The Year the Vulnerability Backlog Changed Shape

As we enter the AI era, few among us have found themselves so entrenched in the throes of the shifting landscape as CISOs. With the threat landscape shifting, they are up against increasing rates of vulnerabilities and exploits alongside rapidly scaling demands for ever more secure environments. As CISOs, with our role as the protectors of an organization, we are expected to deliver guidance and security visibility. This is not a vision of the future, it’s today’s reality. Thankfully, we see some emerging solutions designed to address these problems, which we can unpack more deeply. 

For those of you who may not recognize the pain day-to-day, the Cloud Security Alliance reports that AI-assisted developers generate code commits three to four times faster than their peers but introduce security findings at ten times the rate, creating security debt that outpaces remediation efforts. Georgia Tech’s Vibe Security Radar tracked 35 CVEs in one month alone, directly linked to AI coding tools, with researchers estimating the true number in open source packages is five to ten times higher, with the latest AI Frontier models intensifying this trend. 

The UK’s NCSC warned of a “wave of patches” this spring, attributing it to years of prioritizing rapid delivery over vulnerability debt. AI-driven discovery is now exposing the consequences. Disclosed vulnerabilities surged 92% in 2025, even before this acceleration. The backlog is not only longer, but is growing faster than traditional release cycles can manage.

The Shift

The more significant shift is qualitative, disrupting established security models. For two decades, triage relied on severity: patch criticals, schedule highs, and let mediums and lows wait. This approach is now obsolete. The Australian Cyber Security Center‘s April guidance on frontier models advised organizations to reassess risk tolerance and patch vulnerabilities regardless of severity, especially when lower-rated flaws are combined into so-called chained vulnerabilities, resulting in an aggregated, more serious threat level.

These new AI agents excel at chaining vulnerabilities together. For example, a medium-severity information disclosure and a low-severity path traversal, previously deprioritized, can now be exploited together making the combined threat more severe than either of them on their own.

Compounding this issue is the reality that the response window has also narrowed. Unit 42 reports attackers scanning for new CVEs within fifteen minutes of publication, often before security teams review advisories. For example, a critical pre-auth RCE in the Marimo notebook project was exploited within ten hours of disclosure, using only the advisory text.

This shift has major implications for software producers. Accepting medium- and low-severity issues is not just an internal risk; these vulnerabilities ship to customers, get flagged by their scanners, and may trigger support tickets, security alerts, or contractual remediation obligations. 

This situation results in a severe operational mismatch: while customers technically possess the tools to filter findings by severity, their zero-tolerance policies mean they rarely do. Consequently, engineering teams must remediate the full spectrum of vulnerabilities across their products – with no increase in resources.

CISOs in the AI Era

For CISOs, the ramifications of this new reality are massive. Proofpoint’s 2026 Voice of the CISO found that 79% are expected to manage AI risk without additional resources. 

That idea is difficult to conceive, and that kind of approach to management reminds me of the experience of one very well-known CISO with whom I had the pleasure of speaking at this year’s swampUP conference. Tim Brown, who served as SolarWinds’ CISO during the massive SUNBURST malware attack that compromised the data of thousands of the company’s customers, exemplifies the resilience that CISOs must exhibit. That attack was caused by 3,000 lines of malicious code inserted into the build environment, bypassing source control. The fallout required massive efforts to rebuild trust, forcing the creation of a complete pipeline overhaul: three separate builds for development, staging, and production; access reduced from 400 to 36 to 2 people; binary comparisons across builds to detect tampering; and a six-month pause on new features. 

It also severely impacted Tim on a personal level. The SEC considered filing charges against not SolarWinds, but Tim as an individual. The stress caused him physical and mental deterioration, and he eventually wound up in the hospital, suffering from a heart attack. After 3 years, the case against him was dropped, but his insistence on fighting the charges created a lasting impact. It also signals something much broader about the role of CISOs in today’s changing AI-first environment. 

The SolarWinds attack was conducted before AI became a common tool and tactic used by malicious actors. With it in their arsenal, the damage they can do is much, much greater.

In the context of AI-assisted development, Tim’s experience points not only to fears about new attacks, their methods and scale, but also about ill-placed trust. Assuming code is safe, environments are secure, or systems are properly configured without verification led to the SolarWinds incident. 

Controls that once ran on a six-month audit cycle must now run continuously. In the agentic era, threats include not only rogue employees but also autonomous agents acting independently. This makes non-human identity management and agent sandboxing critical – especially with rising demands for regulatory compliance based on an exact record of software artifacts –  regardless of whether it’s agentic or man-made.

Facing the new application security reality

The most important advice considering all of this is to avoid a pessimistic outlook and stay positive. All the pressures described above point to a single essential capability: maintaining precise, continuous knowledge of your software artifacts and their deployment and ownership, and having a response plan in place if they are compromised. Without this, meeting regulatory requirements, like the 24-hour early warning needed in accordance with the EU Cyber Resilience Act and the 72-hour notification to ENISA for actively exploited vulnerabilities for EU market products, for example, will be impossible. Adopting a severity-agnostic patching approach or discussing AI-driven risk at the board level will continue to be goals yet to be realized. A unified plan is a single, strategic investment, and only the CISO can advocate for it, given their unique position across executive, engineering, and customer domains. 

But this responsibility is a shared one, and should not just fall on the sole shoulders of the CISO. When I asked Tim about his experience at SolarWinds and the lessons he took away from it, he emphasized that the roles of communication and collaboration can’t be understated.

JFrog’s Role in Building a Secure Software Supply Chain Future

On a personal note, this is a very challenging time for keeping applications secure. The strings noted above are all connected because governance, risk, compliance, security, and development are no longer separate discussions with different tools and owners. They have become a unified conversation, and organizations that have already recognized this and acted accordingly are managing the current challenges most effectively. The belief that we must be in the field with our customers, as I am privileged to be daily, in this challenging environment is a key reason I am proud to be at JFrog.

From a product perspective, we are not just building another scanner that reports issues after the fact. Instead, we offer a platform where the software supply chain is integrated: artifacts, provenance, governance, dependencies, policy, and release paths are all part of a centralized, trusted, single system of record.

This is designed to enable organizations to respond quickly to vulnerabilities and regulatory notifications efficiently, before they require crisis management. The rise of autonomous agents strengthens the need for robust controls. As agents increasingly write, review, and release code, the control point must be the platform itself, including:

• Managing non-human entities
• Ensuring verifiable provenance
• Enforcing policy at pipeline speed
• Maintaining an immutable build record

Tim’s statement about operating at machine speed without compromising compliance best defines today’s engineering challenge.

Importantly, this centralized approach is definitely not about slowing development to satisfy auditors. When implemented correctly, it lets teams deliver faster release cycles while maintaining standards they can confidently present to regulators, customers, and the board. For this, I am truly grateful for the opportunity to participate in these important discussions.

Adapting your Development Operations to Today’s Requirements

Each month, I engage with organizations across diverse industries, from technology and automotive to finance and beyond. The common challenge is clear: adapting software supply chains, originally designed for a slower, less complex environment, to meet today’s speed, scale, and risk while complying with new governance and regulatory requirements.

No organization has fully met this challenge. Supporting teams through this transition and learning from their experiences is the most rewarding work I have done. It is a privilege to be entrusted with this important knowledge.

Organizations face the challenge of balancing this unprecedented change with the need to continue supporting user and business demands for trust and innovation, but the good news is that most already have the technical know-how, operational expertise, and cultural elements to meet major security and regulatory demands.

Now it’s time for platforms, tools, and processes to step up to meet the challenge, and I see signs at JFrog and across our industry that this is already underway.

For more information on how JFrog can help manage and secure your software supply chain in today’s environment, reach out to me or schedule a meeting with one of our application security experts.

This blog content is provided for general informational and illustrative purposes. It reflects information available as of the publication date shown below and may not be updated. Examples and outcomes are illustrative; actual results may vary depending on your environment and use case. JFrog makes no representation or warranty that the content is complete, accurate, current or suitable for your particular circumstances. You are responsible for evaluating the content before relying on it. This blog does not modify any agreement governing your use of JFrog products or services. Product use remains subject to the applicable JFrog Subscription Agreement or other written agreement between you and JFrog.