Introduction: The AI Warning Businesses Cannot Ignore
AI Threat:In January 2024, a finance employee in the Hong Kong office of Arup, the British engineering firm behind the Sydney Opera House, received a message that seemed to come from the company’s UK chief financial officer. It asked for a confidential payment. The employee suspected phishing. That was the right instinct.
Then came a video call. The CFO and several colleagues were on screen, looking and sounding familiar. According to Hong Kong police, everyone on the call except the employee was a deepfake. The doubts faded. He made 15 transfers worth HK$200 million, about US$25.6 million, to five Hong Kong bank accounts. The fraud surfaced only when he checked with head office. (CNN)
Arup confirmed the case in May 2024 and said none of its internal systems were compromised. Nothing was hacked. A person was persuaded by a face and a voice.
That is why the warning that companies are ignoring AI threat deserves a fair hearing. If a video call can no longer prove who is speaking, what is your company relying on to stay safe?
Table of Contents
The Hidden AI Threat
AI has changed the speed, scale and believability of business risk. Convincing phishing emails can now be produced in minutes and tailored to one person. Voices can be cloned and faces faked.
Three different problems are often mixed together:
- Malicious use: criminals using AI for fraud, phishing and impersonation.
- Poor adoption: staff pasting confidential data into public tools, or trusting AI output without checking it.
- Competitive risk: rivals using AI to work faster while you stand still.
Each needs its own response.
Why Companies Are Ignoring AI Threat
Not every company is ignoring AI. In the World Economic Forum’s Global Cybersecurity Outlook 2026, the share of organisations with a process to assess AI security rose from 37% to 64%. Yet assessing a tool is not governing it, and roughly a third still had no process.
Four causes keep appearing:
- Complacency: “it has not happened to us”.
- Short term budgets: AI risk sends no invoice until something goes wrong.
- Skills gaps: nobody owns the topic.
- False confidence: “we have antivirus”.
Size changes the picture. Large firms have security teams, but staff adopt tools faster than policy. Small firms rarely have a security person. Startups favour speed over rules. Traditional organisations may delay AI, so staff use it privately.
The New Face of Cybercrime
Old warning signs, such as clumsy grammar, are fading. ENISA’s Threat Landscape 2025 analysed 4,875 incidents, mostly European, from July 2024 to June 2025. It reported that AI supported phishing made up more than 80% of observed social engineering by early 2025. ENISA says “reportedly”, so treat it as a strong signal, not an exact count.
The NSA, FBI and CISA warn that deepfakes can impersonate leaders and financial officers. Their advice: verify through another channel, protect executive communications, train staff and rehearse responses.
Arup shows why. The attackers did not need to break in. They needed one person to believe.
AI Is Also a Competitive Threat
OECD data show firms using AI across member countries rising from 8.7% in 2023 to 14.2% in 2024 and 20.2% in 2025, where data were available. Large firms stood at 52.0%, small firms at 17.4%.
Most firms are still early, so time remains. But the size gap is wide, and speed gaps compound: a rival that quotes faster gains ground quietly. Waiting also does not remove AI risk, because staff, customers and attackers already use it.
AI Adoption Is Not AI Readiness
| Area | AI unprepared company | AI ready company |
|---|---|---|
| Everyday AI use | Staff pick their own tools | Approved tools and clear rules |
| Security approach | Reacts after an incident | Reviews risks before launch |
| Threat mindset | Trusts voice, video and email | Verifies requests through a second channel |
| Ordinary vs AI enabled threats | Looks for typos and odd wording | Expects polished, personal, cloned messages |
| Adoption goal | Speed and cost savings | Speed with ownership, testing and training |
Adoption is buying and using tools. Readiness means someone is accountable when they fail.
The SMART AI Defence Framework
Built for this article, it suits a ten person firm or a large group.
S: Secure sensitive information and systems. Decide which data must never enter public AI tools. Use approved accounts, multi factor login and tight access controls.
M: Monitor AI use and emerging threats. Keep a live list of every AI tool in use, official or not. Review it every quarter.
A: Assess risks before deploying AI. Ask what data the tool touches, what happens if it is wrong and which laws apply. The free NIST AI Risk Management Framework gives a structure.
R: Retain human oversight and accountability. Name one owner per AI use. Keep people in charge of payments, hiring, legal and customer decisions.
T: Train employees and test preparedness. Run short deepfake and data handling sessions. Simulate fraud attempts and rehearse your response.
An Illustrative Example: The Urgent Voice Message
This is a hypothetical scenario, not a documented incident.
A company with about 300 staff receives a voice message, apparently from its finance director, who is travelling. The voice sounds right. It asks an accounts executive to pay a supplier urgently into a new bank account and to keep it quiet.
The warning signs are urgency, secrecy, a new account and a request outside normal channels.
Proper verification stops it. Policy says no payment to a new account without a call back on a known number, and two approvers for large sums. The executive calls the real director, who knows nothing about it. The loss is zero.
The lesson: do not depend on staff to spot perfect fakes. Build rules that work even when the fake is perfect.
What the Evidence Tells Us
- IBM, Cost of a Data Breach 2026: 602 breached organisations worldwide, March 2025 to February 2026. One in four malicious breaches was AI enabled, averaging $6 million against a $4.99 million global average for all breach types. Limit: only breached firms were studied, so this is not an overall attack rate.
- Same report: over 20% of organisations had a breach targeting an AI model or application, and 92% of those lacked proper AI access controls. Compromised APIs, plug ins and cloud misconfigurations were leading causes.
- IBM 2025: 20% of studied organisations had a breach tied to shadow AI, and high shadow AI added about $670,000 to average costs. Different year and sample.
- Stanford AI Index 2026: a record 362 reported AI incidents in 2025. Better reporting may explain part of the rise.
- WEF 2026: 87% of respondents named AI related vulnerabilities the fastest growing cyber risk (804 respondents, 92 countries). This is opinion, not measured loss.
Our reading: AI risk is often a familiar security failure on a new tool. Access control, inventories and verification are cheap compared with one successful fraud.
Advantages and Limitations
AI delivers real gains in efficiency, productivity, innovation and decision support. IBM found extensive AI and automation in security saved $1.93 million per breach on average. The limits are real too: bias, hallucinations (one Stanford benchmark showed rates from 22% to 94% across models), privacy exposure, implementation costs and over trust in automated answers. Job roles will change, so plan reskilling.
A Common Misconception
“We have antivirus, bought an AI security product and banned public AI tools, so we are safe.” Not quite. Tools do not fix weak access controls or untrained staff, and a ban may push use out of sight. Technology, people, process and governance must work together.
Who Should Act, and How
- Small businesses: approve two or three AI tools and verify every change of bank details.
- Large corporations: keep an AI inventory with named owners and regular audits.
- Startups: set data and access rules early, before habits form.
- HR and management: publish plain AI rules and train everyone.
- Finance: two approvers, known number call backs, waiting periods for new accounts.
- IT and security: secure APIs, plug ins and non human accounts, which fewer than half of organisations actively secured in IBM’s data.
The Future Belongs to Prepared Companies
Over three to five years, readiness should feel routine: a live AI inventory, verification as a habit, regular audits, AI literacy and board level reporting. Check local data protection and AI law. No plan removes risk. It reduces it and shortens recovery.
Conclusion: The Cost of Looking Away
The evidence does not say all companies are ignoring AI threat. It says many move slower than the risk. Start this week:
- List every AI tool your people use.
- Write one page on what data must never enter public AI tools.
- Require a second channel check for any payment or bank detail change.
- Name one owner for AI risk.
- Run a deepfake briefing and test it within 30 days.
Ignoring AI threats is not a strategy. Preparing intelligently is a business necessity.
Frequently Asked Questions
Why are companies underestimating AI threats? Complacency, tight budgets, skills gaps and false confidence in basic tools. AI risk often shows only after a loss.
How can AI be used in cyberattacks? To write convincing phishing, clone voices, fake video, build malware and scale attacks. ENISA and IBM both report rising use.
Can AI generated deepfakes cause financial losses? Yes. A deepfake video call led an Arup employee to transfer about US$25.6 million in 2024.
What is shadow AI in the workplace? Employees using AI tools without approval, often pasting in company data. IBM linked high shadow AI to higher breach costs.
How can small businesses protect themselves from AI threats? Approve a few tools, set a one page data rule, verify payment changes by call back and train staff.