How Cisco Live Protect Uses eBPF Runtime Shields to Close the AI Exploit Window

Anthropic's disclosure of Claude Mythos Preview in April 2026 rewrote the rules of vulnerability management overnight. The unreleased frontier model autonomously discovered and exploited thousands of zero-day vulnerabilities across every major operating system and web browser, including a 27-year-old bug in OpenBSD and a 17-year-old remote code execution flaw in FreeBSD . It solved a 32-step corporate network attack simulation end-to-end, succeeded at 73% of expert-level capture-the-flag challenges, and chained multiple browser vulnerabilities into exploits that escaped both renderer and operating-system sandboxes .

Anthropic deemed the model too dangerous for general release and instead worked with select partners — including Cisco — to use it defensively . But the demonstration proved a point Cisco's security leadership had been dreading: frontier AI models can now find and exploit software flaws faster than any human team, compressing the traditional vulnerability-to-patch cycle from weeks to minutes .

At Cisco Live 2026, the networking giant responded with Live Protect, a runtime security platform that doesn't patch software — it shields running infrastructure from specific vulnerabilities in real time.

What Is Cisco Live Protect?

Cisco Live Protect is a runtime security capability embedded directly into NX-OS, the operating system powering Cisco Nexus switches. It allows administrators to deploy Cisco-validated compensating controls — called "shields" — against specific vulnerabilities on live, running systems without reboots, software upgrades, or maintenance windows .

Cisco is explicit about what Live Protect is not: it is not a patch and does not replace permanent software fixes. It is a temporary, targeted emergency control designed to mitigate risk immediately after a vulnerability is disclosed, while infrastructure teams test, schedule, and deploy the permanent remediation through normal change-control processes . Cisco describes it as a "digital immune system" for its products .

How eBPF and Tetragon Power the Shields

Under the hood, Live Protect runs on eBPF (extended Berkeley Packet Filter) and Tetragon, technology developed by Cisco's Isovalent team — the same team behind the open-source Cilium project . eBPF allows security policies to execute directly inside the kernel of the switch control plane, operating with full system context and minimal latency .

This kernel-level approach means Live Protect can observe and enforce behavior at the point of execution, rather than relying on external monitoring or delayed response workflows. It protects against privilege escalation vulnerabilities and network control-plane DDoS attacks by intercepting system calls, monitoring processes, and tracking file activity .

Operational Modes and Shield Lifecycle

Live Protect shields operate in three distinct modes, giving administrators control over how aggressively they respond to threats :

• Monitor mode: Detects and logs potential exploit activity without blocking it. This mode is useful for validation and testing before enforcement.
• Enforce mode: Actively blocks the targeted vulnerability at runtime, preventing exploitation of the live system.
• Disable mode: Turns off a specific shield, useful when a vulnerability is no longer a concern or the shield interferes with legitimate operations.

A critical design principle is the automatic retirement of shields. Once a permanent software patch is applied and the vulnerability is fully remediated, the corresponding runtime shield is automatically retired . This prevents compensating controls from becoming permanent infrastructure cruft — a common risk with temporary fixes in production environments.

Live Protect is now generally available on Cisco Nexus 9000 Series switches, with support for N9300 Series Smart Switches and plans to expand across the broader Cisco portfolio .

Why Claude Mythos Forced Cisco's Hand

The urgency behind Live Protect's launch is directly tied to Anthropic's Claude Mythos Preview. In early April 2026, Anthropic disclosed that its unreleased model had autonomously identified and written working exploits for thousands of zero-day vulnerabilities across every major operating system and web browser — including flaws that had survived decades of human code review and millions of automated security tests .

Cisco was one of 12 launch partners in Project Glasswing, Anthropic's restricted-access program for using Mythos defensively to find and patch software vulnerabilities before they could be weaponized . The exercise demonstrated that AI-accelerated vulnerability discovery had fundamentally changed the threat landscape.

The company's own statement was blunt: "Frontier AI models like Anthropic's Claude Mythos can now comprehensively identify software vulnerabilities faster than any human team. The speed of AI-enabled attacks has fundamentally broken the traditional patch cycle and collapsed the exploit window from weeks to minutes" .

Twice-Monthly Vulnerability Disclosures: A Policy Overhaul

Cisco also used Cisco Live 2026 to announce a fundamental shift in its vulnerability disclosure model. Starting in July 2026, the company moved from ad-hoc emergency patching to a scheduled, twice-monthly cadence :

• Disclosures will occur on the first and third Wednesdays of each month.
• Cisco will provide seven days of advance notification of which technologies will be covered in each release.
• The goal is to give customers predictability to plan maintenance windows, rather than reacting to emergency advisories on unpredictable timelines .

This change was directly driven by the acceleration of AI-enabled vulnerability discovery, which made the traditional reactive model unsustainable .

Cisco Cloud Control: Agentic AI for Infrastructure

Live Protect was not the only announcement. Cisco introduced Cisco Cloud Control, an agentic AI platform for operating and defending critical IT infrastructure . Built on the Cisco Data Fabric powered by Splunk, Cloud Control provides cross-domain telemetry across all Cisco products, platforms, services, and agents, using purpose-built AI models to improve operations, telemetry, and security posture . It also addresses post-quantum readiness by preparing networks and encryption for quantum-era threats .

Resilient Infrastructure Services: A Three-Phase Framework

Cisco also launched Resilient Infrastructure Services, a structured three-phased service framework designed specifically for the AI-accelerated vulnerability era . Available starting July 2026, the framework includes :

• Phase 1 — Exposure Assessment: Immediate discovery and risk mapping. Cisco maps the attack surface and threats to infrastructure, providing prioritized security and operational actions.
• Phase 2 — Structured Remediation: Applying compensating controls and building a path to full patch deployment while product teams develop permanent fixes.
• Phase 3 — Continuous Resilience: Ongoing hardening and adaptation as Cisco expands its capabilities and new threats emerge.

The phases are iterative rather than strictly sequential, forming a cycle that adapts as the threat landscape evolves .

The announcement reflects a broader shift in Cisco's security philosophy: from treating patching as a periodic maintenance activity to treating vulnerability defense as a continuous, always-on operational responsibility. Live Protect's embedded runtime shields, combined with the predictive disclosure cadence and structured resilience services, represent a coordinated response to a world where AI models can find and exploit vulnerabilities faster than organizations can patch them.

It's not a patch. It's a shield — and for network operators facing the reality of AI-powered zero-days, it's the breathing room they didn't have before.