Loading
Skip to content
Miracle Minds AI
  • Try “HIPAA”, “model risk”, “FinOps” or “agentic”.

Responsible AI Enablement

Policy alone does not change how people use AI. Bans and long rulebooks tend to push use out of sight; enablement pays off when the approved path is easier than the workaround and each role knows the few checks it owns.

Part of Security, Assurance & Governance, in the Adopt mode. 7 sources, checked .

of employees say they have uploaded company information, such as financial, sales or customer data, into public AI tools (47 countries, 2024 to 2025)
48%
of employees say they have uploaded company information, such as financial, sales or customer data, into public AI tools (47 countries, 2024 to 2025)
The University of Melbourne and KPMG, 2025(opens in a new tab)
of people who use AI at work had received AI training from their company (31 markets, 2024)
39%
of people who use AI at work had received AI training from their company (31 markets, 2024)
Microsoft, 2024(opens in a new tab)
of organizations named knowledge and training gaps as the top obstacle to responsible AI in 2025, up from 51%
59%
of organizations named knowledge and training gaps as the top obstacle to responsible AI in 2025, up from 51%
Stanford Institute for Human-Centered Artificial Intelligence, 2026(opens in a new tab)

The problem

Use has outrun guidance. In Microsoft's 2024 survey of 31,000 knowledge workers in 31 markets, 75% used generative AI at work and 78% of AI users brought their own AI tools, yet only 39% of people using AI at work had received AI training from their company.2 A 2025 University of Melbourne and KPMG study of 48,340 people in 47 countries found that only two in five employees said their organization had a policy or guidance on generative AI use.1

The risk shows up in behavior. About half of employees in that study (48 to 49%) said they had uploaded sensitive company information or copyrighted material into public AI tools, and 56% said they had made mistakes in their work because of AI.1 The telling finding is about rules: these behaviors were most common where the organization had banned generative AI (67%) or had a policy (56%), against 33% where there was none. The authors conclude that bans may be ineffective and that having a policy does not guarantee compliance.1

Organizations know where the gap is. In the 2025 survey behind the Stanford AI Index, knowledge and training gaps were the most cited obstacle to implementing responsible AI, rising from 51% to 59% of respondents.3 Regulation now asks for measures, not certificates. The EU AI Act's literacy article, as rewritten in July 2026, requires providers and deployers to take measures to support the AI literacy of their staff, and states that this does not require guaranteeing any specific level for any individual.4

What good looks like is role-based. The NIST AI RMF expects personnel and partners to receive AI risk management training that fits their duties.5 US federal policy requires periodic training for operators of high-impact AI that is specific to the system and how it is used.6 ISO/IEC 42005 gives teams a method to identify, evaluate and document how an AI system may affect individuals, groups and society,7 which is the backbone of a review checklist a builder fills in before launch.

What you get

  1. 01

    Role-based responsible AI standards

    Short, specific standards for builders, approvers, business users and buyers, each stating what to do, what to check and when to escalate, traced back to your AI policy.

  2. 02

    Review checklists and impact assessment templates

    Pre-launch checklists scaled by risk tier and an impact assessment template based on ISO/IEC 42005, so a team can show what it considered and what it changed as a result.

  3. 03

    Approved tools and a fast request path

    A published list of sanctioned AI tools with clear rules on what data may go into each, and a quick route to request a new one, so people have a legitimate option instead of a personal account.

  4. 04

    Training by role, measured by behavior

    Short modules and hands-on labs for each role, including how to check AI output before relying on it, followed by measures of behavior such as checklist quality and policy exceptions rather than attendance alone.

  5. 05

    Champions network

    A trained champion in each business unit who answers first-line questions, collects use cases and feeds recurring problems back to the governance council.

Where it goes wrong

The failure modes we design against from the first week, and the control for each.

  • Banning tools instead of offering safe ones

    Bans push use onto personal accounts where you see nothing. In survey data, uploading company data to public AI tools and breaking AI policy were most common where generative AI was banned.1 Offer approved tools with enterprise data protections and a quick way to request more.

  • One generic course for everyone

    A single annual module teaches principles but not what a developer, an underwriter or a buyer must actually do on Monday. Write standards and training per role, as the NIST AI RMF and federal system-specific training requirements imply.5,6

  • Teaching the rules but not judgment

    Two in three employees report relying on AI output at work without critically evaluating it.1 Training that covers only what is forbidden misses the bigger risk. Teach people how to check output, spot errors and decide when AI should not be used for a task.

  • Measuring attendance, not behavior

    Completion rates say nothing about whether people stopped pasting customer data into public chatbots. Track what changed: AI-related data-loss alerts, approved versus unapproved tool use, checklist quality on sampled launches and the volume of exception requests.

Standards and rules

How to measure it

  • Share of staff in AI-using roles who completed role-specific training, and knowledge-check pass rates
  • AI use through approved tools versus unapproved tools detected on the network
  • AI-related data-loss alerts per thousand users, quarter on quarter
  • Share of launches with a completed checklist and impact assessment that pass quality sampling
  • Median time to approve a new AI tool request

Questions buyers ask

Is AI literacy training still mandatory under the EU AI Act?

Yes, in a softer form. The Digital Omnibus adopted in July 2026 rewrote Article 4: providers and deployers must take measures to support the AI literacy of staff and others operating AI on their behalf, considering their knowledge and the context of use. It no longer requires you to guarantee a specific level of literacy for any individual. Keep records of what you provided and to whom, because you may need to show the measures were taken.

What should a responsible AI review checklist include?

Only the checks that change decisions: intended use and users, data sources and permissions, known failure modes, test results for relevant user groups, human oversight, how users report problems, and who signs off. Tier the checklist so a low-risk internal tool takes minutes and a customer-facing decision system gets the full review with an impact assessment. Attach it to the release process so it is finished before launch, not after.

How do we stop employees using unapproved AI tools?

Give them a better option first. Approve a small set of tools with enterprise data protections, publish clear rules on what data may go into them, and make requests for new tools quick. Then use data-loss prevention and network controls to spot the rest, and treat what you find as a sign of unmet demand as well as a policy breach. Bans on their own tend to hide use rather than stop it.

Sources

Every figure on this page comes from the sources below. We checked each link on .

  1. 1.Trust, attitudes and use of artificial intelligence: A global study 2025(opens in a new tab)The University of Melbourne and KPMG, 2025 · assets.kpmg.com
  2. 2.2024 Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part(opens in a new tab)Microsoft, May 8, 2024 · microsoft.com
  3. 3.The 2026 AI Index Report, Chapter 3: Responsible AI(opens in a new tab)Stanford Institute for Human-Centered Artificial Intelligence, 2026 · hai.stanford.edu
  4. 4.Regulation (EU) 2026/1744 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI)(opens in a new tab)European Parliament and Council of the European Union, July 8, 2026 · eur-lex.europa.eu
  5. 5.Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1(opens in a new tab)National Institute of Standards and Technology, January 2023 · nvlpubs.nist.gov
  6. 6.M-25-21: Accelerating Federal Use of AI through Innovation, Governance, and Public Trust(opens in a new tab)Executive Office of the President, Office of Management and Budget, April 3, 2025 · whitehouse.gov
  7. 7.ISO/IEC 42005:2025 Information technology, Artificial intelligence (AI), AI system impact assessment(opens in a new tab)International Organization for Standardization / International Electrotechnical Commission, May 2025 · iso.org