Article 4 of the EU AI Act entered into force on February 2, 2025 – making AI literacy the first EU AI Act obligation to apply in practice. It requires providers and deployers of AI systems to “take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf, taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in.” As Linklaters’ analysis notes, no direct fines apply specifically for Article 4 violations – but non-compliance creates civil liability exposure (from August 2025 when sanctions provisions activate), and regulators will cite obvious non-compliance in any later investigation. The standard is flexible: not every employee needs the same depth of training, and the Act calls for a layered, role-differentiated approach.
Any organization with operations in the EU, employees based in the EU, or AI systems deployed to EU users should treat Article 4 compliance as an immediate operational requirement, not a future aspiration.
Workplace Monitoring and Surveillance
AI-enabled workplace monitoring – tracking employee keystrokes, screen activity, communication patterns, and even facial expressions – is expanding rapidly and attracting regulatory attention.
The EU AI Act explicitly prohibits AI systems that use biometric data to infer emotions of natural persons in the workplace (with narrow exceptions), under its prohibited AI practices provisions (Article 5), which entered into force on February 2, 2025.
Ontario’s Bill 149 requires employers to disclose AI use in hiring processes – a transparency norm that is beginning to extend to broader employment contexts as regulators become more focused on the employee experience of AI surveillance.
The Illinois Biometric Information Privacy Act (BIPA) applies directly to employers collecting biometric data (including facial geometry and voiceprints) from employees, with a private right of action and statutory damages per violation that has already generated class action settlements in the tens of millions of dollars.
Labour Relations and Collective Bargaining
The 2023 strikes by the Writers Guild of America and SAG-AFTRA produced the first major collective bargaining agreements to address AI in detail. The WGA agreement prohibits AI from writing or rewriting literary material; AI-generated content provided to writers is not considered “assigned material” for compensation purposes; and companies must disclose to writers whether any provided content was generated by AI. The SAG-AFTRA agreement requires explicit informed consent before studios can use digital replicas of actors’ images or voices, and establishes compensation obligations for such use. Both agreements require an ongoing AI monitoring committee with authority to propose updates to keep pace with technological developments.
These agreements are models for how AI governance concerns will be integrated into collective agreements across industries as AI becomes a standard feature of work. Canadian executives in unionized environments – including in financial services – should anticipate AI-related bargaining demands when agreements come up for renewal.
Change Management and Workforce Reskilling
The data on workforce adaptation to AI is consistent: investment in training is insufficient relative to the pace of deployment. BCG’s AI at Work 2025 report identifies underinvestment in training as a leading barrier to realizing AI’s value: “Stop underestimating the importance of training. Commit appropriate levels of investment, time, and leadership support.” McKinsey’s workforce research identifies training and knowledge gaps as a leading barrier to responsible AI implementation. Deloitte’s 2026 Global Human Capital Trends frames AI-driven workforce transformation as a structural challenge requiring reskilling investment to be integrated into business strategy, not treated as a one-off training exercise.
Human-in-the-Loop: What “Meaningful” Actually Means
“Human-in-the-loop” (HITL) is a governance term for design approaches that preserve the ability of a human to review, modify, or override an AI system’s output before it becomes a binding decision. The term is frequently invoked in governance policies without meaningful operationalization.
EU AI Act Article 14 provides the most precise regulatory definition of what effective human oversight actually requires for high-risk AI systems. It specifies that human overseers must be enabled: to properly understand the system’s capacities and limitations; to remain aware of the risk of automation bias (the tendency to over-rely on AI outputs); to correctly interpret the system’s outputs; to decide in any given situation not to use the system or to override its output; and to intervene and stop the system through a “stop button or similar procedure.” This is a demanding standard – it requires not just that a human is nominally present in the workflow, but that they have sufficient information, training, and authority to actually exercise meaningful oversight.
Training Program Design
Effective AI literacy programs are role-differentiated, not one-size-fits-all. A practical curriculum structure follows:
| Audience | Core Topics |
| All staff | What AI tools are approved, what shadow AI means and why it is prohibited, how to report AI-related concerns |
| Managers | Oversight obligations, how to evaluate AI outputs, escalation procedures, regulatory requirements in their function |
| Developers and data teams | Responsible AI design, bias testing methods, documentation requirements, model cards |
| Legal, compliance, privacy | Current regulatory landscape, EU AI Act Article 4 compliance, sector-specific obligations |
| Board and senior executives | Strategic risk, governance oversight obligations, board-level reporting on AI risk |
Training completion rates and currency (ensuring training is refreshed annually or when AI systems change materially) are among the most trackable KPIs in an AI governance program.
