In this week’s digest, we reported that the DoL defined baseline AI literacy, and House lawmakers proposed a 30 percent tax credit for qualified AI training. At the same time, public companies disclosed formal AI governance committees and risk-tiered controls. For senior enterprise L&D leaders, AI training is shifting from discretionary programming to finance- and audit-classified investment requiring tighter scope and documentation.
1. How Do This Week’s Federal Signals Change the Risk Profile of Enterprise AI Training?
Accenture has trained roughly 500,000 employees on classical AI. Cognizant reports training 260,000. EPAM says more than 80 percent of its workforce has completed AI upskilling. Scale has become the signal. This week, two developments changed what that scale means.
On February 13, the Department of Labor released a voluntary AI literacy framework defining baseline competencies, including directing AI systems effectively and using them in ethical and secure ways.
The same week, House lawmakers introduced the AI Workforce Training Act, proposing a 30 percent tax credit for qualified employer-funded AI training, including data literacy, prompt engineering, machine learning fundamentals, and AI ethics.
Neither development mandates enterprise training design. However, both establish external reference points that can influence procurement, reporting, and tax treatment.
Public company earnings calls indicate that AI spending is already being classified in financial terms.
Two Harbors’ CFO stated that “a lot of what we’re doing is going to be expensed rather than capitalized,” and added that capitalization rules are “quite strict.”
Citizens Financial Group described a “meaningful chunk of overlay for tech spend” that will “wind up in our depreciation line over time.”
Innodata management referred to incremental Gen AI-related operating expenses as costs they “think of as investments.”
These examples show that AI-related expenditures are being parsed as operating expense, capitalized investment, and depreciation impact.
If AI training becomes tax-eligible or benchmarked against federal literacy definitions, senior enterprise L&D leaders should expect categorization, documentation, and defensibility requirements to increase.
Broad enterprise bootcamps can complicate this environment because they increase aggregate spend while reducing clarity about which roles require which level of capability. This is an analytical conclusion based on the interaction between external definitions and financial classification practices disclosed in earnings calls.
For a senior L&D leader in a large enterprise where the Chief Information Officer drives AI deployment and the Chief Financial Officer oversees capital allocation, AI training may shift from programmatic initiative to finance-classified line item. The immediate risk is expanding scope before clarifying role-based need, documentation standards, and alignment with external definitions.
2. How Are Boards and Risk Committees Structuring AI Governance in 2025-2026?
Public filings from late 2025 and early 2026 indicate that AI governance is being formalized through defined committees, risk-tiered controls, and policy acknowledgment requirements.
WEX disclosed in its 10-K the establishment of an “AI Systems & Model Governance Committee” composed of representatives from risk, compliance, privacy, security, and technology. Its Technology Committee reviews AI use to ensure alignment with ethical and regulatory obligations.
IQVIA described differentiated controls for “higher risk AI activities,” which require formal evaluation for error, bias, and hallucinations. For “lower risk activities,” IQVIA relies on standard operating procedures and employee training as part of its control framework.
ACI Worldwide requires 100 percent of employees and contractors to review and acknowledge its Artificial Intelligence Policy. This requirement positions training and policy acknowledgment as compliance artifacts rather than engagement initiatives.
Risk language in 10-K filings also reflects elevated scrutiny. SoFi cites obligations under the EU Artificial Intelligence Act and references regulatory attention to generative AI. Upstart identifies regulatory uncertainty around model “explainability” as a risk factor and notes that regulators may require demonstrations of robust governance and validation.
MSCI reports that its Audit Committee receives updates on AI-related risks. WESCO warns that documentation or testing gaps could result in compliance failures and enforcement exposure.
These disclosures show that boards, audit committees, and cross-functional governance bodies are treating AI as a formal risk domain. Training appears in filings as a component of control frameworks, especially where risk tiering differentiates high-risk and lower-risk use cases.
For enterprise L&D leaders, the implication is that the capability under scrutiny is supervisory and procedural: escalation protocols, validation standards, documentation practices, and alignment with internal AI policies. If AI governance structures are formalized outside the L&D function, training that is not explicitly tied to risk tiering and accountability may be peripheral to core oversight mechanisms.
3. How Should Senior L&D Leaders Redesign AI Training Portfolios Under Finance and Governance Scrutiny?
The practical response to increased financial classification and governance oversight is to segment AI training into risk- and value-based tiers rather than treating AI capability as a single enterprise-wide category.
Recent disclosures show three converging pressures:
Finance leaders are parsing AI spend as operating expense versus capitalized investment.
Governance committees are formalizing AI oversight structures.
Federal frameworks are defining baseline literacy externally.
In this environment, a defensible AI training portfolio can be structured into three tiers.
Tier 1: Documented Baseline LiteracyBaseline literacy aligns to the Department of Labor framework and internal AI policy. This tier typically consists of short modules, policy acknowledgment, and completion tracking. ACI Worldwide’s requirement that 100 percent of employees review and acknowledge its AI policy illustrates how baseline literacy can function as documented compliance coverage. The objective at this tier is documented coverage rather than depth of technical skill.
Tier 2: Workflow-Specific EnablementWorkflow-specific enablement targets business processes materially altered by AI during the current planning period, such as underwriting, customer support scripting, sales pipeline analysis, claims review, or code generation. Cohorts are limited to roles directly tied to those workflows. Measurement links to defined performance indicators, such as cycle time, error rates, productivity metrics, or revenue per representative. When Innodata frames Gen AI-related expenses as “investments,” the implied expectation is measurable return; narrowing enablement to specific workflows supports observable impact.
Tier 3: Managerial Oversight and AccountabilityManagerial oversight training targets supervisors responsible for higher-risk AI use cases. In IQVIA’s risk-tiered framework, higher-risk activities require additional controls; training architecture can mirror this differentiation. Oversight-focused training addresses escalation protocols, validation requirements, documentation standards, and acceptable-use boundaries. When WEX establishes an AI Systems & Model Governance Committee spanning risk and compliance, governance authority is distributed across functions. Without explicit supervisory training, control mechanisms may operate independently of L&D strategy.
For senior L&D leaders in large enterprises, the objective is to reduce ambiguity in scope and classification. Before the next budget cycle, leaders can pressure-test their portfolio by asking:
Can each AI training dollar be mapped to a defined role cohort and risk tier?
Is baseline compliance coverage clearly separated from value-generating enablement?
Are governance-supporting programs distinct from productivity-oriented initiatives?
If segmentation is unclear, expansion may increase financial and audit vulnerability. If segmentation is explicit, narrowing can support defensibility under finance and governance review.
Large-scale AI training disclosures from 2025 reflect a phase of rapid expansion. Public filings and policy signals from early 2026 indicate a shift toward classification, oversight, and return on capital. Designing a structured, role-specific AI capability portfolio is a risk-management and capital-allocation response to that shift.
Learning and Development Executive Intelligence is for CHROs, CLOs, and senior L&D buyers investing in internal talent development, training, and reskilling.
This is one of our six education and learning-related publications spanning K-12, Higher Education, and Workforce. Our education newsletters reach tens of thousands of senior decision-makers across the U.S. and key international markets.
Ping us at [email protected] if you’d like to learn more, explore Enterprise Subscriptions, or would like to partner in other ways.
The Intelligence Council is a next-gen B2B media and business intelligence platform built for people who make strategy, allocate capital, and carry operating risk.