> ## Content Index > Fetch the complete content index at: https://www.aiacceleratorinstitute.com/llms.txt > Use this file to discover other available public pages before exploring further. # 7 questions your board will ask about AI before you're ready to answer them - URL: https://www.aiacceleratorinstitute.com/7-questions-your-board-will-ask-about-ai-before-youre-ready-to-answer-them/ - Published: 2026-09-17T14:41:37.000Z - Updated: 2026-09-17T14:41:37.000Z - Description: Boards used to nod along when AI came up in the strategy deck. That era ended. Directors now show up with specific questions, and a vague answer about "efficiency gains" reads to a board the way "the check is in the mail" reads to a landlord. - Author: Andrew Lovell - Tags: AI Infrastructure, Articles [**Protiviti and BoardProspects**](https://www.protiviti.com/au-en/press-release-ai-board-meeting-discussions-global-survey) surveyed 772 board members and executives in late 2025 and found that only 26% of boards discuss AI at every meeting, yet organizations with AI on every agenda see meaningfully higher returns than those that treat it as an occasional item. The gap runs deeper: [**the same survey**](https://www.protiviti.com/au-en/press-release-ai-board-meeting-discussions-global-survey) found 63% of boards in high-ROI organizations put AI on every agenda, against just 13% in the low-ROI group. [**Nasdaq's Governance Pulse survey**](https://www.governance-intelligence.com/boardroom/awareness-accountability-how-boards-are-approaching-ai-oversight) found AI and machine learning jumped to a top-issue concern for 40% of respondents, up from 19% two years earlier. The direction is clear. The questions are getting sharper along with the attention. --- ## The stakes just turned legal, well beyond reputational This shift runs deeper than looking prepared in a meeting. Under [**Delaware's Caremark doctrine**](https://www.mondaq.com/unitedstates/new-technology/1748354/managing-legal-risk-in-the-age-of-artificial-intelligence-what-key-stakeholders-need-to-know-today), directors can face personal liability for a sustained failure to implement or monitor adequate oversight systems. Courts have increasingly extended that doctrine to technology risk. Legal analysts now describe AI, where it touches core business functions, as exactly the kind of mission-critical risk Caremark was built to cover. [**Common deployment mistakes**](https://www.aiacceleratorinstitute.com/6-mistakes-ai-leaders-keep-making-with-agentic-deployments/) compound this exposure further, since ungoverned rollouts are precisely the pattern regulators and plaintiffs' lawyers are now targeting. [AI leaders who skip this meeting keep losing budget in Q3Nearly every organization running AI in production is overspending against forecast. The fix rarely means a smaller budget. It means a meeting most AI leaders keep skipping.![](https://storage.ghost.io/c/26/b3/26b323cb-c378-4831-bc7d-27e29def746a/content/images/icon/AIAI-ICON-fc3097bb-61c3-4dca-a622-25560dec8984.png)AI Accelerator InstituteAndrew Lovell![](https://storage.ghost.io/c/26/b3/26b323cb-c378-4831-bc7d-27e29def746a/content/images/thumbnail/AIAI_Website_Article_Images_Doodles--3--6-1bd13503-9ca0-4dfe-b133-677d7a2f3b98.png)](https://www.aiacceleratorinstitute.com/ai-leaders-who-skip-this-meeting-keep-losing-budget-in-q3/) ## The disclosure gap is already producing lawsuits The numbers make the exposure concrete: - [**A recent analysis**](https://www.alston.com/en/insights/publications/2026/06/how-boards-can-shrink-the-ai-governance-gap) found that 72% of S&P 500 companies now identify AI as a material risk in their Form 10-K filings, up sharply since 2023. - [**The same analysis**](https://www.alston.com/en/insights/publications/2026/06/how-boards-can-shrink-the-ai-governance-gap) found that roughly 8% of the 3,048 Russell 3000 and S&P 500 companies reviewed by Institutional Shareholder Services disclose any board-level AI oversight at all. - [**ISS and Glass Lewis**](https://aiireland.ie/2026/03/06/the-ai-literacy-gap-why-its-now-a-fiduciary-duty-for-every-board/) have both updated their stewardship guidance for the 2026 proxy season, putting withhold-vote recommendations on the table for directors who struggle to demonstrate documented AI literacy. That gap between adoption and oversight has already produced [**real litigation**](https://www.alston.com/en/insights/publications/2026/06/how-boards-can-shrink-the-ai-governance-gap). Shareholder suits have targeted companies accused of overstating AI capabilities or understating the costs and risks of AI initiatives, arguing the board lacked the internal controls to ensure public statements matched operational reality. A board member walking into a 2026 strategy review has read these headlines too, which is exactly why the seven questions below carry more weight than they did eighteen months ago. [Why AI on data fails — and how PromptQL fixes ItBy Rajoshi Ghosh, Co-founder, PromptQL![](https://storage.ghost.io/c/26/b3/26b323cb-c378-4831-bc7d-27e29def746a/content/images/icon/AIAI-ICON-4b029873-e640-488e-a8f7-2a82bccd0b68.png)AI Accelerator InstituteRajoshi Ghosh![](https://storage.ghost.io/c/26/b3/26b323cb-c378-4831-bc7d-27e29def746a/content/images/thumbnail/AIAI_Website_Article_Images_Author_Highlight--5--51532ce5-2c3d-4ff7-a959-08ffaaabeafa.png)](https://www.aiacceleratorinstitute.com/why-your-ai-on-data-projects-keep-failing-and-what-fixes-it-2/) ## 1\. What is our actual return so far, and against what baseline? A board member asking this wants a number, a comparison point, and a date. [**Defining reliability**](https://www.aiacceleratorinstitute.com/ais-new-rule-demonstrating-reliability/) as the measurable standard rather than a vibe is what separates a confident answer from a stammering one. If the honest answer is "too early to tell," say that directly and name the date you will have a real figure. Come with the baseline already defined, well ahead of improvisation in the room. If the team skipped agreeing on what "before AI" looked like, the "after" number lacks anything credible to compare against, and a sharp director will notice the gap immediately. --- ## 2\. Who owns this if it goes wrong? Directors have sat through enough incident postmortems to know that "the team" makes a weak substitute for an owner. They want a name, a title, and a chain of accountability that survives someone being on vacation when the system misbehaves. [**Recent governance research**](https://www.schellman.com/blog/ai-governance/who-should-own-ai-governance) found that in most organizations, the same executive who approves an AI purchase is also the person left holding the liability when it fails, a structure that concentrates risk rather than genuinely managing it. A board that asks this question twice, once about the decision and once about the consequences, is testing whether those two roles sit with one person or a genuinely accountable structure. [**The AI architect role**](https://www.aiacceleratorinstitute.com/the-emergence-of-the-ai-architect-engineering-the-future-of-tech/) is one attempt at formalizing that structure, rather than leaving it to whoever happened to sign the vendor contract. --- ## 3\. What data is it touching, and did we have the right to use it? This question has teeth now that regulators are watching. A board wants to hear that provenance, consent, and access controls were part of the design, rather than a patch applied after Legal asked about it. State-level AI regulation is proliferating quickly enough that a policy adequate in January can lag behind requirements by the following board cycle. The stronger answer names the specific framework guiding your data governance, whether that is the EU AI Act's risk-tiering approach or the NIST AI Risk Management Framework, rather than a general assurance that "we take this seriously." [**Governing shadow AI**](https://www.aiacceleratorinstitute.com/turn-shadow-ai-into-safe-agentic-workforce-barndoor-ai/) matters here too, since a tool that entered the organization outside procurement rarely has provenance documented anywhere at all. --- ## 4\. How do we know it is still working the way it did on day one? Model drift feels like a concrete threat to a director who read about a competitor's chatbot going sideways. [**Operational stability**](https://www.aiacceleratorinstitute.com/operational-stability-for-mission-critical-ml-systems/) built for mission-critical systems is the honest answer here, backed by monitoring dashboards a non-technical board member could actually read. [**Testing tool-selection reliability**](https://www.aiacceleratorinstitute.com/the-canary-test-ai-agents-keep-failing/) directly, rather than trusting a benchmark score from launch day, is how leaders catch drift before a director does. [8 stats that show AI got smarter faster than it got saferSWE-bench just crossed the 100% line and code security is still stuck at 56%. Eight numbers that show exactly where AI’s capability outran everyone’s ability to trust it, and what to do about the gap.![](https://storage.ghost.io/c/26/b3/26b323cb-c378-4831-bc7d-27e29def746a/content/images/icon/AIAI-ICON-b4d1cab3-683f-4b9c-9ab0-1a6946a98d48.png)AI Accelerator InstituteAndrew Lovell![](https://storage.ghost.io/c/26/b3/26b323cb-c378-4831-bc7d-27e29def746a/content/images/thumbnail/AIAI_Website_Article_Images_Doodles--3--4-394fdd65-18b0-4065-8688-7b9f31cdf0f1.png)](https://www.aiacceleratorinstitute.com/8-stats-that-show-ai-got-smarter-faster-than-it-got-safer-2/) ## 5\. What happens if we turn it off tomorrow? A surprising number of AI leaders have skipped running this exercise entirely. If the honest answer is "the workflow collapses," that is useful information for the board, and slightly alarming information for you. Test the kill switch before a director asks about it in the room. **A genuine test involves more than confirming a toggle exists:** - **Time the full revocation**, well beyond the moment the switch flips. - **Check for cached permissions** in downstream systems that might outlast the toggle itself. - **Confirm a human can step into the workflow** inside a single shift, rather than a multi-day ramp-up. Directors increasingly ask for the timing specifically, since "we can turn it off" and "we can turn it off in under an hour" are very different answers. --- ## 6\. How does this compare to what our competitors are doing? Boards read the same trade press everyone else does. They have seen the Wall Street headlines about agents running trading desks and coding pipelines. A grounded answer names where your organization sits on that curve, honestly, rather than overselling a pilot as a transformation. --- ## 7\. What is this costing us, really, once everything is counted? This is the question that catches people out most often. [**LLMOps enterprise costs**](https://www.aiacceleratorinstitute.com/llmops-optimizing-towards-enterprise-value-in-the-llm-agentic-era/) rarely get estimated accurately at the outset. Compute, licensing, integration, and the engineering hours spent babysitting a fragile pipeline rarely live in one spreadsheet. A board member who has read about budget overruns elsewhere in the industry will ask for the fully loaded number, well beyond the one from the original business case. --- ## The pattern across all seven Every one of these questions traces back to the same underlying demand: proof over promise. Boards spent the last two years hearing about potential. They are done with potential. They want evidence, ownership, and a plan for the moment something breaks. [**AI hallucinations**](https://www.aiacceleratorinstitute.com/ai-hallucinations-understanding-why-sometimes-machines-get-it-wrong/) matter here too, since a boardroom deck built on a confidently wrong AI-generated number is its own kind of unprepared. The AI Accelerator Institute's [**Chief AI Officers**](https://www.aiacceleratorinstitute.com/top-20-influential-chief-ai-officers-in-the-silicon-valley-area-2026/) community is built on exactly this shift: leaders comparing notes on what actually survives a board meeting, rather than guessing at what sounded good in a vendor pitch. The most useful preparation is boring on purpose: - **Write the answers down before the meeting**, in plain language a non-technical director can follow easily, glossary aside. - **Bring one number that stings a little**, alongside the plan to fix it. Boards trust leaders who volunteer the bad news before it surfaces on its own. - **Rehearse the kill-switch answer specifically**, since it is the one most leaders skip, and most directors now ask. [Bridging the gap from supercomputing to AI factoriesA comprehensive industry report on modernizing high-performance computing for production AI, featuring insights from NVIDIA and WEKA leaders.![](https://storage.ghost.io/c/26/b3/26b323cb-c378-4831-bc7d-27e29def746a/content/images/icon/AIAI-ICON-50c12b9f-fa27-4c52-8a8d-d52e5c1a3052.png)AI Accelerator InstituteAIAI![](https://storage.ghost.io/c/26/b3/26b323cb-c378-4831-bc7d-27e29def746a/content/images/thumbnail/AIAI_Supercomputing-to-AI-factories_Supporting-Assets_Meta-be134f63-a598-4d16-a366-7f5c024e7613.png)](https://www.aiacceleratorinstitute.com/bridging-the-gap-from-supercomputing-to-ai-factories/) ## Where AI leaders prepare for exactly this conversation The [**Chief AI Officer Summit Boston**](https://world.aiacceleratorinstitute.com/location/caioboston) brings roughly 250 director, VP, and C-level AI leaders together at the Westin Boston Seaport on October 29, 2026, built around governance conversations that hold up in front of a board. - **Governance frameworks** tested against real deployments, ready to adapt rather than build from a blank page under deadline pressure. - **Peer conversations** with leaders from around 175 companies who have already sat through this exact grilling. - **Direct access** to sessions on translating technical reality into board-level language that actually lands. Request a seat today: [**world.aiacceleratorinstitute.com/location/caioboston**](https://world.aiacceleratorinstitute.com/location/caioboston).