KESHO PARTNERS
Enterprise AI governance and planning

AI governance

Enterprise AI Readiness Guide

In most companies, AI did not arrive through a strategy. It arrived through staff using public tools, vendors switching on features, and teams running pilots. The company is accountable for all of it.

By Kesho Partners

10 minute read

Few countries have a general AI law in force, and several AI laws were delayed, replaced or withdrawn in 2026. It would be easy to conclude that AI governance can wait. It cannot. A tribunal held an airline responsible for its chatbot's advice. A court let a hiring-bias case against an HR software vendor proceed on behalf of applicants nationwide. A finance employee paid out US$25 million to a deepfake. None of these needed an AI law. They needed ordinary law, applied to systems the companies had not governed.

AI is everywhere. Value and control are not.

In McKinsey's 2025 survey, 88% of organizations used AI in at least one business function. Only 39% reported any effect on enterprise-level profit, and most of those put it below 5%.

Meanwhile, staff are not waiting. In the Microsoft and LinkedIn Work Trend Index, 78% of AI users brought their own tools to work, and over half were reluctant to admit using AI for their most important tasks. IBM found that one in five organizations had suffered a breach linked to this "shadow AI", and that high levels of it added about US$670,000 to the average breach cost. Of organizations with an AI-related security incident, 97% lacked proper AI access controls.

In plain terms: most companies have more AI than they know about, and less value than they hoped for.

Sources: [1], [2], [3]

Four ways ordinary companies get caught out

Your chatbot speaks for you.
An Air Canada chatbot told a customer he could claim a bereavement refund after travel, contrary to the airline's policy. The airline argued it was not responsible for the chatbot. A tribunal disagreed and ordered it to pay.
Buying a tool does not transfer the risk.
In Mobley v. Workday, applicants allege an AI screening tool discriminated against older, Black and disabled candidates. A US court allowed the age claim to proceed as a nationwide collective action in 2025 and rejected most of Workday's latest motion to dismiss in June 2026. The merits are undecided. Employers still own their hiring decisions.
Seeing and hearing is no longer proof.
An Arup employee in Hong Kong made 15 transfers totaling HK$200 million after a video call where every other participant was a deepfake. The control that failed was a payment process that trusted a face on a screen.
Generative AI needs a reviewer, not just a user.
Deloitte Australia refunded part of a government contract after a report was found to contain non-existent references and a fabricated court quote. The department said the findings stood. The damage was to trust.

Sources: [4], [5], [6], [7], [8]

The rules are moving, not going away

What changed in 2026
JurisdictionWhat happenedWhat it means
European UnionHigh-risk AI Act duties, including AI used in hiring and HR, moved to December 2027A delay, not an exemption. Use the time.
United KingdomNew rules on automated decisions in force from February 2026Allowed with safeguards: notice, the right to contest and meaningful human review.
United StatesColorado replaced its AI Act with a narrower law from January 2027; NYC hiring-tool audits under pressure to be enforcedMost exposure comes from existing discrimination and consumer law.
KenyaDraft data protection guidance on AI in July 2026; AI Bill rejected by a Senate committee in October 2026Data protection law already applies, and the draft names hiring, monitoring and generative AI as high-risk.
South AfricaDraft national AI policy withdrawn in June 2026POPIA's limits on automated decisions still apply.

The common thread is people decisions. AI used to hire, manage, monitor or decide what customers get is the area almost every jurisdiction treats as high-risk.

Sources: [9], [10], [11], [12], [13]

Agents change the question from "what does it say" to "what can it do"

A chatbot gives advice. An agent sends the email, changes the record or makes the payment. The security community's list of AI risks names "excessive agency", meaning too much functionality, permission or autonomy, as a leading cause of damaging actions. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027 because of cost, unclear value or weak risk controls.

The controls are not exotic: give each agent only the access it needs through its own identity, require approval for consequential actions, cap spend and volume, treat content it reads as untrusted, log every action, and test the off switch. We cover these in more depth in our agent risk assessment article.

Sources: [14], [15]

Five questions every board should ask

  1. Show me every AI system we use, including AI inside our software, with an owner and a risk rating.
  2. Which AI tools are staff using that we have not approved, and what data goes into them?
  3. Where does AI influence a decision about an applicant, an employee or a customer, and how do we know it is fair?
  4. Which AI systems can act on their own: send messages, change records or spend money?
  5. If an AI system made a costly mistake tomorrow, how would we find out, and who would decide what to do?

Where to start

Start with what is already there. Reconcile an inventory against licenses, expenses and network logs, and ask staff what they use. Give people an approved tool good enough to replace personal accounts. Find every AI tool that screens or scores people and test its outcomes. Limit what agents can do. Fix payment approvals so a convincing voice is never enough. Then take the results to the board as decisions with owners and dates.

Our guide covers the full assessment for companies in any sector, with the evidence to request, red flags, 42 questions and a 100-day plan. Banks, insurers and healthcare providers will find sector-specific guides alongside it.

The Enterprise Guide to AI Readiness and Governance

Sources

  1. McKinsey & Company. The state of AI in 2025, November 2025.
  2. Microsoft and LinkedIn. 2024 Work Trend Index, May 2024.
  3. IBM. Cost of a Data Breach Report 2025, July 2025. Press release.
  4. American Bar Association. BC tribunal confirms companies remain liable for information provided by AI chatbot, February 2024.
  5. Holland & Knight. Federal court allows collective action lawsuit over alleged AI hiring bias, May 2025.
  6. Freeman Mathis & Gary. Court largely denies Workday's motion to dismiss, June 2026.
  7. South China Morning Post. Arup confirmed as victim of HK$200 million deepfake scam, May 2024.
  8. The Register. Deloitte refunds Aussie gov after AI fabrications, October 2025.
  9. Taylor Wessing. The EU Digital Omnibus on AI, May 2026.
  10. Bratby Law. Data (Use and Access) Act 2025: changes now in force, 2026.
  11. Ropes & Gray. Colorado scales back AI law, May 2026.
  12. ODPC Kenya. Draft Guidance Note on AI, July 2026. Techweez: Senate committee rejects AI Bill, October 2026.
  13. SAnews. Cabinet approves withdrawal of AI policy, June 2026.
  14. OWASP Top 10 for LLM Applications 2025. Summarized by HackerOne.
  15. Reuters, via The Star. Over 40% of agentic AI projects will be scrapped by 2027, Gartner says, June 2025.

This article is general information, not legal advice.

Related service

AI governance and assurance

Independent AI readiness and governance assessments that separate normal gaps from real risks, and turn findings into a 100-day plan.

Explore the service