AI Breaks the Assumption Cybersecurity Was Built On

Key points

• Determinism was the quiet foundation of every security model built over the last thirty years; agentic AI dismantles that assumption at runtime.
• A Harvard Business School study found a 12.4% rise in coding activity after GitHub Copilot access, alongside a roughly 25% drop in project management time, where informal security review lives.
• Attackers are using AI to operationalize vulnerabilities that once required rare expertise, making "attackers probably won't bother" a poor planning assumption.
• Runtime visibility must become a primary control, not a fallback, because pre-deployment checks cannot contain an agent that adapts mid-task.
• Defensive AI is moving from experiment to production; the gap between attacker automation and defender automation is the real risk to watch.

Why does this break security now?

Determinism was the quiet foundation of every security model built over the last thirty years. Harden before deployment. Lock down permissions. Map trust boundaries. The system behaves consistently, so you can reason about where the perimeter is.

AI agents don't behave consistently. That's the whole point of them.

Large language models produce context-dependent outputs. Agents interact with external environments in ways their developers genuinely cannot predict ahead of deployment. When a system can adapt its behavior at runtime, the classic "keep bad things out" posture doesn't stop working entirely. It just stops being sufficient.

Cloud transformation gave us a preview. Security teams spent years chasing misconfigurations and identity sprawl, only to find that real failures happened in live environments: permissions that drifted, APIs that changed behavior, identities that accumulated unexpected access paths. AI compresses that staleness problem to near-zero latency.

Should you worry about the pace of software generation?

Software volume makes everything harder. When AI-assisted development tools push a team from hundreds of thousands of lines of code per month to millions, the window for security review shrinks. The Harvard Business School study on GitHub Copilot found a 12.4% increase in coding activity alongside a roughly 25% drop in project management time. That's where a lot of informal security review quietly lived.

On the offensive side, vulnerabilities that previously required significant expertise to chain together are becoming operationalizable at scale. Prioritization models built around "attackers probably won't bother" deserve a hard look.

What does runtime visibility actually require?

Runtime visibility has to become a primary control. If an AI agent can generate novel actions, access multiple systems, or adapt mid-task, pre-deployment controls alone cannot contain the blast radius of something going wrong. We've tracked this shift across seven runtime-security stories since 28 May, and the pattern is consistent: the teams ahead of this problem instrumented their agents before they needed to, not after.

Security teams need to know what data AI systems touch, which identities interact with sensitive environments, what actions agents are taking, and how fast an unintended behavior can spread before containment is possible.

Can defenders close the gap?

The defensive opportunity is real. Security teams cannot scale on human labor when software generation and vulnerability volume accelerate simultaneously. AI-assisted triage and automated investigation workflows are moving from experiment to production. Attackers are using automation to increase speed and scale; defenders need to match that or accept a permanent gap.

The organizations that adapt well won't be the largest or best-funded. They'll be the ones that stop treating runtime monitoring as an edge-case safety net and start treating it as the control plane for an environment that is, by design, always changing.