Earning Public Trust Through Public Service and Verifiable Results
Frederick AI Lab (FAL) | Policy Proposal No. 1 | November 6,2026
1. A Time for AI to Serve Society
Cyberattacks can reach beyond information theft and computer outages to disrupt the physical systems that sustain everyday life.
Attacks on refrigeration systems in food retail can lead to food waste and supply disruptions. Unauthorized access to cold-storage systems for blood products and other medical materials threatens continuity of care. Disruptions to government services, financial systems, and payment networks directly affect communities and economic activity.
Whether AI contributed to any particular attack must be established through evidence. Equally, when AI is used for defense, its effectiveness and safety must be demonstrated through measurable results.
At a time when society faces these challenges, the AI industry should work together to protect people and demonstrate its usefulness and trustworthiness through action.
2. Establish a Public-Interest AI Cyber Defense Consortium
FAL proposes establishing a public-interest consortium bringing together AI developers, cloud and semiconductor companies, cybersecurity firms, equipment manufacturers, ethical hackers, and research institutions.
Participating organizations would make sustained contributions of funding, computing resources, technology, and expertise. Under neutral governance, the consortium would coordinate the development and testing of defensive technologies, threat information sharing, and recovery support.
Priority should go to sectors where disruption directly threatens life and essential services: government, finance, healthcare, and food storage and distribution. The consortium should also establish ways to reach smaller facilities with limited security resources.
3. Build a Shared Foundation: A “Linux for Cyber Defense”
The objective is an open defensive foundation that organizations develop together and that a broad range of specialist providers can use.
Common specifications should support threat information sharing, anomaly detection, the distribution of defensive measures, and response records. Products and AI systems from different vendors should be able to work together. Existing international standards and collaborative initiatives should provide a starting point, with interoperability testing and independent verification supporting wider international adoption.
Software and specifications suitable for public release should be available for specialist providers to use, inspect, and improve without licensing fees. Personal data, confidential facility information, and details of unpatched vulnerabilities must be protected and shared responsibly with appropriate parties.
Specialist firms would use this shared foundation to deliver deployment, maintenance, monitoring, and recovery services. A common foundation would allow their expertise and experience to contribute to the defense of a much wider community.
4. Keep Defensive AI Under Clear Control and Accountability
AI used for defense must operate within defined limits on its authority and actions.
Deployment must be authorized by the facility owner, with access restricted to what is necessary. Actions must be recorded, auditable, and traceable. Authorized people must be able to intervene and stop the system. Operations that could shut down or alter critical equipment should require approval procedures proportionate to their potential consequences.
AI monitoring should be supported by independent safeguards, including temperature alarms, local manual controls, and safety mechanisms that continue to function during communication failures. Facilities must retain the ability to protect people and equipment if the defensive system itself fails.
5. Measure Effectiveness and Examine Limitations and Failures
Deploying AI is not, by itself, evidence of success.
Evaluation should compare AI-assisted defense with existing methods, measuring attack detection, missed threats, false alarms, reductions in damage, recovery time, and operating costs. Failures and limitations should be examined alongside successes, with findings reported in a form that does not compromise security.
Assessment should include independent third-party verification rather than remain solely within participating companies. Expansion should proceed in stages, based on demonstrated results within clearly defined sectors and evaluation conditions.
6. Prioritize Public Benefit and Accept Responsibility for Community Costs
Initial projects should use existing computing infrastructure to establish effectiveness. Where additional AI data centers are necessary, applications that protect society should receive priority in infrastructure planning.
A public-interest mission must still meet requirements concerning electricity, water, the environment, and local living conditions. Developers should bear the costs of the additional power supply and supporting infrastructure their projects require, with safeguards against shifting those costs onto household electricity bills.
Public service and responsibility to host communities should advance together.
Earn Trust Through Demonstrated Protection
Public trust in AI cannot rest on promises of future potential alone.
Protecting food supplies, supporting healthcare, and keeping government services and payment systems functioning require practical cooperation, verified effectiveness, and accountability for results.
FAL therefore calls for a public-interest cyber defense infrastructure jointly sustained by the AI industry.
At a time when society needs help, the AI industry should bring its resources together to protect people—and demonstrate its usefulness and trustworthiness through verifiable results.