Skip to content
Securafy AI Lab by Securafy
AI Compliance & Regulated Industries

The EU Lowered the AI Literacy Bar. Your Risk Did Not Get Smaller.

The Digital Omnibus replaced Article 4's duty to ensure AI literacy with a duty to support it, just as national regulators gained enforcement power. Ric Hall explains why an effort standard puts the burden of proof on employers, when the rule reaches U.S. companies, and why Article 26's oversight competence duty for high-risk systems was left intact.

In this article

"The EU softened its AI literacy rule. Can we stop worrying about it?" It is a reasonable question for any company with a European subsidiary, EU-based staff, or customers there. The short answer is no, and the reason is not the one most people expect.

The EU AI Act's AI literacy duty did get lighter on paper. But the amendment arrived days before national regulators gained the power to enforce it, and the new wording shifts the question from "did your staff reach a level" to "what did you actually do." For an employer, that is a harder question to answer without records.

What changed in the EU AI Act's AI literacy rule?

Article 4 originally required providers and deployers to ensure a sufficient level of AI literacy among staff. The amended text requires them to "take measures to support the development of AI literacy" and states that this "does not require providers or deployers to guarantee any specific level of AI literacy of any individual." The obligation itself remains, has applied since February 2025, and is now enforced nationally.

The new wording appears in the European Commission's AI Act Service Desk text of Article 4, which also adds that the Commission and member states will support organizations, particularly smaller ones, including by publishing practical examples of compliance. The change was part of the Digital Omnibus on AI, which EU institutions provisionally agreed on May 6, 2026, according to Gibson Dunn's analysis of the agreement. It was then published as Regulation (EU) 2026/1744 on July 24 and entered into force on July 27, 2026, ahead of the Act's August 2 milestone.

The same package delayed the heavier obligations. Standalone high-risk systems, including AI used in employment, education, and credit assessment, moved from August 2, 2026 to December 2, 2027, and AI embedded in regulated products moved to August 2, 2028.

Why a lower bar can mean more exposure, not less

Here is the part that gets lost in the "relief" headlines. The old wording was demanding, but it was at least a target: show that staff reached a sufficient level. The new wording replaces the target with an effort standard. Effort standards are judged after the fact, by whoever is asking, against whatever the organization can prove it did. That moves the burden of definition from the regulator to the employer.

In practice, "we took measures" is only as strong as the evidence behind it. A company with dated, role-specific training records can show proportionate measures. A company that ran one optional webinar and has an attendance list for a third of its staff cannot, and the softer statute gives it fewer bright lines to point to when explaining why that was enough. The bar dropped, but the floor became your own documentation.

Who enforces AI literacy now, and what will they look for?

National market surveillance authorities in each member state, not the EU's AI Office. The Commission's AI literacy questions and answers state that supervision and enforcement of these rules apply from August 2, 2026, and that penalties rest on national laws member states were due to adopt by August 2025.

The same guidance is clear about what counts as evidence: no certificate is required, and organizations can keep "an internal record of trainings and/or other guiding initiatives." It also describes enforcement as proportionate, weighing the nature and gravity of an infringement and whether it was intentional or negligent, and it notes that sanctions might be more likely where there is proof of an incident caused by a lack of appropriate training and guidance. That last point is the real risk. Regulators are unlikely to audit your training calendar for its own sake. They are far more likely to ask about training after an AI-related incident, when the absence of records reads as negligence.

Does the EU AI literacy rule apply to U.S. companies?

It can. The Act's scope provisions in Article 2 cover deployers established or located in the EU, and providers and deployers in third countries "where the output produced by the AI system is used in the Union." A U.S. company with an EU subsidiary, staff working from the EU, or AI-generated output delivered to EU clients should assume Article 4 reaches at least part of its workforce.

That reach often arrives through ordinary operations rather than a deliberate EU strategy. A shared Microsoft 365 tenant with Copilot enabled for a European sales team, an AI-assisted support desk answering EU customers, or a recruiting tool screening EU applicants can each bring EU staff and EU output into scope. If your compliance map still treats AI rules as a U.S. state problem, our overview of which AI rules actually apply to your business is a useful place to recheck the assumption.

The high-risk duty the amendment did not soften

For employers, the more consequential competence rule is not in Article 4 at all. Article 26(2) requires deployers of high-risk AI systems to "assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support." The Omnibus delayed when that obligation starts for standalone systems such as hiring and credit tools. It did not dilute the standard.

That gives organizations a planning window, not a reprieve. By December 2027, a company using AI to screen candidates or assess creditworthiness for EU-related decisions will need named overseers who understand the system, have been trained on it, and can actually intervene. Building that capability takes longer than writing a policy, and the general literacy measures you take now under Article 4 are the foundation it sits on.

What the workforce data says about the gap

The distance between AI use and AI preparation is still wide. LexisNexis surveyed 1,400 professionals across more than 20 sectors for its Future of Work 2026 report and found that 53 percent had used generative AI without formal approval, 28 percent said their organization had no formal generative AI policy, and 19 percent had received no AI training, even as 64 percent described themselves as very or extremely confident using the tools responsibly.

That combination of high confidence and thin preparation is precisely the profile that produces incidents: regulated or personal data pasted into an unapproved tool, an unreviewed output sent to a client, a biased recommendation accepted because it looked authoritative. It is also the profile that makes "we took measures" hard to defend afterward, because unapproved use means the organization often cannot say which tools were involved, let alone who was trained on them. We dig into that visibility problem in the AI skills gap that is really shadow AI you cannot see.

What proportionate AI literacy measures look like

Article 4 itself tells you what the measures should account for: staff's technical knowledge, experience, education and training, the context the systems are used in, and the people the systems affect. Turned into an operating program, that points to a short set of steps:

  • Inventory the AI systems in use, including features embedded in existing platforms, and which staff use each one.
  • Give everyone a common baseline, then deeper training for roles whose AI use affects customers, applicants, or regulated data.
  • Keep a dated record of who received which training, on which tool, with what content.
  • Identify now the people who will need Article 26 oversight competence when high-risk obligations begin.
  • Refresh content when tools or uses change, and record that you did.

The tiering matters because a proportionate program is not a uniform one. The difference between AI training for employees and AI training for leaders is a good model: leaders need to understand accountability and oversight, while frontline staff need practical judgment on the tools they use every day. Data handling deserves its own emphasis, since the most common AI incident is still sensitive information going where it should not, a theme we cover in data privacy in the age of generative AI.

None of this is unique to the EU. Treating AI literacy as an operational discipline rather than a compliance memo is the same shift we argue for in why AI governance is becoming an operational discipline, and it is why AI adoption is a governance problem, not a technology one. A program built well for Article 4 will also serve you under U.S. state rules, insurer expectations, and client due-diligence questionnaires.

How Securafy helps organizations with EU exposure

For clients with any EU footprint, we start by establishing which people and which AI outputs actually fall within scope, using tenant configuration, licensing, and application data rather than an org chart. That usually surfaces a smaller, more specific population than leadership expects, plus a few AI features that nobody realized were reaching EU users.

From there we help build tiered literacy measures matched to that population, set up training records that live alongside the AI system inventory, and flag the roles that will need Article 26 oversight competence before December 2027. The goal is a file that answers a market surveillance authority's question in one pass: which systems, which people, which training, and when.

Where To Go From Here

Treat the softer Article 4 as an invitation to define your own standard, then document it. The companies best positioned under the amended rule are the ones that can show proportionate measures by role, not the ones waiting for Brussels to say what counts.

If your team is moving faster with AI than your guardrails are, start with structured training rather than another tool. AI University gives your people role-based AI training with security built into the material, not bolted on afterward.

If you would rather talk through your specific environment first, book a strategy call with Securafy and we will walk your current AI usage, exposure, and the fastest path to safe adoption.

Ric Hall
Ric Hall

Ric Hall is the Chief Revenue Officer at Securafy, with decades of experience in enterprise infrastructure, cloud technology, sales leadership, and business strategy.

He writes for leaders trying to make sense of big technology decisions without getting trapped in vague promises or polished sales language. His articles cover provider selection, IT budgeting, co-managed services, cybersecurity investments, modernization, and the questions businesses should ask before signing a contract.

Ric’s strength is connecting technical decisions to business outcomes, helping leaders understand not just what they are buying, but why it matters and whether it will still make sense 3 years from now.

Writes about: IT budgeting, provider evaluation, cybersecurity ROI, co-managed IT, cloud modernization, vendor selection, technology strategy

More from Ric Hall

Learn AI by building with it

AI University helps teams move beyond AI curiosity through practical lessons, secure workflows, guided experiments, and real projects built for everyday business use.

Explore AI University

Stay current on practical business AI

Get practical updates on AI security, governance, tools, compliance, and implementation without the daily hype cycle.

Join the conversation

Have a question or a different take on this? Add it below.