove a universal standard, but they would validate the market-for-governance brokers hypothesis.

On the other hand, if a universal framework emerges from a cloud provider or a regulatory body—an unlikely but possible development—the broker thesis would face an existential test. The paper’s emphasis on differing interpretations and local practice suggests that a single, one-size-fits-all solution is improbable, and the economics of modular governance would instead decide who wins in procurement terms, who carries liability, and how rapidly audits can be completed across ecosystems.

The coming year will tell whether policy enforcement becomes a portable capability or remains a market for bespoke orchestration.