AI Is Rewriting Cyber Risk. Is Your Testing Keeping Up?
Industrial Inspection & Analysis (IIA) is proud to announce a partnership with Kaiju Security to deliver penetration testing and offensive security services. IIA already helps manufacturers bring connected products to market with confidence, testing them against the cybersecurity requirements of the EU Radio Equipment Directive (RED) and the forthcoming Cyber Resilience Act (CRA), and is expanding that work to support the U.S. Cyber Trust Mark.
Compliance proves that a product meets the standard. Penetration testing shows how that product, and the networks, cloud services, and people behind it, hold up against a real attacker. Through this partnership, IIA clients get both from one trusted partner: the certification that opens markets in the US and EU, and the adversarial testing that finds exposure before an attacker does. That second need is growing fast, because AI is reshaping the threat landscape faster than most security programs can adapt.
In this post, we will look at what is driving that change, using the latest research from IBM, McKinsey, and Black Kite. We cover the four top trends reshaping cyber risk:
- record ransomware activity,
- AI as both the attacker's tool and its target, and
- a surge in new vulnerabilities.
In this post, we will look at what is driving that change, using the latest research from IBM, McKinsey, and Black Kite. We cover the four top trends reshaping cyber risk:
- record ransomware activity,
- AI as both the attacker's tool and its target, and
- a surge in new vulnerabilities.
In April 2026, a frontier AI model found thousands of high-severity vulnerabilities, including flaws in every major operating system and web browser. IBM calls it a warning to security teams. In attackers' hands, these tools collapse the time between discovering a flaw and exploiting it.
At the same time, AI went from pilot to production. 88% of organizations now use AI in at least one business function, up from 55% in 2023. Security has not kept pace. Most organizations deployed AI before defining the controls, testing, and response plans needed to run it safely.
The question is no longer whether a system can be compromised. It is what an attacker can do once they control an AI-enabled system, and how far the damage spreads.
AI adoption has outrun AI security
Organizations using AI in at least one business function, by McKinsey survey report
The Threat Landscape in Four Numbers
1. Ransomware is at a record high
Black Kite tracked 7,551 publicly named ransomware victims in the year ending March 2026. That is up 25% in a single year and nearly triple the count three years earlier. Manufacturing has been the most-targeted sector four years running.
Ransomware victims have nearly tripled in three years
Publicly named victims on ransomware leak sites, by Black Kite report year
The victims were not unlucky. 43.5% still carried critical unpatched vulnerabilities, and nearly a third had exposure to vulnerabilities already known to be exploited in the wild. These are exactly the gaps a penetration test finds first.
IBM sees the same trend. Ransomware hit 39% of breached organizations, up from 24% in 2023. Attackers have also changed how they extort. Threatening the victim's brand is now the most common tactic. 23% of attacks went after operational systems, through server encryption or OT disruption.
Ransomware has become an extortion business
What attackers threatened, share of ransomware attacks (more than one allowed)
2. AI is now the attacker's tool
In IBM's 2026 study, more than one in four organizations hit by a malicious attack said it was AI-driven, up 56% in a year. Deepfake impersonation drove 45% of them and AI-built malware 19%. Any approval process that trusts a familiar voice or email is now a target.
Those attacks cost more. AI-driven breaches averaged $6.04 million, against $5.03 million for malicious breaches without AI. Critical infrastructure absorbed 62% of AI-driven attacks. Industrial-sector breaches averaged $5.50 million, up from $5.00 million.
AI-driven attacks add about $1M per breach
Average total cost of a malicious data breach, USD
3. AI is now the target
21% of organizations reported a breach involving an AI model or application, up from 13%. Of those, 92% lacked proper AI access controls. Only 25% reported using adversarial testing or red teaming to govern AI.
The costliest incidents are the ones traditional testing never looks for. Model inversion averaged $6.07 million per breach, and prompt injection $5.89 million.
Attacks on AI systems carry a premium
Average breach cost by type of AI security incident, USD. Blue = above the $4.99M global average
4. Vulnerabilities are outrunning manual triage
A record 48,185 CVEs were published in 2025, up 21%. 2026 is on pace for roughly 70,000. Yet only 18% of security teams using AI agents point them at vulnerability management.
No team can patch by severity score alone. What matters is which weaknesses are present, reachable, and capable of real harm.
Published vulnerabilities are on pace to nearly triple since 2022
CVEs published per year
(forecast)
Testing Pays for Itself
IBM measured 30 factors that raise or lower breach costs. Offensive security testing cut the average breach cost by $211,339, one of the largest reductions in the study. It covers red teaming, penetration testing, and vulnerability testing.
Offensive security testing lowers breach costs
Average reduction from the $4.99M breach average, by factor, USD
Yet among breached organizations planning to raise security spending, the share putting it toward offensive testing fell from 32% to 26%. Attackers are speeding up while testing budgets slip.
The attack paths testing exposes are also the expensive ones. Voice and SMS phishing was the costliest entry point at $5.29 million per breach. Help-desk impersonation and MFA fatigue averaged $5.23 million. A compromised business partner added $227,250, more than any other factor.
Where AI Breaks Traditional Testing
Traditional penetration testing is still essential. But AI systems hold sensitive data, connect to other systems, and act with little human oversight. Testing has to cover the model, its data, its integrations, and the real-world consequences of compromise.
- Prompt injection. Malicious inputs that make an AI system ignore its instructions or safeguards.
- Model inversion and data exposure. Sensitive training or customer data extracted from a model or leaked through AI workflows.
- Data poisoning. Tampered training or retrieval data that quietly changes how a model behaves.
- Over-privileged agents. AI agents and machine identities with too much access. Fewer than half of organizations secure these identities in AI workflows.
- Weak integrations. Compromised apps, APIs, and plug-ins, and misconfigured cloud workloads. Each was behind 27% of AI-related breaches.
AI Expands Your Compliance Scope
AI does not replace existing obligations. It extends them. An AI system that touches regulated data, a cardholder-data environment, or a third-party service can pull new assets into SOC 2, PCI DSS, and ISO/IEC 27001 scope. ISO/IEC 42001 and the NIST AI Risk Management Framework are fast becoming what customers, regulators, and insurers expect.
The stakes are measurable. Noncompliance with regulations added $201,112 to the average breach. Security testing produces the evidence that your controls hold: access paths, data flows, segmentation, logging, and behavior under malicious input.
IIA + Kaiju Security: Testing Risk Where It Matters
Industrial Inspection & Analysis (IIA) brings deep testing, inspection, certification, and compliance experience across connected products and industrial environments. Kaiju Security brings threat-informed offensive security, testing how real adversaries find weaknesses, gain access, move laterally, and reach critical objectives.
Together, we find the vulnerabilities, prove the attack paths, and prioritize the fixes that matter.
| Service | What you get |
|---|---|
| Network & Application Penetration Testing | Establish a security baseline and find immediate risk across networks, applications, APIs, cloud, connected products, and AI integrations. |
| Red Team Operations | See how a determined adversary would operate in your environment, where controls fail, and what goes unnoticed. |
| Kaijin Ransomware Attack Simulation | Replay attack paths reverse-engineered from active ransomware campaigns to show where defenses break and what is exposed. |
| Social Engineering | Test your people against impersonation, deepfake-style pretexting, and credential theft before an attacker does. |
| Physical Security | Assess facility access, workstations, and on-site safeguards that protect systems and sensitive data. |
| Breach Data | Find your company's data on dark-web and open platforms, so you can assess impact, act, and communicate from a position of control. |
AI is changing cyber risk. You do not have to face it alone. Contact us about testing your environment the way a real attacker would.
