The Security Playbook Is Dead – Here's What Replaces It

The shift. AI tools have collapsed the window between vulnerability disclosure and exploitation. According to Mandiant's M-Trends report, the average time-to-exploit is now negative 70 days: exploits land before patches. In 2018 that gap was 63 days. The trend accelerated sharply in 2023 and 2024, coinciding with the rise of frontier models that can chain multiple small vulnerabilities into reliable exploits.

Why it matters. This changes the economics of software security. The old model—scan for CVEs, prioritize by severity, remediate on a 30-day cycle—assumes patches exist before attacks. That assumption is dead. Attackers are using AI to write ransomware, pick targets, and manage campaigns from a single laptop. A 17-year-old used ChatGPT to steal 4.4 million records from a Japanese internet cafe. Semi-technical attackers hit 85 targets in months via FunkSec. The bar for effective attacks has dropped while the bar for defense has risen.

The supply chain trap. Every organization sits between its users and its upstream dependencies. Users trust you to deliver secure software, but you trust thousands of maintainers you can't vet. A two-hour attack on Axios achieved 600,000 downloads. The Shai Hulud worm in the JavaScript ecosystem spawned copycats. Malware now appears daily, not monthly. These attacks chain because everyone shares everyone else's dependencies. A single compromise propagates.

New attack surfaces. The easiest entry points are no longer production code. Attackers target CI/CD pipelines, developer laptops, and the emerging agent skill layer. Agent skills—like code that runs on AI assistants—have a 5% active malware rate and another 30-40% with serious security issues. Most organizations haven't hardened these pipelines. They run misconfigured, full of dependencies, with no signature verification.

What you should do. The talk recommends concrete tactics: implement cooldown periods for new package versions (days or weeks to catch malware), sign every artifact (Sigstore, SLSA), use minimal base images to shrink the attack surface, and prefer built-from-source repositories over raw upstream packages. For zero-day exploits, consider consortium-level intelligence sharing. The goal is to shift from reactive patching to build-time integrity.

The vendor lens. The speaker works for Chainguard, which sells these exact solutions: Wolfi minimal Linux images, Chainguard Libraries (built-from-source replacements for PyPI/NPM), hardened CI/CD actions, and agent skill hardening. The advice is sound and aligns with wider industry direction, but the optimal solution set may not be a single vendor. The principles—sign, minimize, isolate, cooldown—apply regardless. Teams should start with the free layers of Sigstore and SLSA before evaluating commercial options.