5 AI and Cybersecurity Questions Every Leader Is Asking Right Now: True Stories and Lessons Learned

From a $25 million deepfake to a chatbot that lost in court, here’s what I’m telling boardrooms, executive teams, and audiences around the world.

By Theresa Payton, CEO of Fortalice Solutions and former White House Chief Information Officer  |  Published 21 October, 2026

In January 2024, a finance employee in Hong Kong joined a video call with his company’s chief financial officer and several colleagues. They talked. He listened. He followed instructions. By the time it was over, he had made 15 transfers worth about $25.6 million.

Not one of the other people on that call was real.

I start with that story because it captures exactly where we are with AI. The technology is dazzling. The risks are real. And the best answers? They’re surprisingly human.

Over the last 18 months, I’ve advised our clients at Fortalice Solutions, given keynotes, sat in fireside chats, and briefed boardrooms and executive teams across the United States and in ten countries on six continents. The clients, the industries, and the audiences change. The questions don’t.

Here are the five questions I get asked nearly every time:

1.    “Could a deepfake of me or my CEO trick our team into wiring money?”

2.    “Should I let my customers’ AI agents use our website? How do I know it’s really my customer?”

3.    “If the AI gets something wrong, who’s responsible?”

4.    “Are employees sabotaging AI projects, and why? And is AI even cheaper than the way we do it now?”

5.    “With all the chips, data centers, and electricity AI needs, is it moral to use it? And what happens when it breaks?”

Below are the true stories that answer them, and the thing I think most people miss. Then I’ll take you around the world, because leaders in ten countries are wrestling with the very same questions, and some of their answers are worth borrowing.

The Quick Answers

•    Deepfakes: Yes, a deepfake can fool your team. The fix isn’t sharper eyes; it’s modifying your verification processes to do something the criminals don’t expect.

•    AI agents: Let them in on your terms. Verify the agent, confirm the customer’s permission, and keep the human in charge of high-risk actions.

•    Responsibility: If your AI gets it wrong, you own it. Give every AI system a named human owner.

•    ROI and “sabotage”: Measure today’s process before you judge AI. Most resistance is fear or frustration, and both are fixable.

•    Ethics and resilience: Ground your AI use in ethics, use it where the value justifies the cost, right-size the tool, and practice running without it.

STORY 1: THE CFO WHO WAS NEVER ON THE CALL

1. Could a deepfake of me or my CEO trick our team into wiring money?

Short answer: Yes. It has already happened, and the price tag can reach eight figures. But the best defense isn’t expensive technology. It’s a simple habit: verify before you act, through a channel the criminal doesn’t control.

Let’s go back to Hong Kong. The employee at global engineering firm Arup actually had doubts at first. The request arrived by message and talked about a confidential transaction. So he did what many careful people would do: he joined a video call to check. The faces looked right. The voices sounded right. His doubts melted away.

The criminals know our corporate processes, and they turned our processes into an attack. We must outthink and outsmart them.

Bloomberg broke a story with a different outcome on July 26, 2024. A Ferrari executive got WhatsApp messages that appeared to come from CEO Benedetto Vigna, hinting at a big confidential acquisition. Then came a phone call with a convincing AI-cloned voice, down to his Southern Italian accent. The executive noticed slight inconsistencies in the caller’s tone and asked about a book Vigna had recommended days earlier. The caller couldn’t answer, and hung up.

Two similar stories with different outcomes. The difference wasn’t a better tool. It was a better habit.

By the numbers

•    1 in 4 malicious breaches were AI-enabled in IBM’s 2026 Cost of a Data Breach Report, a 56% jump in a year, costing an average of $6 million each.

•    45% of AI-enabled attacks involved deepfake impersonation, the single largest category in IBM’s 2026 findings.

•    $3.05 billion in business email compromise losses were reported to the FBI in 2025, up about 10% from 2024.

What most people miss: The deepfake is problematic, but it isn’t the vulnerability. The process is. And criminals build these fakes from what’s already public: keynote videos, earnings calls, podcasts, even your family’s social media. Hong Kong police said the Arup attackers used publicly available footage. Your digital footprint is their raw material. That’s why your leaders’ personal online exposure is now a business risk, not just a personal one.

Your move this week: throw the criminals off track and go analog where possible

•    Pick a verification question or phrase for executives, finance teams, and your family. Make it something only the real person would know.

•    Call back on a known number, never one supplied in the request, before any payment change or urgent transfer.

•    Use a two-person rule on wires above a set amount, use a verbal passphrase on the call, and make it safe for anyone to slow down. Urgency plus secrecy is the oldest red flag in the book.

•    Shrink your leaders’ footprint. Find out what’s publicly available about them and their families, then reduce it.

STORY 2: YOUR NEXT CUSTOMER MIGHT NOT BE HUMAN

2. Should I let my customers’ AI agents use our website, and how do I know it’s really my customer?

Short answer: Increasingly, yes, because your customers are already sending them. Let AI agents in on your terms: confirm which agent it is, confirm the customer actually authorized it, and require the human for anything high-risk.

Picture a Tuesday night. Your customer is making dinner and tells her AI assistant, “Find me the best fare to Denver on Friday and book it with my usual card.” The assistant visits your website, compares options, and checks out while she’s stirring the pasta. She never sees your homepage.

That isn’t science fiction. Adobe data cited by Visa found AI-driven traffic to U.S. retail sites jumped 4,700% in a single year.

Now picture the evil twin: a fraud bot dressed up to look exactly like that helpful assistant, testing stolen cards or scraping your prices. To most websites today, the two look almost identical. That’s why Visa introduced its Trusted Agent Protocol in October 2025, a way for AI agents to carry a kind of digital ID badge that proves who they are and that a real customer gave them permission.

What most people miss: Many websites are blocking legitimate AI shoppers with bot defenses built for an earlier era, while the clever fraud bots slip through by acting human. And here’s the twist: AI agents can be tricked too. Hidden instructions planted on a web page or in a product review can steer an agent off course. So you’re not just protecting yourself from agents. You’re protecting your customer’s agent.

Your move: if you allow AI agents to act on your customers’ behalf, consider tiered access

•    Browse and compare: open to verified agents.

•    Purchase: allowed within limits, with your normal fraud checks.

•    Change a password, bank account, shipping address, or payout details: the human, every time.

•    Keep a separate log of agent activity, so you can show a customer exactly what their agent did.

“The question used to be ‘Human or bot?’ Now it’s ‘Whose bot, with what permission, doing what?’”

— [[TEAM NAME]], Fortalice Solutions

STORY 3: “THE CHATBOT DID IT” IS NOT A DEFENSE

3. If the AI gets something wrong, who is responsible?

Short answer: You are. Courts and regulators are treating AI as part of your organization, not a separate party you can blame.

In November 2022, a grieving customer went to Air Canada’s website to book a last-minute flight after his grandmother died. He asked the airline’s chatbot about bereavement fares. The chatbot told him he could book now and apply for the discount afterward.

That wasn’t the airline’s policy.

When he asked for the partial refund he’d been promised, Air Canada argued, in effect, that the chatbot was responsible for its own words. In February 2024, a British Columbia tribunal wasn’t having it. A chatbot, it reasoned, is still part of your website, and Air Canada hadn’t taken reasonable care to make sure its chatbot was accurate. The airline was ordered to pay.

The check was small, about $812 Canadian. The precedent was not.

By the numbers

•    92% of organizations that suffered an AI-related breach lacked proper access controls for their AI systems (IBM, 2026).

•    More than 20% of organizations reported a breach that targeted their AI models or applications directly (IBM, 2026).

What most people miss: The customer won partly because he kept screenshots. Your customers will keep a record of what your AI told them. Will you have the same record? And when everyone “owns” AI, no one does. Every AI system that touches customers, money, or safety needs one named human owner.

Four questions every board should be asking

1.    Who is the named owner of each AI system that touches customers, money, or safety?

2.    When it’s wrong, how do we find out, fix it, and make it right with the customer?

3.    Do our vendor contracts cover AI errors, data use, logging, and liability?

4.    Can we produce a record of what our AI said or did, and why?

“Accountability needs a name, not a committee.”

— Theresa Payton, CEO, Fortalice Solutions

“Treat every AI tool and agent like a high school intern or a new employee who just got the keys to the building: give it only the access it needs, log what it does, and review it often. That 92% is a fixable problem, and fixing it starts this week.”

— Blue Team / Security Engineering, Fortalice Solutions

STORY 4: THE QUIET REBELLION IN YOUR ORG CHART

4. Are employees sabotaging AI projects, and why? And is AI even cheaper than the way we do it now?

Short answer: Yes, some employees are resisting. Usually it’s fear or frustration with the tools, not malice, and both are fixable. And sometimes AI isn’t better or cheaper. You won’t know until you measure your current process first.

Here’s a scene playing out in a lot of companies right now. Leadership announces an exciting AI rollout. Training is scheduled. Dashboards are built. And down the hall, an employee closes the company’s official AI tool, opens a personal AI account he or she pays for personally, and pastes in a customer contract.

The employee is misguided, but not a villain. The employee is trying to get work done. But in doing so, they create two problems at once: the company’s AI investment isn’t being used, and its data just walked out the door.

By the numbers

•    31% of employees, including 41% of Gen Z, admit to undermining their company’s AI strategy, for example by refusing to use AI tools or outputs (Writer and Workplace Intelligence, 2025).

•    35% of employees pay for their own AI tools because their company doesn’t offer the ones they prefer (Writer, 2025).

•    75% of executives think their AI rollout has been successful, but only 45% of employees agree (Writer, 2025).

•    95% of organizations in MIT NANDA’s 2025 research saw no measurable return from generative AI, despite an estimated $30 to $40 billion in enterprise spending.

•    $670,000: the extra average breach cost at organizations with high levels of “shadow AI,” meaning unapproved tools (IBM, 2025).

What most people miss: Sabotage and shadow AI are two sides of the same coin. The fix for both is the same: involve your people early, give them tools they actually want to use, and be honest about what AI means for their jobs. And here’s a second miss: most organizations never measured what the current way costs, so AI “wins” or “loses” on gut feel. MIT also found the biggest returns in unglamorous back-office work, even though more than half of AI budgets go to sales and marketing.

Your move: a simple test before you scale

•    Measure your baseline: time, cost, error rate, and risk of the current process.

•    Run side by side for 30 to 60 days on real work, not a demo.

•    Count the hidden costs: human review, security, licensing, and the time spent fixing AI mistakes.

•    Ask your employees where AI would help them most. They know exactly where the pain is.

“Employees don’t resist AI. They resist uncertainty. When 75% of executives think the AI rollout is working and only 45% of employees agree, the gap isn’t technology. It’s trust.”

— Human Resources, Fortalice Solutions

“You can’t decide whether AI is better or cheaper until you know what the current way costs. MIT’s research found 95% of organizations saw no measurable return on generative AI, and in our experience, the missing piece is usually a baseline. Measure time, cost, error rate, and risk first, then let the numbers make the case.”

— Risk and Threat Assessment Team, Fortalice Solutions

STORY 5: THE DAY THE CLOUD WENT DARK

5. With all the chips, data centers, and electricity AI needs, is it moral to use it, and what happens when it breaks?

Short answer: It’s a fair question, and a leadership one. Use AI where the value clearly justifies the cost, choose the smallest tool that does the job well, and never build an operation that can’t function without it.

One year ago this week, just before midnight Pacific time on October 19, 2025, two automated processes inside Amazon Web Services tripped over each other. The result was a blank entry in the internet’s “phone book” for one of AWS’s core databases in Northern Virginia.

That tiny blank set off a chain reaction. Apps millions of people rely on, including Slack, Snapchat, and Atlassian, slowed or stopped for roughly 15 hours. No hacker. No attack. Just one hidden flaw in one region that so much of the internet depends on.

Now layer AI on top of that. More and more organizations are routing more and more of their work through a handful of AI models, running in a handful of cloud regions, drawing on power grids that are already stretched.

By the numbers

•    More than double: global data center electricity use is projected to more than double by 2030 to about 945 terawatt-hours, slightly more than all of Japan uses today (International Energy Agency).

•    Almost half of U.S. electricity demand growth between now and 2030 is projected to come from data centers (IEA).

What most people miss: Morality and resilience are the same conversation. When every task gets routed to the biggest model, in one cloud region, on one power grid, you are wasting resources and building a single point of failure at the same time. And here’s the hopeful part: the IEA also sees real potential for AI to help the energy sector itself cut costs and emissions. Used wisely, AI can be part of the solution.

Your move

•    Match the model to the task. Not every email summary needs the most powerful model on the market.

•    Ask vendors about energy efficiency and make it part of how you buy.

•    Run an “AI-off” drill: could you serve customers for 72 hours if your AI vendor or cloud region went dark?

•    Keep manual fallbacks, a second provider for critical functions, and copies of the data you’d need to keep operating.

Three More Questions I Hear in the Hallway

Is it safe to let our employees use AI tools?

Yes, if you give them approved tools, clear rules about what data can never go into them, and controls that enforce those rules. Banning AI doesn’t stop it. It just drives it underground.

We’re a city, county, or small business without a big budget. Where do we start?

Start with the basics that cost little or nothing: multifactor authentication everywhere, a callback rule for payment changes, an approved AI tools list, and a tabletop exercise that walks your team through a deepfake scenario. And if money goes out the door, report it fast. Speed matters.

“The Ferrari executive who stopped a deepfake had a better reflex: he asked a question only the real CEO could answer. Tabletop exercises build that reflex before the real call comes. And speed matters after a fraud, too. The FBI’s Recovery Asset Team reported freezing funds in 66% of the cases it worked in 2024.”

— Strategic and Crisis Communications Team, Fortalice Solutions

How much of my personal life is exposed online?

Usually more than you think. Home addresses, family members’ names, and old photos are often a few clicks away on data broker and people-search sites. The good news: your exposure can be measured, and it can be reduced.

“Every podcast clip, home address, and family photo online is raw material for a deepfake. In the $25.6 million Arup case, Hong Kong police said the attackers built their fakes from publicly available footage. Shrinking a leader’s digital footprint is essential. It’s a critical security strategy.”

— Digital Executive Protection Team, Fortalice Solutions

Ten Countries, Common Themes

Over the last 18 months, I’ve met with government and private sector leaders in ten countries, and in Jordan and Malaysia, with members of the royal family as well. I expected to hear ten different philosophies about AI.

Instead, I heard one theme across them all: AI must be human-centered.

.

Here’s my takeaway from each country, and what I think U.S. leaders can borrow.

Switzerland: Ethics as a bedrock. AI without ethical guardrails is like a Ferrari without brakes: fast, but dangerous. Guardrails aren’t what slow you down. They’re what let you go fast safely.

Italy: Cultural preservation meets innovation. Blend AI with your cultural roots, and double down on design and high-touch experiences, so you can innovate without losing who you are. AI should amplify what makes you distinctive, not flatten you into sameness.

Japan: Balancing tradition and innovation. Respect tradition, embrace innovation, and remember that AI must serve people, not replace them. Lead with augmentation, and adoption follows.

Jordan: A human-centric ethical framework. Jordan emphasizes transparency, fairness, and bias mitigation, positioning AI as an accelerator for society. You don’t have to be the biggest to lead. Setting clear principles early is its own kind of leadership.

Malaysia: Building the foundation for a regional AI future. Malaysia is fast becoming a technology and data center hub, and the IEA projects data centers could drive as much as one-fifth of the country’s electricity demand growth by 2030. The lesson: an AI strategy is also an energy, infrastructure, and security strategy.

Mexico: Productivity versus poverty. Upskill to turn fear into opportunity. Training is the bridge between AI’s promise and people’s worries, and it’s the very same fix for the quiet rebellion in Story 4.

Brazil: Scaling with privacy. Scale AI boldly, but lock down data like it’s Fort Knox. Privacy isn’t the brake on scale. It’s what makes scale possible.

South Africa: Empowering the future. AI can transform even the most remote communities, if you invest in people. That’s a lesson for rural America, too.

Australia: Practical capability and investment to scale AI safely. Build skills, invest to grow an AI industry, and use light-touch, principle-based governance. Let principles guide the rules, not the other way around.

New Zealand: Practical, human-centered adoption. New Zealand pairs public-service frameworks and targeted investment with principle-based governance, using AI responsibly to improve productivity and public services. It’s a strong model for U.S. city, county, and state governments.

The common thread

Big or small, every country I visited is asking the same question U.S. boardrooms are asking: how do we capture AI’s benefits without losing trust? None of them has it perfectly figured out. But the leaders furthest ahead share three habits. They put people first. They set principles before they write rules. And they treat security and privacy as the foundation for speed, not a brake on it.

The Bottom Line

Go back to those two stories at the top: the employee in Hong Kong and the executive at Ferrari. Both faced the same technology. One lost $25.6 million. The other asked about a book.

That’s why I’m hopeful. Every one of these five questions has a practical answer, and most of those answers start with people, relationships, and good habits, not expensive technology. The organizations that will win with AI aren’t the ones moving the fastest. They’re the ones that can move fast and still be trusted.

Blessings,

Theresa

Need help putting this into practice?

At Fortalice Solutions, we help organizations and their leaders use AI and technology with confidence, including AI risk and governance assessments, deepfake and social engineering tabletop exercises, incident response, and Digital Executive Protection to reduce leaders’ and families’ online exposure and take down imposter accounts. Contact the Fortalice team at https://www.fortalicesolutions.com.

Every story in this post comes back to one thing: the people who stopped the attack had someone to call. If you’d like Fortalice Solutions in your corner, reach us at 877.487.8160 or email us at Watchmen@Fortalicesolutions.com for a complimentary consultation. We’ll tell you honestly whether we’re the right fit.

Want these lessons learned, facts from the front lines, and relevant stories at your next event?

To book Theresa Payton for a keynote, fireside chat, or to moderate a discussion contact Kelly Skibbie at KPA Speaker Management: email info@kpaspeakermgt.com, call 703.217.0569, or visit https://www.kpaspeakermgt.com.

Frequently Asked Questions

Can a deepfake trick employees into wiring money?

Yes. In 2024, an Arup employee in Hong Kong transferred about $25.6 million after a video call in which every other participant was a deepfake. IBM’s 2026 research found deepfake impersonation was the largest category of AI-enabled attacks.

How can I tell if a video or voice call is a deepfake?

Don’t rely on spotting it; deepfakes are getting too good. Verify through a separate channel instead: call back on a known number, ask a question only the real person would know, and require a second approver for payments. That is how a Ferrari executive stopped a deepfake of the company’s CEO in 2024.

Should businesses allow AI agents on their websites?

Yes, with controls. Verify which agent is visiting, confirm the customer authorized it, allow low-risk actions like browsing, and require the human for high-risk actions like changing passwords or payment details.

Who is legally responsible when AI makes a mistake?

The organization that deploys it. In Moffatt v. Air Canada (2024), a Canadian tribunal held Air Canada responsible for incorrect information its chatbot gave a customer and rejected the argument that the chatbot was responsible for its own words.

Why do most AI projects fail to deliver a return?

MIT NANDA’s 2025 research found 95% of organizations saw no measurable return, mostly because of how AI was integrated into everyday work rather than the quality of the models. Measuring the current process first and focusing on back-office work improves the odds.

Are employees sabotaging AI projects?

Some are. A 2025 Writer survey found 31% of employees, including 41% of Gen Z, admit to undermining their company’s AI strategy, often out of job fears or frustration with the tools. Involving employees early and offering tools they want reduces resistance.

What is shadow AI?

Shadow AI is employees’ use of AI tools their organization hasn’t approved. IBM found that one in five organizations reported a breach tied to shadow AI in 2025.

How much electricity does AI use?

The International Energy Agency projects global data center electricity use will more than double by 2030 to about 945 terawatt-hours, slightly more than Japan uses today, with AI as the biggest driver.

How can organizations stay resilient when they rely on AI?

Avoid single points of failure. Keep manual fallbacks, use a second provider for critical functions, keep copies of essential data, and practice operating without AI for at least 72 hours.

Who is Theresa Payton?

Theresa Payton is the CEO of Fortalice Solutions, a Digital Trust and Protection firm, a Board Member, a former White House Chief Information Officer, and an author and keynote speaker on AI, cybersecurity, and digital trust.

Want to learn more about these topics? Follow us here!

AI cybersecurity questions

AI Security

Cybersecurity

Deepfakes

Deepfake Fraud

AI Agents

AI Governance

Shadow AI

AI Risk Management

Executive Protection

deepfake fraud

AI agents

AI liability

shadow AI

Sources

•    IBM, 2026 Cost of a Data Breach Report press release (July 29, 2026): https://newsroom.ibm.com/2026-07-29-ibm-study-one-in-four-malicious-breaches-are-ai-enabled,-costing-companies-6-million-on-average

•    IBM, 2025 Cost of a Data Breach Report press release (July 30, 2025): https://newsroom.ibm.com/2025-07-30-ibm-report-13-of-organizations-reported-breaches-of-ai-models-or-applications,-97-of-which-reported-lacking-proper-ai-access-controls

•    Baker Donelson, Ten Takeaways from IBM’s 2026 Cost of a Data Breach Report (source for the 45% deepfake figure): https://www.bakerdonelson.com/ten-takeaways-from-ibms-2026-cost-of-a-data-breach-report

•    FBI Internet Crime Complaint Center, 2025 IC3 Annual Report: https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf

•    FBI Internet Crime Complaint Center, 2024 IC3 Annual Report: https://www.ic3.gov/AnnualReport/Reports/2024_IC3Report.pdf

•    CNN, Arup revealed as $25 million deepfake scam victim (May 16, 2024): https://edition.cnn.com/2024/05/16/tech/arup-deepfake-scam-loss-hong-kong-intl-hnk

•    Bloomberg, Ferrari narrowly dodges deepfake scam (July 26, 2024): https://www.bloomberg.com/news/articles/2024-07-26/ferrari-narrowly-dodges-deepfake-scam-simulating-deal-hungry-ceo

•    MIT Sloan Management Review, How Ferrari Hit the Brakes on a Deepfake CEO: https://sloanreview.mit.edu/article/how-ferrari-hit-the-brakes-on-a-deepfake-ceo/

•    Visa, Trusted Agent Protocol press release (October 14, 2025): https://usa.visa.com/about-visa/newsroom/press-releases.releaseId.21716.html

•    Visa Developer Center, Trusted Agent Protocol: https://developer.visa.com/use-cases/trusted-agent-protocol

•    British Columbia Civil Resolution Tribunal, Moffatt v. Air Canada, 2024 BCCRT 149: https://decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do

•    Fox Business, Air Canada chatbot ruling and award amount: https://foxbusiness.com/lifestyle/air-canada-chatbot-gave-customer-wrong-info-airline-pay

•    Writer, 2025 AI Survey press release (March 18, 2025): https://www.businesswire.com/news/home/20250318607263/en

•    TechRepublic, coverage of Writer’s 2025 survey (source for the 75% vs. 45% figure): https://www.techrepublic.com/article/news-ai-adoption-gap-workplace/

•    Fortune via Yahoo Finance, MIT report on generative AI pilots (August 2025): https://finance.yahoo.com/news/mit-report-95-generative-ai-105412686.html

•    International Energy Agency, Energy and AI, Executive Summary: https://www.iea.org/reports/energy-and-ai/executive-summary

•    International Energy Agency, Energy and AI news release: https://www.iea.org/news/ai-is-set-to-drive-surging-electricity-demand-from-data-centres-while-offering-the-potential-to-transform-how-the-energy-sector-works

•    Power Technology, coverage of IEA Energy and AI (source for the Malaysia figure): https://power-technology.com/?p=382210

•    Amazon Web Services, Summary of the Amazon DynamoDB Service Disruption (October 2025): https://aws.amazon.com/message/101925

•    ThousandEyes, AWS Outage Analysis, October 20, 2025: https://www.thousandeyes.com/blog/aws-outage-analysis-october-20-2025