Sharper: Countering Cyber Threats

Advanced AI and quantum capabilities are increasing the cyber threat to critical systems, making the development of proper resilience and responses a national security imperative. These new threats could come from adversaries but also from misaligned or uncontrolled AI systems, with the Hugging Face incident demonstrating the potential danger of capabilities outpacing constraints. CNAS experts continue to develop policy recommendations to ensure critical infrastructure, private data, and essential systems remain secure while preserving America’s innovation edge. Continue reading this edition of Sharper to learn more.

Features:

Report | Red Lines

The Chinese government treats frontier AI systems as an instrument of statecraft, creating a significant cyber threat to U.S. national security as Chinese AI models become increasingly capable. In Red Lines: Understanding the National Security Risks of China’s Advanced AI, Daniel Remler argues for an analysis of these threats outside of the framework of an “AI race,” emphasizing AI-enabled offensive cyber operations as a critical area of risk in the kinetic domain. “PLA cyber campaigns are where AI capabilities will likely have the most immediate effect on U.S. national security,” Remler writes, especially as other adversarial actors can take advantage of China’s open-weight releases to carry out further cyber threats.

Report | Tipping the Scales

After first allowing defenders to better scale cyberdefenses, emerging AI capabilities have the potential to disrupt this balance to disproportionately empower attackers. In Tipping the Scales: Emerging AI Capabilities and the Cyber Offense-Defense Balance, Caleb Withers examines these dynamics and envisions how the net advantage of AI-powered cyber tools might ultimately be determined, suggesting how policymakers can strengthen cyber defenses and proactively shape the landscape in favor of U.S. security.

Virtual Event | The AI Open-Weight Debate: Balancing Innovation and Security

China’s recent cyber campaigns have demonstrated the ability of AI to heighten adversarial operations in cyberspace, and the release of leading Chinese models as open-weight software opens the door for other actors to take advantage of these capabilities at the expense of U.S. interests. At a recent event on The AI Open-Weight Debate, CNAS Senior Fellow Daniel Remler was joined by Chris McGuire of the Council on Foreign Relations, Stella Biderman of EleutherAI, and Ian Reynolds of Hugging Face—the open-source AI platform at the center of OpenAI’s recent rogue model incident—to discuss this critical cyber risk.

Report | Cyber Crossroads in the Indo-Pacific

The cyber domain has long been a critical nexus of the U.S.-China rivalry, but recent years have seen China both escalate and expand its reach geographically. In Cyber Crossroads in the Indo-Pacific: Navigating Digital Potential and Cyber Peril, Vivek Chilukuri, Lisa Curtis, Janet Egan, Morgan Peirce, Elizabeth Whatcott and Nathaniel Schochet assess how Washington and its Indo-Pacific partners have approached cybersecurity and recommend mechanisms to strengthen resilience—an increasingly pressing pursuit as AI capabilities are now heightening China’s ability to inflict harm in cyberspace.

Report | The Entanglement Edge

In The Entanglement Edge: U.S. Strategic Priorities in Quantum Networking, Constanza M. Vidal Bustamante and Morgan Peirce explain that quantum networking technologies could both enhance and resist cyber attacks: They could link quantum computing modules into a powerful computer capable of breaking common forms of encryption, while other applications could help secure communications, such as via quantum key distribution (QKD). The authors emphasize that, despite heavy Chinese investment in QKD, its practical limitations have led the National Security Agency and allied cybersecurity agencies to conclude that post-quantum cryptography—classical encryption algorithms designed to resist quantum attacks—represents the best defense against quantum computers.

CNAS Insights:

Washington Can’t Afford to Ignore AI’s Warning Shot

The OpenAI–Hugging Face incident was the first publicly disclosed case of an AI agent taking a harmful action against its operator’s intent—a “warning shot” for AI alignment. With the speed of AI advancement, Ruby Scanlon and Janet Egan warn that Washington should take immediate steps to align frontier models with human intentions, outlining mechanisms for both policymakers and the AI industry to ensure that future incidents don’t end in catastrophe.

Governing Jailbreak Incidents

While AI developers have made significant progress in defending against jailbreak incidents, they remain fundamental to machine learning. To mitigate the risk of these jailbreaks, Ben Hayum argues that the government must develop an institutionalized process that monitors the jailbreak attack-defense balance, centralizes intake of new jailbreak incidents, assesses their severity, and calibrates the response to meet future cases with a measured reaction rather than a crisis.

The Case for Long-Term CISA 2015 Reauthorization

The Cybersecurity Information Sharing Act of 2015 is one of the United States’ most important cyber defense laws, but its sunset provision and repeated lapses have produced substantial uncertainty and weakened the country’s ability to defend against cyber threats. Carrie Cordero and Morgan Peirce write that lawmakers should permanently authorize or substantially extend the authorization cycle so the United States is better prepared for a new generation of threats in cyberspace.

Commentary:

Analysis from Janet Egan, Michelle Nie, Gary Corn, and Morgan Peirce.

In the News:

Insights from Daniel Remler, Caleb Withers, Morgan Peirce, and Lt Gen Jack Shanahan (Ret.), USAF.