The Board Bot Is Only Getting Started

When AI moves from preparing directors for meetings to interrupting them inside one, everything we understand about boardroom accountability changes overnight.

Audio summary: https://youtu.be/toA5iY4lrEY

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://open.substack.com/pub/johanosteyn/p/the-board-bot-is-only-getting-started?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

In April 2026, Lloyds Banking Group became what is believed to be the first FTSE 100 company to deploy a specialist AI agent in its boardroom. The tool, built by London-based firm Board Intelligence, ingests confidential board packs, helps directors interrogate hundreds of pages of material in conversational language, flags inconsistencies across papers, and surfaces analysis spanning cybersecurity, sustainability, financial performance, and mergers and acquisitions. Nicola Putland, Lloyds’ corporate governance director, described it as enabling “faster analysis and access to a broader range of perspectives.” The coverage that followed was almost uniformly celebratory.

Nobody asked the harder question. What comes next?

CONTEXT AND BACKGROUND

Board Intelligence CEO Pippa Begg was candid about the roadmap. The current deployment is, in her words, “step one” — AI as a preparation tool, helping individual directors consume information and test their judgements before they enter the room. Step two is something altogether different. In a later phase, she suggested, the agent could sit inside live meetings and, in her words, “almost interrupt and say: ‘Hang on, I think you’re falling into this trap.’ Or: ‘I disagree.’” Begg ruled out giving the AI a formal legal vote, calling that “a dangerous leap.” She did not rule out the rest.

The distance between a document summarisation tool and an AI that interrupts the chair of a listed company’s board to say it disagrees is not a technical increment. It is a governance revolution. The first of those two things can be justified relatively easily within existing frameworks — boards have always used external analysis, advisers, and research tools to inform their preparation. The second has no precedent in corporate governance anywhere in the world, and the frameworks that will eventually govern it do not yet exist.

The Lloyds deployment lands in a broader landscape that is moving faster than most boards realise. Research presented at Governance Intelligence’s 2026 briefing on AI in the boardroom found that 64 per cent of US public company directors are already using AI in some form to support their board work. More consequentially, 42 per cent of those directors are relying on free, consumer-grade tools — systems with no data isolation, no audit trail, and no governance architecture — to process materials that include the most confidential information their companies produce. The boardroom AI story is not only about Lloyds. It is about what is already happening informally, invisibly, and without any of the safeguards that make the Lloyds deployment defensible.

INSIGHT AND ANALYSIS

The governance questions that step two raises are not hypothetical. They are structural, and they touch the foundations of how boards are supposed to work.

Corporate governance frameworks — the UK Corporate Governance Code, South Africa’s King V, and equivalent codes across major economies — are built on a set of assumptions about the nature of board deliberation. Directors are expected to exercise independent judgment. They are expected to challenge management. They are expected to bring diverse perspectives into the room and test them against each other in a process of genuine deliberation. The minutes of a board meeting are a legal record of who said what, and why, and what was decided. That record is the accountability infrastructure through which shareholders, regulators, and courts can assess whether a board discharged its duties.

When an AI agent intervenes in that deliberation — flagging cognitive traps, surfacing contradictory data, expressing disagreement — several things happen simultaneously. The agent’s contribution is not that of a director who can be identified, held accountable, or asked to explain their reasoning under oath. It is not that of an adviser whose report can be requested and scrutinised. It occupies a new category, one that existing governance frameworks have no language for and no mechanism to govern. What goes in the minutes when the AI interrupts? Whose judgment is it when a director changes their position in response to the agent’s intervention? And when the agent is wrong — as it will sometimes be — who answers for the consequences of a board decision that was materially shaped by an incorrect AI output?

I have previously written about the risk of AI sycophancy, the tendency of AI systems to tell powerful people what they want to hear rather than what is accurate or useful. That risk is especially acute in the boardroom, where the most senior and self-confident people in an organisation are making decisions under pressure. A board agent that intervenes in live meetings is operating in precisely the environment where sycophancy is most dangerous. If it consistently validates the direction the chair is already moving, it does not challenge groupthink. It amplifies it, with the additional authority of an AI’s apparent objectivity.

IMPLICATIONS

South Africa’s King V Code on Corporate Governance, published by the Institute of Directors in South Africa on 31 October 2025 and effective for financial years beginning on or after 1 January 2026, speaks directly to this moment. King V’s Principle 10 requires boards to oversee AI systems with clear accountability for designs, actions, outputs, and outcomes, and mandates that all automated technologies be subjected to human oversight and override mechanisms commensurate with the level of risk to the organisation and its stakeholders. It is no longer sufficient, King V states, for directors to defer AI oversight to IT or compliance teams. Boards now have a duty to understand the technology, its implications, and its impact on stakeholders.

That duty, applied to a board agent operating in live meetings, raises questions that King V’s drafters almost certainly did not anticipate. How does a board demonstrate human oversight of a tool that is itself participating in the oversight process? How does it document the AI’s contribution to a decision in a way that satisfies the Code’s “apply and explain” transparency requirement? And how does it ensure that the override mechanism — the board’s ability to disregard or contradict the AI’s intervention — is genuinely exercised, rather than allowed to erode under the accumulated weight of an agent that is always present, always confident, and always available?

The liability dimension has sharpened considerably in 2026. Governance experts now warn that the risk calculation has inverted: boards that fail to leverage secure, structured AI tools for research, scenario analysis, and decision support could themselves be seen as falling short of their duty of care. But that argument applies to the document preparation use case, not to an agent that intervenes in live deliberation. The second use case introduces a category of AI-assisted decision-making for which no liability framework yet exists, no case law has been established, and no regulator has issued guidance.

Lloyds has estimated that AI tools generated £50 million in value for the organisation in 2025, with a target of £100 million in 2026. That financial logic will accelerate boardroom AI adoption across the FTSE 100 and, in time, across South African listed companies. The pressure from institutional investors and governance advisers will follow. The question for South African boards is not whether to engage with this technology. It is whether the engagement will be structured, governed, and genuinely accountable — or whether the tools will arrive before the frameworks, as they almost always do.

CLOSING TAKEAWAY

Step one is defensible. A tool that helps directors read faster, ask sharper questions, and enter the boardroom better prepared is a sensible application of AI to a genuine problem. Step two — the agent in the room, interrupting, disagreeing, shaping deliberation in real time — is a different proposition entirely, and the governance frameworks that should govern it do not yet exist in any jurisdiction.

The Lloyds deployment has opened a door. What matters now is not whether other boards will walk through it — they will — but whether they will do so with their eyes open to what the next step actually requires. Corporate governance exists to ensure that power is exercised with accountability. When an AI begins to exercise influence over the decisions of the most powerful people in an organisation, accountability cannot be assumed. It must be designed, documented, and enforced. The board bot is only getting started. The governance frameworks that should govern what it becomes need to start now.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net