Anne Neuberger Proposes AI-Driven Cybersecurity Plan for U.S. Critical Infrastructure

Author: Anne Neuberger, a16z

Compiled by: Deep潮 TechFlow

Deep潮导读: AI is drastically shortening the time window between the discovery and exploitation of software vulnerabilities, while a large number of operational technology systems running physical devices in U.S. critical infrastructure—such as power, banking payments, water supply, telecommunications, and hospitals—have been in service for over 20 years and simply cannot be patched at this speed.

Anne Neuberger, partner and head of global affairs at a16z, drawing on her experience as U.S. Deputy National Security Advisor, proposes a public-private partnership cybersecurity defense framework: the White House’s superintelligence unit sets national objectives, top CEOs across industries sign commitment letters, and existing industry information sharing and analysis centers are transformed to scale AI-driven defense.

The core logic of this plan is not to fully replace outdated equipment, but to first identify hundreds of critical operators and shared suppliers whose failure would trigger nationwide cascading effects, and contain the risk within six months through network isolation, 72-hour manual operation drills, and AI-driven compensatory protective measures.

a16z partner and head of global affairs Anne Neuberger wrote today in The Wall Street Journal:

AI is continuously shrinking the time between vulnerability discovery and exploitation. Defenders still responding at human speed will inevitably lose this race.

The same AI can also be used by defenders: identifying exposed systems, assessing which targets are most critical, and generating remediation solutions. For outdated devices that cannot be patched, AI can generate alternative mitigation measures, such as firewall rules or parameter adjustments. The question is whether we can deploy it quickly enough.

Fortunately, the most significant national-level risks are highly concentrated. Risks that could trigger nationwide cascading failures are primarily concentrated among hundreds of operators and their shared suppliers: large transmission substations, Fedwire and ACH payment systems, water infrastructure serving major cities, core telecommunications networks and 911 systems, regional trauma centers, and health records providers.

Small operators are equally important; they are often the weak points of defense, but prioritizing the protection of these key few can buy valuable time for repairing the rest of the system.

Drawing on her experience as former U.S. Deputy National Security Advisor for Cyber and Emerging Technologies, Anne outlines a new model for advancing national cyber resilience: reporting to the White House Super Intelligence Force and led by the CEOs of U.S. core enterprises responsible for critical infrastructure operations.

The United States must confront the significant risk that cyberattacks could disable critical infrastructure upon which millions of people rely. This cyber defense plan outlines the required governance structure and specific actions to identify and mitigate risk exposure from these hundreds of core operators and shared vendors, preventing their failures from triggering nationwide cascading effects.

This action plan aims to: (a) establish robust cybersecurity defenses across industries, while accounting for the large number of legacy and unmanaged systems; (b) leverage superintelligence (SI) technologies to identify and patch vulnerabilities in industry-specific, widely-used code; (c) build AI-guided or protective frameworks for vulnerabilities that enterprises have yet to address, enabling rapid identification of required patches when new models are deployed.

The second phase of this work will involve building digital replicas (i.e., "digital twins") of critical shared systems across various industries to enable regular testing of these shared systems.

This public-private initiative fully engages leading companies across industries and clarifies their responsibilities, as only they understand the most critical operational risks in their respective sectors; without their participation, it would be difficult to achieve rapid and widespread progress given the vast scale of U.S. infrastructure.

The White House Superintelligence Unit will establish a CEO committee responsible for overall effectiveness. Each industry’s Information Sharing and Analysis Center (ISAC) will be transformed into an AI-driven execution engine and deliver outcomes to the designated regulatory bodies for each industry.

Coverage: Five major industries, including power, banking and payments, water supply, communications, and healthcare hospitals.

Core objective: Hundreds of operators and shared suppliers whose failures could trigger nationwide cascading effects.

Technical core: Use AI to detect vulnerabilities at machine speed and generate tested alternative protection solutions, eliminate internet exposure of control systems, implement network segmentation, and verify 72-hour manual operation capability.

At the same time, clarify the nationwide scalable defense capabilities that telecommunications operators, infrastructure-as-a-service (IaaS) providers, and cloud network services (such as Cloudflare) can offer, along with the legal protections required to activate these national cybersecurity defenses.

Governance core: The CEO signs industry goals and reports progress monthly via the ISAC scorecard. Large operators assist smaller operators.

Potential legal amendments: Provide liability protection for information sharing and establish potential “safe harbor” mechanisms for participating companies pursuing outcome-based objectives.

AI is continuously reducing the time between vulnerability discovery and exploitation. Meanwhile, a large number of critical infrastructure systems cannot be patched within this timeframe. Operational technology (OT)—systems that control physical devices—often operate continuously for over 20 years, and patching some of these systems carries the risk of downtime.

Current regulatory frameworks were not designed to address this threat; they rely on fixed patch windows and compliance-oriented thinking, assessing whether controls are formally in place rather than the actual attack surface size or the speed of risk mitigation.

National governance model for rapidly scaling network defense

Reporting to the White House Superintelligence Unit, led by the CEO and operating through existing industry coordination committees (SCC), Information Sharing and Analysis Centers (ISAC), and regulatory agencies.

The United States currently lacks a governance model that enables targeted, rapid action to protect its most critical infrastructure. This initiative establishes such a model within a public-private partnership framework. In short, the White House Superintelligence Task Force sets national objectives, CEOs commit on behalf of their industries, and the SCC and ISACs drive implementation and measure outcomes within each sector.

The United States actually has a governance system in place to promote such efforts. Established two decades ago by executive order, this system mandates that each industry have a Strategic Sector Coordination Committee (SCC) at the strategic level and an Information Sharing and Analysis Center (ISAC) at the operational level, but this system urgently requires modernization.

SCC and ISAC are both private-sector-led nonprofit entities operating under federal directives and established public-private partnership frameworks. The most effective organizations typically feature a dual leadership structure of a CEO and a Chief Information Security Officer (CISO) to balance strategic decision-making across the industry with operational execution.

Currently, ISACs in the electricity and financial services industries have achieved CEO-level participation, but ISACs in the water supply, telecommunications, and hospital industries have not yet established this level of engagement.

The White House’s newly established Superintelligence Unit should fully leverage these ISACs to coordinate engagement with critical infrastructure industries. In industries where a CEO-level structure has not yet been established, the Superintelligence Unit must collaborate with the most influential corporate CEOs to drive the enhancement of relevant ISACs and the establishment of CEO-level oversight, thereby enabling necessary actions to be implemented.

In addition to the existing ISAC, this initiative will activate two additional forces: (a) cloud service providers, cybersecurity firms, and telecommunications operators, who will identify technically scalable defense measures under appropriate legal liability protections; and (b) AI labs and technology companies, which will collaborate with critical infrastructure operators to scan their code and firmware and generate remediation solutions and alternative mitigation strategies.

The AI laboratory should also collaborate with industrial control system manufacturers to build digital twins of standard equipment, enabling repair testing before actual connection to operational power grids or water plants; related findings will be consolidated into the government’s newly established AI-enabled vulnerability information exchange center, Gold Eagle, which should expand its scope from software to industrial equipment.

Device manufacturers (many of which are foreign companies) must be required to take action on identified vulnerabilities within a specified timeframe, as a prerequisite for selling devices to critical operators. Success metrics should be based on how quickly vulnerabilities are contained and whether core services can continue operating during an attack.

The chart below summarizes the participants and their roles in this national cyber defense initiative:

How does the accountability mechanism work?

Sign the commitment letter. Each CEO on the committee must sign on to industry-specific goals: achieving zero internet exposure for control systems, passing a 72-hour manual operation drill, and containing critical vulnerabilities within the specified timeframe (via patching or deploying compensating controls).

Monthly scorecard. Each ISAC submits an anonymous, industry-wide metrics report to the White House Superintelligence Unit. Every CEO can view their company’s position relative to the industry. If an industry falls behind its established targets, a CEO-level review with the Superintelligence Unit lead is triggered.

Signed off by the board. For publicly traded companies, the board certifies outcome-based progress metrics, aligning with current U.S. Securities and Exchange Commission (SEC) cybersecurity disclosure rules. Regulators will accept this certification in lieu of redundant audits, incentivizing companies to actively participate.

Large enterprises support small businesses. Large utility companies support cooperatives and municipal utility agencies within their service areas; large healthcare systems support rural hospitals; major banks provide financial support to community banks through the Financial Services Information Sharing and Analysis Center (FS-ISAC); and national telecommunications operators provide backup support to rural operators.

Cross-industry joint exercise. All industries are highly dependent on electricity and communications. Therefore, within 3 to 6 months, the CEO Committee will organize a joint exercise, modeled after the GridEx large-scale grid attack drill, to enable electricity, telecommunications, banking, and healthcare sectors to respond together to cyberattacks at AI speed.

Functional transformation of ISACs across industries

Currently, ISAC primarily shares threat alerts at the CISO level. Under this initiative, its role will be expanded to coordinate industry-wide efforts:

Shared AI tools: Provide detection rules, exposure surface scanning for member organizations, and AI-generated alternative mitigation solutions (which will ultimately be tested on digital twins that are exact replicas of the live system software to ensure operational reliability).

Coordinated remediation channel: Establish a unified contact channel for vendors to address AI-discovered vulnerabilities affecting the entire industry.

Scalable mutual assistance mechanism: Expanded upon the Cyber Mutual Assistance program in the power industry and the Sheltered Harbor data recovery standard for banking.

Metric summary: The ISAC generates a unified scorecard, so companies only need to report their data once.

Note: Without legal protection, the CEO will not authorize the sharing of sensitive data, and companies may also face legal liability due to vulnerabilities they identify themselves. Therefore, the Superintelligence Corps must establish clear legal safeguards for both scenarios: the former can be addressed by reauthorizing the Cybersecurity Information Sharing Act of 2015, and the latter can be achieved through some form of time-limited liability protection mechanism.

Strategic Technology Path: Contain Risk, Validate Resilience, Leverage AI for Scalable Adoption

The overall technical strategy prioritizes protecting the most critical assets, implements risk mitigation for systems that cannot be patched, and leverages AI to enable a small number of defenders to cover thousands of operators.

Classify targets by industry and identify key participants

Rank operators based on the severity of harm caused by failures and the number of other entities dependent on them, rather than solely on company size. Shared suppliers often carry greater systemic risk than single operators. Below are recommended approaches for the first five industries.

Electricity: large transmission substations, power balancing dispatch centers, and black-start power plants used to restart the grid after a major outage.

Banks and payments: Fedwire, ACH, CHIPS, clearinghouses, and widely shared cloud services and core banking system providers.

Water supply: Water supply systems serving large populations (50,000 or more U.S. residents), particularly their chemical dosing control systems.

Telecom: Internet routing, core mobile networks, 911 emergency call systems, and timing signals relied upon by various industries.

Hospitals: regional trauma centers, electronic medical record system providers, and medical equipment manufacturers.

Isolate systems that cannot be patched

Identify and close exposure surfaces: Use AI-assisted scanning to discover all internet-facing control systems, VPNs, and remote access paths. Disconnect each one, place them behind monitored access points, or protect them with phishing-resistant multi-factor authentication.

One-way communication connection: Deploy hardware devices that allow data to flow only outward from the most critical control network.

Reduce the blast radius: Isolate the IT network from the control network and segment it by facility or site to prevent a single point of compromise from spreading throughout the entire utility or healthcare system.

Verify manual operational capability: Each core objective must demonstrate during drills its ability to operate for 72 hours without IT systems. Hospitals use paper medical records, utilities rely on manual controls, and banks activate backup settlement channels.

Offline backup isolation: Maintain an offline copy of the control system configuration and verify the ability to rebuild the IT system from a bare-metal state, ensuring rapid system recovery.

Resilient emergency preparedness: Pre-secure regulatory exemptions and authorization documents to quickly mobilize critical resources in the event of a major operational disruption.

Scale defense with AI (implemented through ISAC)

Vulnerability Discovery and Remediation: Conduct AI-driven vulnerability mining on open-source code, vendor code, and firmware prior to release. Findings are reported to the National Vulnerability Exchange Center, requiring vendors to provide patches or alternative mitigations within a specified timeframe, with dedicated windows reserved for critical infrastructure to complete deployment and protection before vulnerability details become widely known.

Mitigate risk with alternatives: For systems that cannot be patched, AI converts each vulnerability into specific compensatory measures—such as firewall rules, monitoring alerts, or configuration changes—and tests them in a digital twin environment before deployment.

Block malicious traffic: CDNs, firewall providers, telecom operators, and network service providers can redirect and block malicious traffic before it reaches operational systems. Large-scale deployment of such measures may require government authorization.

Monitoring for operators unable to build teams: Subsidized AI security operations services for small water systems, rural hospitals, and electric cooperatives (nonprofit funding is currently available to support this).

Conduct drills against real-world attacks: Perform AI-driven red team exercises on digital twins across various industries to help defenders practice responding to tools operating at attacker speed.

Security safeguards: The AI defense tool is granted read-only access to the network by default; any changes require manual approval.

Industry-specific action guide

Each industry has a primary risk, a priority plan, an ISAC responsible for execution, and a shortlist of CEOs selected based on company size and industry influence.

Cross-industry members

All industries operate on major cloud providers (Microsoft, AWS, Google) and are increasingly reliant on cutting-edge AI labs. These institutions will join the Superintelligence Force CEO Committee as members, responsible for providing AI defense tools and identifying vulnerabilities in vendor products.

Policy levers, financial support, and proposed timeline

The White House Superintelligence Unit must address policy, legal liability, and regulatory issues to incentivize the necessary participation of all parties. The following key points outline the urgent institutional adjustments required.

Measure outcomes, not paperwork: Regulators track zero exposure to control systems, time to contain critical vulnerabilities, and verified human operational hours, rather than checking paper checklists.

Performance-based regulatory relief: Companies that achieve outcome-based goals and obtain board approval will be granted safe harbor legal protection.

Enforce supplier accountability: Incorporate “security by default” requirements into federal procurement; suppliers must disclose the support lifecycle of their products and must provide patches or alternative mitigations for vulnerabilities identified by AI within a specified timeframe.

Information sharing protection: Advocate for the long-term reauthorization of the Cybersecurity Information Sharing Act of 2015 (CISA 2015) to enable operators to rapidly share AI-generated threat data under legal protection.

Defense computing power support: The federal government coordinates support for participating enterprises and the National Vulnerability Information Exchange Center to ensure access to cutting-edge models and computing power.

Emergency Response Talent Reserve: Establish a vetted emergency response reserve team composed of personnel seconded from major operators and suppliers, to be deployed to the most critical infrastructure entities in the event of a security incident.

Schedule and Metrics

The primary goal should be to contain the risk within six months. While a full replacement of unsupported legacy equipment has been proposed, this approach is costly and could take 2 to 5 years, making immediate containment measures a priority.

The cost of addressing nationwide technical debt is expected to be substantial and must be prioritized for operators and shared vendors whose system failures would trigger nationwide cascading effects and who lack the resources to upgrade their systems independently.