9 minutes reading
Cyber attacks and data breaches have held the number one spot in three consecutive Aon global risk surveys, and AI is the reason that ranking is unlikely to move. Generative AI did not invent a new category of attack. It removed the cost and skill barriers around social engineering, which was already the vector running through most breaches.
That distinction matters for how a security program responds. If AI created new threats, the answer would be new tooling. If AI made existing threats cheaper and more convincing at scale, the answer is to fix the layer those threats have always targeted, and to measure whether the fix works.
Why Cyber Attack Ranks as the Top Global Business Risk
Cyber attack and data breach ranks first because risk decision-makers expect it to stay first. Aon’s tenth Global Risk Management Survey, published in October 2025, surveyed 2,941 decision-makers across 63 countries and 16 industries and found cyber attack or data breach ranked as both the top current risk and the top future risk, a position it has now held across three survey cycles.
The rest of the table moved considerably. Business interruption sat at number two in 2025 but is forecast to fall to seventh by 2028. Geopolitical volatility climbed 12 places to enter the top ten for the first time at number nine, and is projected to reach fifth by 2028. Cyber is the constant in a list where almost everything else is in motion.
One boundary is worth stating plainly. This survey measures what senior risk owners believe, not how many incidents occurred. It is a useful signal of where budget and board attention will go, and it should be read alongside incident data rather than in place of it.
What AI Actually Changed About Cyber Risk
AI changed the economics of attacks that already worked, not the techniques themselves. Verizon’s 2026 Data Breach Investigations Report found that the median threat actor applies generative AI across roughly 15 documented techniques rather than inventing new ones. The attack chain is familiar. The speed, volume, and quality of the lure are not.
The World Economic Forum’s Global Cybersecurity Outlook 2026, built on survey responses from more than 800 leaders across 92 countries, reports that 94 percent of respondents view AI as the most significant driver of change in cybersecurity, with attackers using generative and agentic AI to scale social engineering, automate exploitation, and lower the barrier to sophisticated campaigns.
For a security program, that reframes the question. A campaign that once required a fluent writer, a researched pretext, and hours of preparation per target now requires a prompt. Defenses that depended on attacker effort as an implicit rate limiter no longer have one. Defenses that depend on a person recognizing a manipulation attempt still work, provided the person has practiced against the manipulation actually in use.
Why the Human Element Has Stayed at 62 Percent for Three Years
The human element appeared in 62 percent of breaches in the 2026 DBIR, up slightly from 60 percent the year before, and that figure has not materially moved across three consecutive editions. Verizon analyzed more than 22,000 confirmed breaches across 145 countries for the 2026 report, its largest dataset to date, and found social engineering to be the third most common breach pattern at 16 percent of all breaches.
A flat number over three years, during a period of record security spending, is the most useful finding in the report. It says the controls being added are not the controls that address this share of breaches. Most organizations can describe their email gateway, their EDR coverage, and their patch cadence. Far fewer can state, with a number, how susceptible their workforce is to a voice pretext or a cloned executive.
That gap is what human risk management exists to close: treating employee behavior as a measurable exposure with a baseline, an owner, and a trend line, rather than as a training completion percentage.
How AI-Enabled Attacks Changed Breach Economics
AI-enabled breaches cost roughly a million dollars more than the global average, which changes the business case for addressing them. IBM’s 2026 Cost of a Data Breach Report found that one in four malicious breaches was AI-enabled, a 56 percent increase over the prior year, and that those breaches averaged $6 million against a global average of $4.99 million. Deepfake impersonation accounted for the largest share of AI-driven attacks, ahead of AI-enabled malware and AI-generated phishing.
The vector ranking in the same report is where the human layer shows up in currency:
| Initial attack vector | Share of breaches | Average breach cost |
| Phishing, including voice and SMS | 17% | $5.29 million |
| Social engineering via help desk impersonation | 13% | $5.23 million |
| Global average, all vectors | Not applicable | $4.99 million |
| AI-enabled malicious breaches | 25% of malicious breaches | $6.00 million |
Phishing has led all initial access vectors for four consecutive years, and its voice and SMS variants now carry the highest average cost of any vector IBM tracks. The counterweight in the same dataset is that organizations running AI and automation across prevention, detection, investigation, and response closed breaches roughly two months faster and paid close to $2 million less than organizations running none.
The practical read is that deepfake phishing is no longer an emerging category to monitor. It is the single largest slice of AI-driven attacks, priced above the average breach.
Where Social Engineering Moved After Email
Social engineering moved to phone, SMS, and collaboration platforms while most awareness programs stayed on email. The 2026 DBIR found that 41 percent of social engineering breaches involved vectors other than email, with roughly a quarter of social action vectors coming from social media or phone-based channels. Large organizations saw a median of 48 SMS-based phishing campaigns per year against mobile devices, and smaller organizations a median of 12.
The simulation data is sharper still. Voice phishing succeeds around 40 percent more often than email phishing in the DBIR’s simulation figures, at a median click-equivalent rate of 2 percent versus 1.4 percent for email. Verizon’s analysts also noted difficulty finding organizations running voice-based simulations at all, which is the point: the success rate is higher partly because almost nobody has practiced against it.
This is a measurement gap before it is a training gap. An awareness program that reports a declining email click rate while AI vishing attacks go untested is reporting on a shrinking share of the actual attack surface. Closing it means running voice-based simulations with the same cadence and the same scoring discipline already applied to email.
People Security Management
Book a Free
Demo Call
with Our Expert
Discover how Threatcop protects your workforce from modern cyber threats.
Your Details
Why Shadow AI Creates an Unmapped Employee Attack Surface
Shadow AI creates exposure that security teams cannot see because employees adopt the tools faster than governance covers them. IBM’s 2026 breach cost research found shadow AI incidents at 43 percent of breached organizations, more than double the 20 percent recorded a year earlier, with an average incident cost of $5.39 million against $4.63 million the prior year. Sixty-eight percent of organizations lacked governance to manage AI use or detect shadow AI, up from 63 percent, and only 38 percent required IT approval before AI tools were deployed, down from 45 percent.
Governance is improving in one respect. The WEF outlook reports that the share of organizations assessing the security of AI tools before deployment nearly doubled from 37 percent in 2025 to 64 percent in 2026. The gap that remains sits between approved tooling and everyday behavior, where an employee pastes customer data into a consumer chatbot to save twenty minutes.
That behavior does not respond to a policy document. It responds to the same intervention set as any other risky habit: a baseline of who is doing it, a specific and non-punitive explanation of the consequence, and a sanctioned alternative that is easier than the workaround.
What Separates Cyber-Resilient Organizations From the Rest
Board engagement is the clearest structural difference between resilient organizations and the rest. In the WEF’s 2026 resilience data, 99 percent of respondents from highly resilient organizations reported board involvement in cybersecurity. Within that group, 52 percent said board members receive regular security updates, 48 percent reported boards actively engaged with the security function, and 45 percent said the board has a clearly defined oversight role.
The report also documents a widening cyber inequity gap, with large enterprises absorbing AI capability and defenses faster than smaller organizations, particularly in emerging markets. Resource constraints and skills shortages compound the difference rather than simply reflecting it.
For a security leader, the actionable part is not the correlation. It is the specificity. A board that receives a quarterly slide showing training completion at 96 percent is technically engaged and materially uninformed. A board that receives a susceptibility rate by vector, a median time to first report, and a trend against the prior quarter can ask the second question.
How to Put the Human Layer on the Risk Register
The human layer belongs on the risk register as measured lines with owners and cadences, not as a single entry called security awareness. Each line below needs a metric that a board can read and a security team can move.
| Risk line | Metric that proves it | How to baseline it | Review cadence |
| Email phishing susceptibility | Click rate and report rate, tracked together | Multi-template simulation across all business units | Monthly |
| Voice and SMS susceptibility | Engagement rate on vishing and smishing simulations | Voice and SMS simulation against high-exposure roles | Quarterly |
| Deepfake and pretext susceptibility | Compliance rate on a synthetic-media or help desk pretext scenario | Scenario exercise for finance, HR, IT support, and executive assistants | Twice yearly |
| Credential and MFA behavior | Reuse rate and MFA prompt approval without context | Credential exposure check plus policy audit | Quarterly |
| Shadow AI usage | Share of employees using unsanctioned AI tools with company data | Anonymous survey plus egress or SaaS discovery data | Quarterly |
| Reporting speed | Median time from delivery to first employee report | Simulation timestamps | Monthly |
| Third-party human risk | Share of critical vendors with a documented awareness program | Vendor questionnaire | Annually |
Two of these lines do most of the work. Reporting speed is the only metric on the list that improves outcomes during a live incident, because a report at four minutes gives a SOC options that a report at four hours does not, which is why incident reporting culture is worth measuring separately from click rate. And per-employee risk scoring is what turns the other lines into targeting, since a workforce-wide average hides the twenty people who account for most of the exposure. Guidance on building that scoring sits in this breakdown of employee cyber risk scores.
Set the baseline before running any training. A program that starts with training and measures afterward cannot attribute the change, and cannot tell a board whether the spend worked.
Where People Security Management Fits in an AI-Era Program
People Security Management is Threatcop’s framing for the layer this data keeps pointing at: employee vulnerabilities, behaviors, and decision-making treated as a security domain with its own controls and metrics, rather than as a training obligation.
It runs as a repeating cycle of four stages. Assess establishes the baseline through multi-vector simulation and behavioral analysis, which is where TSAT operates. Aware builds recognition of the specific risks that the baseline surfaced, delivered through TLMS. Empower turns the workforce into a reporting layer, which is TPIR’s job.
Three capabilities matter specifically for the 2026 threat picture of AI-scaled social engineering across multiple channels. AI-driven simulation templates keep scenarios current with the lures actually in circulation. Multi-language content delivery matters because a workforce spread across regions cannot be assessed fairly in one language. And WhatsApp and SMS threat reporting closes the gap identified in the DBIR vector data by giving employees a way to report an attack that never touched their inbox.
What to Do Before the Next Board Risk Review
Cyber risk will hold its position at the top of the register through this planning cycle, so the question worth answering is not whether the risk is real. It is whether the organization can state its exposure in numbers. Most can do that for infrastructure and cannot do it for people, which is the same gap the DBIR has been reporting at 62 percent for three years.
Start with a baseline across more than one vector. Run an assessment that covers email, voice, and SMS, score susceptibility per employee rather than per organization, and bring those three numbers to the next risk review alongside the infrastructure metrics that are already there. Book a Threatcop assessment to establish that baseline.
Frequently Asked Questions
How is AI changing cyber attacks?
AI is scaling existing attack techniques rather than creating new ones. The 2026 DBIR found the median threat actor using generative AI across roughly 15 documented techniques, mostly to produce more convincing lures at higher volume. IBM found deepfake impersonation to be the largest single category of AI-driven attack, ahead of AI-generated malware and phishing content.
What percentage of breaches involve the human element?
The 2026 Verizon DBIR puts the human element in 62 percent of breaches, a slight increase from 60 percent the previous year. That figure has stayed within a narrow band across three consecutive editions, which suggests current interventions are holding the number steady rather than reducing it.
Are AI-enabled breaches more expensive?
Yes. IBM's 2026 Cost of a Data Breach Report found AI-enabled breaches averaged $6 million against a global average of $4.99 million, a difference of roughly one million dollars per incident. One in four malicious breaches was AI-enabled, a 56 percent year-over-year increase.
Does security awareness training still work against AI-generated attacks?
It works where the training matches the vector. Email-only programs have limited effect on voice and SMS attacks, which the DBIR found succeed around 40 percent more often than email phishing. Programs that simulate voice, SMS, collaboration platform, and synthetic media scenarios, then score susceptibility per employee, address the attack surface that current data actually describes.
Praveen Pal Singh is the Growth Director – North India & ASEAN at Threatcop, with experience spanning cybersecurity, business growth, and People Security Management. He works with organizations to address human-layer risks and strengthen their cybersecurity resilience. His areas of expertise include cybersecurity awareness, social engineering, phishing, email security, human risk management, and People Security Management. He is passionate about helping organizations build stronger, people-centric defenses against evolving cyber threats.
