Operationalizing U.S.-China AI Incident Notification
Following an agreement between the United States and China to establish a “notification mechanism” for AI incidents that could affect national security, this memo outlines a structure to operationalize the channel quickly. We explain why the two sides should establish a text-based standard template for sharing information on serious incidents, describe who should be notified, and provide examples of when the channel should be activated.
Cloud Compute: A Policy Escalation Framework
Chinese developers and hyperscalers are already renting frontier-scale U.S. AI compute from cloud service providers outside China. Washington needs a clear strategy to address national security risks posed by remote access to its most advanced compute. In our new report, we recommend that the Bureau of Industry and Security publishes a cloud compute rule and propose a policy escalation framework to solve this problem.
Honeypots for AI Agent Insider Threats
AI agent adoption at scale presents a new version of an old security problem—the insider threat. Existing approaches to monitoring and controlling AI agents provide important but incomplete defenses. Our new report describes how honeypots, decoys placed inside systems to catch unauthorized actions, could be a promising strategy to secure deployments at scale. To hone the strategy’s effectiveness, we identify five priority areas for deeper technical research.
Embedded Evaluators for Frontier AI: Contractual Foundations for Credible Assessments
Can embedded evaluators provide a real check on frontier AI companies? This depends on the contracts agreed by companies and the independent experts brought in-house to assess safety and security practices. For evaluators’ work to be credible and meaningful, and to identify serious risks before they cause harm, five contractual foundations should guide these arrangements.
Preparing for Offensive Cyber Agents: Piloting AI-Enabled Penetration Testing and Red Teaming
U.S. policy needs to prioritize hardening the cyber landscape, which is highly vulnerable to increasingly capable offensive cyber agents—AI systems that autonomously carry out attacks. AI-enabled pentesting and red teaming supported by federal pilots could address many of the challenges faced by critical but under-resourced defenders, such as public utilities and hospitals, as this report sets out.
Differential Automation: Steering Automated Research and Development Toward Safety and Security
AI companies are increasingly automating research and development (R&D) processes, advancing powerful AI systems and compressing timelines for the implementation of necessary safeguards. To manage the risks posed by fully automated AI R&D, or recursive self-improvement (RSI), our report sets out a differential automation strategy to ensure that an effective share of frontier-model-enabled R&D is directed toward security-critical efforts.
Priorities for Frontier AI Policy
The capabilities of advanced AI models are improving rapidly, raising the stakes for the development of safe, secure, and beneficial AI. To meet this challenge, a comprehensive set of policies is urgently required to govern the most powerful AI systems across their full lifecycle, advance American AI leadership, and strengthen national resilience against AI-enabled threats.
Establishing a U.S.-China AI Risk and Incident Dialogue
As the risk of acute AI incidents with destabilizing impacts increases, now is the time to develop communications channels between the U.S. and China before they are needed in a crisis. This report proposes an AI Risk and Incident Dialogue between the two countries, informed by the experiences of existing mechanisms like the Military Maritime Consultative Agreement.
Near-Term Verification Methods for AI Chip Exports
U.S. export controls on AI chips are being undermined by insufficient enforcement and a lack of resources. This report serves as an implementation guide for policymakers to strengthen enforcement through verification mechanisms that can be active within a year.
Evaluating LLM Capabilities for Commodity Classification
Automating commodity classification, a common task at the Bureau of Industry and Security, could speed up processing of export licenses and provide experience in building in-house AI tools. This brief sets out why a pilot program to test LLMs for this task in a more realistic environment should be created.
The OpenAI/Hugging Face Incident: Challenges in Controlling and Containing Cyber-Capable AI Systems
The OpenAI/Hugging Face incident is the first known instance of an AI system acting outside its developer’s intentions to autonomously identify a target and execute an attack end-to-end. Policymakers should consider it a warning shot that has left critical questions unanswered. In the incident’s wake, Congress should seek industry-wide responses to questions outlined in this memo, which also sets out actions that policymakers can take today to mitigate risks.
Remote Access Security Act (RASA)
U.S. export controls restrict selling advanced AI chips to China, but do not clearly reach the same compute when it is rented remotely through the cloud. The Remote Access Security Act (RASA) would close this gap in the Bureau of Industry and Security’s authorities. This brief assesses how the legislation should be scoped.
IAPS Reacts to the White House Executive Order on AI Innovation and Security
In response to the White House’s new Executive Order, “Promoting Advanced Artificial Innovation and Security,” our experts react to key provisions, outlining both implementation challenges ahead and opportunities to build on the administration’s approach.
Detecting Offensive Cyber Agents: A Detection-in-Depth Approach
AI agents can now orchestrate cyberattacks, dramatically altering the nature of cyber threats. To defend against these emerging threats, actors must first be able to detect them. This report outlines the AI challenge to traditional detection capabilities, introduces a framework to guide the response, and puts this into practice with actionable mechanisms.
After Mythos: A National Security Playbook for Frontier AI
As the White House weighs a range of executive actions to address the national security risks from AI, IAPS proposes strategic policy responses to match the scale and urgency of the challenge.
AI Distillation Attacks: Executive and Congressional Action Can Go Further
The White House and Congress have begun acting on AI distillation attacks, but gaps remain. This memo assesses the OSTP NSTM and the proposed Deterring American AI Model Theft Act of 2026 and recommends further steps.
Advancing America’s Cyber Strategy with Differential Access
Advances in AI-enabled cyber capabilities risk giving threat actors an advantage. To advantage defenders, differential access shapes access to cyber-capable models. The U.S. government should leverage these initiatives to advance the White House Cyber Strategy and U.S. national security.
Risk Reporting for Developers’ Internal AI Model Use
Frontier AI companies run their most capable models internally for weeks before public release. This report offers a harmonized reporting standard for internal use risks across SB 53, RAISE, and the EU Code of Practice.
Mythos and the Evolving Cyber Landscape: Implications and Policy Priorities
Anthropic's Claude Mythos is its most cyber-capable model yet. Policymakers should treat Mythos—and the accelerating trajectory of AI-enabled cyber capabilities—as a national security risk requiring urgent action.
AI Decision Support Systems: A Neglected Source of Military AI Risk
AI decision-support systems are now widely operational in real-world conflicts, yet fall outside DoD policy on lethal autonomy. This memo explores what AI-DSS are, potential failure modes, and policy recommendations to ensure security, reliability, and meaningful human oversight.