The AI security risk gap: building best practice for AppSec
Want to bookmark your favourite articles and stories to read or reference later? Start your Independent Membership today.
Join today
Already a member?
Log in
Checkmarx is a Business Reporter client
AI is accelerating vulnerability discovery and exploitation, leaving organisations struggling to patch weaknesses before attackers can take advantage.
The recently reported security flaws in PHP have sent shockwaves through the global cyber-security community. It was revealed that software used across three-quarters of the world’s websites contained vulnerabilities that let attackers bypass login processes and remotely execute malicious code.
Perhaps the most shocking revelation is not that these vulnerabilities are being exploited, but that the flaws had been known for years. Until recently, exploiting them required highly sophisticated analysis, “at the edge of what human researchers could practically achieve”. The world still seemed safe, until AI delivered the analysis that criminals needed. And criminals are taking full advantage: nearly 900 vulnerabilities have been confirmed as exploited in 2026, which is already nearly double those for all of 2021.
AI and the security risk gap
As life gets ever more connected and digital services become more central to day-to-day activities, organisations must respond proactively to rapidly changing cyber-security requirements.
AI is adding powerful new defensive capability to current security best practice. But AI, with its advanced logic and analysis capabilities, is also layering new risks onto the digital world. The PHP vulnerabilities are just one example. For years, everyone ignored them until it became very clear that they represented a major risk. While AI helps security teams spot anomalies, it also helps attackers find and exploit software flaws, and do so fast.
A major problem is that so much software ships with known vulnerabilities: 91 per cent of companies knowingly release vulnerable applications. But today’s world makes it essential to address critical vulnerabilities that allow attackers to execute unauthorised code, gain control of systems or access sensitive data. As well as defining remediation processes, security policies also need to address the “attackability” of IT systems – their reachability and exploitability.
Security teams can struggle to deal with these developments. Partly because of the sheer volume of vulnerabilities. Partly because fixes are more complex and taking longer than ever: AI-generated code takes longer to fix as the developer isn’t familiar with its construction. As a result, the gap between time to remediate and time to exploit (which can be almost immediate) keeps widening.
This risk gap is a very real and expanding problem. It is defined by more code (AI-assisted developers produce code three times as fast), more risk in code (AI-generated code is frequently insecure), and faster attackers (over the past four years, average disclosure-to-exploit time has fallen from a month to less than a day).
“The vulnerability backlog has always existed. The challenge now is that it’s a whole lot bigger,” says Sandeep Johri, CEO of Checkmarx. “Before, you could ignore some of it and live with the risk. With LLMs being weaponised, you can’t afford that. You have to work through your backlog and keep it as small as possible.”
AI is very good at finding security flaws that were previously missed. AI can spot complex exploits, for example those that involve multiple steps or weaknesses in access control and business logic. But AI can also raise false alarms. And it can be hard to tell whether it is accurate: even the same model can provide very different analyses of the same code. A list of problems, some real and some not, some clear and some uncertain, isn’t a path to security. It’s just a bigger workload.
Nearly secure isn’t secure
Given how AI can support information security , there is a natural tendency for some cyber-professionals, under-resourced and under pressure to move quickly, to respond with optimism: “We did our best, and the important threats were covered.”
But in security, being “nearly protected” still means being exposed. One missed flaw can mean the difference between removing a problem and an attacker getting there first. The speed and capability of AI-powered defences may create an illusion of security. The danger is that a security tool that finds most of the risks can leave an organisation feeling safe when it isn’t.
Rather than simply trusting AI, security teams must demand proof. Faster scanning isn’t enough: better, more accurate scanning with fewer false positives matters, too. The focus must shift from simply finding vulnerabilities to making rapid, effective decisions about how to deal with them.
Ensuring a comprehensive scope of activity is also critical. Most developers have access to security tools, but they are not always used. This can happen for several reasons: a focus on speed and work pressure is often the cause; poor training may also be to blame; and some tools generate so many false positives that developers quickly start to distrust them. As a result, preventable flaws can enter the development funnel. That gap matters because a flaw left unfixed in the IDE moves down the pipeline and costs more to fix at every stage after.
Identifying flaws becomes harder when AI generates code rapidly and at scale: collating vulnerabilities in spreadsheets and tracking the backlog by hand no longer works. Teams need a centralised approach for governance and metrics that they can audit. Unfortunately, only around 20 per cent of companies have formal AI governance policies. And without governance, businesses can’t manage the risks of shadow AI use.
In addition, checking software that in-house developers have created isn’t sufficient when so much open-source software is knitted into systems. In addition, legacy platforms (such as PHP), vendor SaaS products and cloud-native apps must be examined. Are fixes verified from end to end, including build, deploy and runtime behaviour? And then there is remediation: how far does the mitigation process stretch? Into software, customer-facing systems, the supply chain…
On the surface, the barriers to effective cyber-security have fallen, and every vendor and every in-house team now claims to secure code with AI. And the future of AppSec will clearly involve AI.
But you can’t take security on faith, and AI alone isn’t the answer. While frontier AI adds coverage, speed and intelligence, businesses will always need an independent, evidence-based foundation that delivers consistent, repeatable, verifiable results. The hybrid human-machine model is almost always going to be more effective. Holistic security needs the hardcoded logic of traditional deterministic analysis provided by skilled people, while AI introduces probabilistic reasoning, context awareness and advanced analysis.
Ultimately, security isn’t about what you believe is protected. It’s about what you can prove.
To learn more about the trends, threats and innovations shaping tomorrow’s AppSec landscape, read Checkmarx’s The Future of Application Security in the Era of AI.