When Cisco joined Project Glasswing, we foresaw it as the most consequential shift in cybersecurity history. Six months in, I’ve never been more certain.
The numbers tell the story: For the last six months, over 50 specialized engineers have paired frontier AI with human insight to scan more than 6.7 billion lines of code spanning across our portfolio, including third-party assets. To turn that visibility into action, we prioritized hardening our portfolio over adding new features, with roughly 1500 engineers actively fixing bugs. The result is nearly 4000 fixes delivered in our 2026 security hardening releases. We’re moving away from the attacker’s clock and started on our own.
Sharing these figures is a first for us, so let’s be clear: these are fixes to software already in production, which is exactly why we are asking customers to move to our hardened releases. This isn’t a scorecard, a confession, or a finish line. It’s an honest measure of the vulnerabilities hidden within code and proof of what’s possible when you gain more concurrent and holistic visibility.
From flashlight to floodlight
Historically, software security was done with a flashlight: a skilled engineer would pick a component, aim the beam, and examine what was illuminated. It was careful, valuable work, but always limited by time and human resources. What fell outside the illuminated circle stayed shadowed as the flashlight moved.
Frontier AI helped us build automation that is our floodlight. We can now illuminate billions of lines of code across our portfolio at once. When you gain this visibility, the nature of the work changes entirely.
How we find changes. We no longer must prioritize what to inspect. We assess continuously using specialized agents for static analysis, live testing, configuration reviews, and exploit simulations with validation by our engineers. Discovery is no longer the bottleneck.
How we fix changes. The floodlight gives us visibility that helps us resolve an entire class of defects instead of individual instances. One fix does the work of many. Since the heavy lifting is now validation and shipping, this work has become a 1500-engineer and company-wide mission.
How we operate changes. With this new capability, security is now an automated rhythm: continuous assessment, fixed release cadences, and dedicated engineering for hardening. A one-time cleanup is a good start, but we’re committed to the harder, ongoing operational shift that this moment requires.
Every software company will eventually face this reality. Legacy security practices can’t keep up. This is a generational shift–not just a better tool, but a fundamental change in how we find, fix, and operate.
Changing how we build, not only how we check
The other half of this shift is how we build software. Jeetu Patel, our President and Chief Product Officer, has publicly stated that he expects almost 70 percent of our code to be AI-assisted by next year, which is our path to driving continued improvements in our codebase. For example, we recently shrank one security product from approximately eight million lines to under 1.5 million.
From a security seat, this is exciting. AI-written code has the opportunity to leave attackers with fewer places to hide, but it isn’t automatically safe. We’re enforcing a new discipline: automated testing to verify behavior and human accountability for new code. It’s the only way to operate: machine speed, human ownership.
What we owe our customers
Fixing faster only helps if you can absorb the workload. We know that managing hundreds of fixes is a heavy lift, and we are working hard to simplify our reporting and streamline the patching process. We’ve moved to a predictable hardening release schedule to help customers plan and implement fixes. We’re grouping updates so one upgrade does the work of many fixes. The best protection is a current, hardened release, and we’re committed to making that path as simple as possible.
There are no victory laps
With AI-enabled adversaries shrinking the window for defense, the 4000 fixes from our first wave of hardening releases are just a start, not a finish line. We are now scaling continuous adversarial assessments and leveraging AI-driven findings to mature our development lifecycle, reducing reliance on reactive patching tactics to more strategic implementations of security-by-design that’s informed by AI-driven findings, for better security in from day one. This next phase is about further evolving how we work to create more inherently resilient systems, developed with even more focus on security-by-design practices and designed to stay ahead of the evolving threat landscape.
We cannot do this alone. With the emergence of frontier AI cyber capabilities, this work is too important and too urgent for any one company to tackle. The entire ecosystem, maintainers, researchers, and partners, is straining under the same volume. We must scale together. We’ve said before that optimism and vigilance are the same job. Cisco has spent over 30 years being the company in the room when things go wrong on the internet, and we’re here for this moment too.
We aren’t claiming victory. We’re doing the work and reporting honestly what we find. The hard work is just beginning.
