Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Liav Caspi, CTO and Co-Founder, Legit Security
Enterprises are racing to ship AI-powered applications faster than ever without fully understanding the consequences. Development velocity has never been higher, but neither has the risk. AI coding tools compress timelines from months to days while creating new blind spots: dependencies teams didn’t choose, code patterns they didn’t review, and vulnerabilities they didn’t anticipate.
Then there’s the issue of user trust. In a recent Legit Security consumer survey, one-quarter of respondents said they would lose trust if they learned their favorite app uses AI-written code. The contract between software companies and their users is more fragile than most teams realize, and it only takes one high-profile breach to break it.
Next year, we’ll witness the inflection point where velocity, security, and trust collide. The industry can’t keep moving this fast without knowing what it’s shipping. The following three predictions outline how application security will fundamentally shift to address this reality.
Prediction 1: 2026 Will Be the Year of AI-Coded Breaches
The gap between development speed and security comprehension has never been wider. We found that nearly 90% of developers now use AI coding assistants, and 96% of organizations use GenAI-based solutions for building applications.
This creates a specific type of vulnerability: the AI-coded breach – a flaw in code that developers don’t really understand or have never fully reviewed. These aren’t traditional security flaws. They’re vulnerabilities introduced when AI coding tools generate insecure patterns that look legitimate, or create emergent behaviors that slip through human review. When a model suggests a library that is outdated and vulnerable or implements flawed authentication logic, those issues pass through because teams trust the AI’s output implicitly or lack the capacity to review.
This scenario also creates ambiguity around who should be held responsible for AI code failure. Is it the developer who accepted the AI’s suggestion? Is it the security team that should have placed guardrails? Or is it the organization that deployed the coding tool in the first place? The accountability issue will surface as we see the adoption of AI coding tools surge.
Visibility into AI-first SDLC is no longer optional. New innovative tools can monitor AI-generated code and agents, helping teams place guardrails before breaches force the solution.
Prediction 2: Visibility Will Eclipse Velocity as the New Competitive Edge
AI FOMO is driving reckless decisions. Companies are shipping AI-powered features without understanding how they work or what dependencies they’ve introduced. Most are still asking, “How fast can we build this?” when the actual question should be, “Do we actually know what we built?”
The practical visibility gaps are significant. Most organizations can’t answer basic questions around what dependencies the AI agent has introduced and where the code patterns came from. Without this visibility, teams are deploying applications they can’t fully explain or defend.
This is where visibility becomes a competitive advantage. Companies that can confidently explain how their software is built, what’s in it, where it came from, and how it behaves will earn customer trust. Those that can’t will lose users the moment something breaks.
Higher visibility doesn’t equate to slowing down. When teams know exactly what’s in their applications, they can move faster with confidence. They can deploy updates without second-guessing whether an AI assistant introduced a backdoor three sprints ago. And they can especially answer customer questions about data handling without scrambling through codebases.
Prediction 3: Agentic AI Security Will Become Its Own Discipline
A decade ago, DevOps became its own discipline because development velocity outpaced what traditional IT operations could handle. The old model broke, and the industry created something new to bridge the gap. AI security is following the same path.
Traditional AppSec was built for software written by humans, reviewed by humans, and deployed through predictable pipelines. That model doesn’t work when AI is generating code faster than security teams can review it, introducing dependencies they didn’t approve, and creating attack surfaces that didn’t exist six months ago.
Agentic AI security is distinct. The attack vectors are different – prompt injection, model poisoning, unexpected and harmful executions. The skills required are various, such as understanding how AI Agents behave, how they interact with other agents, tools and MCPs, and where context comes from and is stored. And ownership spans security teams, developers, data scientists, and infrastructure engineers. No single team owns the problem, which means everyone needs new capabilities.
Organizations that recognize AI security as its own discipline through dedicated roles, training, and tooling will move faster and more confidently than those trying to bolt it onto existing AppSec teams. 2026 is the year the industry chooses between reactive scrambling and proactive reinvention. Companies that treat AI security as a strategic discipline will define how the next decade of software gets built.
The infrastructure for AI-native development is already here. What’s missing is the security architecture to match it. Next year, that gap will close through either deliberate investment in visibility, discipline, and new capabilities, or through the costly lessons that come from deploying systems no one fully understands.
##
ABOUT THE AUTHOR
Liav leads product management and strategy. Liav previously held leadership positions in product management, platform architecture and engineering at IDF Unit 8200, Checkmarx, and Argus Cyber Security.
