Attackers have accelerated, but most security programs have not. IBM’s 2026 Cost of a Data Breach Report shows what happens when those two realities collide.
According to the new report, the global average cost of a breach reached a record $4.99 million, driven largely by higher costs for detection, escalation, and business disruption. At the same time, AI-driven attacks increased 56%, adding an average of $1 million to the cost of a breach.
The good news, however, is that the report reveals what differentiates those companies that manage costs from those that incur them. Across all findings, a common theme emerges: reducing exposure before threat actors can exploit it yields the greatest financial benefit.
Key Findings
- The global average cost of a data breach reached $4.99 million, up 12% year-over-year.
- AI-driven attacks increased 56% and added an average of $1 million to every breach.
- When businesses extensively use AI and automation, they experience $1.93 million in cost savings compared to organizations not using them.
1. AI Has Changed the Economics of a Breach
IBM’s 2026 Cost of a Data Breach Report makes it clear that AI is not just changing how attacks happen, it’s changing how much they cost.
The report found that AI-driven attacks increased by 56% over the past year, with businesses reporting that these incidents cost an average of $1 million more per breach than those facing non-AI-driven attacks. AI allows attackers to automate reconnaissance, accelerate exploitation, and operate at a scale that manual defenses struggle to match.
IBM also found that shadow AI incidents have become far more common, affecting 43% of breached organizations, up from 20% a year earlier. Those organizations experienced higher average breach costs, greater operational disruption, and increased data loss.
AI has been deployed in so many ways that go well past copilots or development tools with central approval. Models, AI agents, AI-generated code, and third-party AI services are becoming integral parts of software deployment. Each of these elements increases the attack surface unless teams know where and how AI is used and find ways to control it consistently throughout the software development lifecycle.
IBM’s results align with findings from Cycode’s State of Product Security in the AI Era 2026. Although companies are actively incorporating AI into development processes, only 19% know precisely where AI code lives within their environments. Without clear sight, exposure skyrockets before security teams can limit it.
The financial impact IBM measures is the consequence of that visibility gap. Organizations cannot secure the AI assets they don’t know they have.
2. Attackers Are Targeting AI Itself
It’s no news that attackers have been using AI to write phishing emails and generate malware. However, increasingly, they’re going after the AI models and applications that businesses depend on.
IBM’s report found that 21% of breached organizations experienced a security incident involving their own AI models or applications, up from 13% last year. The most expensive incidents weren’t traditional infrastructure compromises. They were model inversion attacks, with an average breach cost of $6.07 million, and prompt injection attacks, with an average breach cost of $5.89 million.
IBM concluded that these incidents were rarely caused by weaknesses in the models themselves, but rather, traced back to compromised APIs, connected applications, plug-ins, and cloud misconfigurations.
Choosing the right model or evaluating its outputs is part of protecting AI, but now teams need to understand every component that supports these models, from the code that integrates them into an application to the identities, secrets, APIs, and third-party services they rely on.
Among those that experienced an AI-related breach, 92% lacked proper AI access controls, despite identity and access management being one of the most effective ways to reduce overall breach costs. Only 40% of organizations reported using access controls to protect AI models and data.
As AI becomes another layer of the application stack, it also becomes another part of the attack surface. Securing it takes the same visibility, governance, and continuous risk management that product security teams already apply across the software development lifecycle. That’s equally true for the agentic development lifecycle, where AI models, agents, and automation become active participants in how software is built and delivered.
Want to hear how CISOs are thinking about these challenges? Roland Cloutier explores the security implications of AI-driven software development with industry leaders on Cycode’s Shift to AI podcast.
3. Supply Chain Risk Has Become a Financial Problem
Software supply chain attacks have always been difficult to detect. IBM’s latest report shows they’re also among the most expensive.
Supply chain compromise ranked as the second most common initial attack vector and tied for the longest breach lifecycle, averaging 258 days to identify and contain. Of the 30 cost factors IBM’s 2026 report analyzed, a business partner or supply chain compromise added the largest increase in breach costs, averaging $227,250 above the global average.
Today’s applications rely on open-source libraries, third-party services, CI/CD processes, cloud technologies, AI algorithms, and ever-increasing sets of agents and plugins. With every additional dependency, there is another relationship based on trust, and another point of entry for threat actors.
Development teams are no longer consuming only packages and libraries. They’re also integrating AI coding assistants, foundation models, retrieval systems, agent frameworks, and AI-generated code into production applications. Product security teams now have two interconnected supply chains to secure: the traditional software supply chain and the emerging AI supply chain.
In both environments, the longer the attacker remains undetected, the higher the cost. Seeing dependencies, risk assessments, and knowledge of how software and AI components fit together throughout the development process are not just compliance processes anymore. They affect how quickly the company can detect, isolate, and recover from an attack.
That’s the case for treating CI/CD pipelines, code leakage, and AI-BOM visibility as one connected supply chain risk problem, not three separate checklists.
4. Reducing Exposure Early Still Delivers the Biggest Return
IBM evaluated 30 different factors that increased or reduced the cost of a breach. The biggest cost saver wasn’t a faster response after an incident. It was adopting a DevSecOps approach, which reduced average breach costs by $253,805. Identity and access management, offensive security testing, and security AI and automation also ranked among the strongest cost reducers.
Each vulnerability identified and fixed before software hits the market means there is one less chance for an attacker to exploit it, one fewer incident to analyze, and one fewer breach to deal with.
The report also shows there is still significant room for improvement. Organizations have embraced AI and automation for detection and investigation, but adoption drops when it comes to prevention. Only about one-third of organizations reported extensive use of AI and automation across the full security lifecycle, despite IBM finding that those that did reduced breach costs by an average of $1.93 million and shortened breach lifecycles by 65 days.
This is crucial, as AI technology has not only transformed the speed of attacks but also the speed at which software products are developed. It means that security teams need to perform continuous assessment of software components, dependencies, AI-generated artifacts, identities, and configuration. Delaying evaluation either until deployment or even later, when an attack happens, will leave defenders on the back foot.
Operationalizing that shift at the speed of AI is what tools like Cycode’s Agentic Workflows and its Context Intelligence Graph are built for: turning scattered signals into a prioritized, continuously updated view of exposure.
5. Security Teams Are Pointing AI at Yesterday’s Problem
The IBM report found that half of breached organizations have deployed AI agents in their security operations centers, but where those agents are being used tells an interesting story. Most deployments focus on threat hunting (56%), response (54%), and containment (53%). Only 18% use AI agents for vulnerability scanning and management, despite IBM identifying that area as one of the most important in responding to frontier AI threats.
That imbalance suggests many firms are still using AI to investigate problems after they’ve already entered the environment, rather than reducing exposure before attackers can exploit them.
IBM’s follow-up research offers an encouraging sign. After learning about the capabilities of frontier AI models to discover large numbers of high-severity vulnerabilities, organizations said they planned to increase the use of AI agents for vulnerability management and penetration testing.
Modern development creates far more code, dependencies, and components generated through AI than any manual process can reasonably hope to assess. It is the role of AI to identify exploitability, triage remediation needs, and minimize exposure before going into production.
TL;DR
IBM’s 2026 Cost of a Data Breach Report measures the financial impact of cyberattacks, and now also shows what happens when attackers operate at a speed that security programs can’t match. AI-driven attacks cost more, supply chain compromises take longer to contain, AI systems have become attractive targets, and organizations continue to invest more heavily in detection than prevention.
For product security teams, the findings reinforce a direction they have been heading towards for several years. Reducing exposure earlier, understanding software and AI dependencies, securing identities and secrets, and embedding security throughout the application and agentic development lifecycle have become measurable business advantages. IBM’s data shows those investments don’t just improve security. They reduce the financial impact when attacks happen.
Cycode’s Agentic Development Security Platform helps organizations reduce software, AI, and supply chain risks before attackers can exploit them. By bringing code, pipelines, AI assets, and third-party dependencies into a single platform, security teams can identify and prioritize the exposures that matter most across the entire development lifecycle.
Want to see what that looks like in practice? Book a demo and see how Cycode helps product security teams reduce risk at the speed modern software is built.
