This episode cross-posts a panel from the event "AI Agent Containment Failures: Technical Realities and Policy Responses," co-hosted by the Wadhwani AI Center and the Institute for Law and AI on August 24. 

Check out the full event recording: https://www.csis.org/events/ai-agent-containment-failures-technical-realities-and-policy-responses

Guests:

Helen Toner, Executive Director, Center for Security and Emerging Technology (CSET) at Georgetown

Mackenzie Arnold, Director of U.S. Policy, Institute for Law and AI (LawAI)

Matt Pearl, Director, Strategic Technologies Program at the Center for Strategic and International Studies (CSIS) 

Timestamps:

What we've learned since the OpenAI-Hugging Face incident (1:15)

Incentives for safety measures at frontier labs (16:24)

Technical talent within government (22:18)

Ensuring compliance to incident reporting requirements (34:58)

Liability for crimes committed by AI agents (38:00)

Liability safe harbors (51:55)

Additional Reading:

"When Reporting an AI Security Incident Is Not Mandatory" by Mackenzie Arnold and Stephan Llerena for Lawfare: https://www.lawfaremedia.org/article/when-reporting-an-ai-security-incident-is-not-mandatory

"Investigating three real-world incidents in our cybersecurity evaluations" blog by Anthropic: https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals

"Incident Report: unsanctioned agent behavior during cyber testing" by U.K. AISI: https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing

"Pacing model development in an era of cyber-critical capabilities" announcement from OpenAI: https://openai.com/index/pacing-model-development-cyber-capabilities/

"Pacing the Frontier" petition: https://www.pacingthefrontier.com/

CSIS Commission on U.S. Cyber Force Generation: https://www.csis.org/analysis/csis-commission-us-cyber-force-generation

Limitations of the current incident reporting regime (7:46)Role of the U.S. government in helping defenders (11:52)What the executive branch can do now (25:29)Concrete policy recommendations (29:31)U.S.-China competition (40:50)Preventing abuse of an incident reporting regime (44:25)De facto regulatory role played by frontier lab employees (49:24)Role of AI-generated code in cyberdefense (54:45)Concluding remarks (55:53)"Out of Bounds: What the U.S. Government Should Do in Response to AI Agent Containment Failures" by Aalok Mehta: https://www.csis.org/analysis/out-bounds-what-us-government-should-do-response-ai-agent-containment-failures"The 'Breaking' News: The OpenAI–Hugging Face Incident" presentation at Black Hat: https://youtu.be/87DyyMV0kCY?si=GWh9MB1plhO-xjYT

This week, we cover updates from the ongoing cyber saga, including OpenAI's two-week pause on RL training and the cyber capabilities of Z.ai's latest model GLM-5.3. We also unpack Anthropic's move to add watermarks to text generated by Claude in compliance with the EU AI Act.

Timestamps:

Mark Zuckerberg's essay on AI (00:20)

OpenAI pauses RL training (5:56)

Cyber capabilities of GLM-5.3 (15:29)

EU AI Act refresher (24:29)

How Anthropic's text watermark works (31:21)

What's driving the backlash against Anthropic (40:59)

Additional Reading:

Zuckerberg's essay "The Future is for Everyone": https://www.meta.com/thefutureisforeveryone/ 

OpenAI announces two-week pause on RL training: https://openai.com/index/pacing-model-development-cyber-capabilities/ 

Letter from Sanders to tech CEOs: https://www.sanders.senate.gov/wp-content/uploads/AI-Pause-Letter-FINAL.pdf 

Letter from House reps to Mike Johnson: https://casar.house.gov/sites/evo-subsites/casar.house.gov/files/evo-media-document/final-letter-to-speaker-johnson-requesting-ai-hearings-1.pdf 

Z.ai blog post "GLM-5.3: Frontier Coding with Emergent Cyber Capabilities": https://z.ai/blog/glm-5.3 

Z.ai article "Preparing GLM-5.3 for Open Release: A Responsible Path to Cyber Defense": https://x.com/Zai_org/status/2088280509474320693 

"The EU's AI Transparency Code of Practice, Explained" (Tech Policy Press): https://www.techpolicy.press/the-eus-ai-transparency-code-of-practice-explained/ 

Anthropic blog post "How Claude's text watermark works": https://www.anthropic.com/news/claude-text-watermark 

"Toward a Federal Framework: Lessons from State and International Frontier AI Regulation" (CSIS): https://www.csis.org/analysis/toward-federal-framework-lessons-state-and-international-frontier-ai-regulation

Check out our upcoming event, "AI Agent Containment Failures: Technical Realities and Policy Responses": https://www.csis.org/events/ai-agent-containment-failures-technical-realities-and-policy-responses

In this episode, we're joined by Adrian Brown, founder and CEO of Windfall Trust, for a conversation about the economic impacts of AI.

Why the economic focus (00:58)

Capturing AI's impact in macro statistics (5:03)

Planning under uncertainty (9:44)

What's driving the AI backlash (17:27)

A new social contract for the AI era (20:31)

Are governments taking AI seriously enough? (31:21)

Why AI is likely to create major wealth concentration (39:10)

Adrian Brown is the founder and CEO of Windfall Trust, a policy accelerator working to prepare governments and societies for the economic impacts of AI. Before founding Windfall, he spent two decades at the intersection of government and technology, including as the founding Executive Director of the Centre for Public Impact in London.

Additional Reading

"Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence" (Stanford Digital Economy Lab): https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recen…

"AI as Normal Technology" (Knight First Amendment Institute): https://knightcolumbia.org/content/ai-as-normal-technology

"Forging a New AGI Social Contract" (Windfall Trust): https://windfalltrust.substack.com/p/forging-a-new-agi-social-contract

"Betting on Humans" by Anton Leicht and Dean Ball: https://writing.antonleicht.me/p/betting-on-humans

Join CSIS’s Aalok Mehta, Director of the Wadhwani AI Center, on a deep dive into the world of AI policy. Every two weeks, tune in for insightful discussions covering AI regulation, economic impacts, national security, geopolitics, and more. CSIS is a bipartisan think tank in Washington, D.C.

In this episode, we touch on Texas' new verification and audit requirement for data center developers seeking connection to the state's grid (1:11) before unpacking Anthropic's disclosure of three incidents in which its models hacked into another company during internal cyber evaluations (11:02). We discuss how these incidents—in addition to a previously disclosed incident at OpenAI—are driving calls for the U.S. government to help "pace the frontier" of AI development (21:53). We also explore the connection between AI safety and open model development, including Nvidia's open letter to U.S. policymakers, "Open Weights and American AI Leadership." (33:18).

Additional Reading:

Gov. Abbott's data center directive: https://gov.texas.gov/uploads/files/press/Thomas_Gleeson_Pablo_Vegas_Data_Centers_Directive_Letter_…

"China's AI Blitz Creates 'Death Zone' for Rival US Model Makers" (Bloomberg): https://www.bloomberg.com/news/articles/2026-08-04/china-s-ai-blitz-creates-death-zone-for-rival-us-model-makers

Anthropic's blog post "Investigating three real-world incidents in our cybersecurity evaluations": https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals 

Hugging Face's technical writeup "Anatomy of a Frontier Lab Agent Intrusion": https://huggingface.co/blog/agent-intrusion-technical-timeline

State Attorneys General letter to OpenAI: https://www.iowaattorneygeneral.gov/media/cms/08_5392C9E17791C.pdf

House Committee on Homeland Security briefing request: https://x.com/HomelandDems/status/2084389406194929918

"How OpenAI Lost Control of an AI Model—and What Needs to Change" (TIME): https://time.com/article/2026/07/24/openai-hugging-face-attack/

"Pacing the Frontier" petition: https://www.pacingthefrontier.com/

Nvidia's open letter "Open Weights and American AI Leadership": https://images.nvidia.com/pdf/Open-Weights-and-American-AI-Leadership.pdf

In this episode, we're joined by Fathom co-founder and President Bri Treece for a conversation about third-party solutions to AI governance. Bri makes the case for Independent Verification Organizations, or IVOs, as the ideal AI governance model (5:32) and explains what historical analogies have to teach us about regulating AI (10:39). We also discuss the role IVOs can play in assuaging the public's concerns about AI (32:27).

Bri is a co-founder and the President of Fathom. She previously served as the Chief Operations Officer of the Center for AI Safety (CAIS), a San Francisco-based nonprofit that works to reduce societal risks from AI through research and field-building. She also founded the Center for AI Safety Action Fund, an advocacy organization that connects technical AI experts with policymakers to advance U.S. leadership in AI.

In this episode, we're joined by Hiroki Habuka for a conversation about Japan's approach to AI sovereignty. We discuss how the Japanese government defines sovereign AI (1:08) and how that definition shifted following the Trump Administration's move to block foreign access to Anthropic's Fable in June (7:58). We also cover Japan's AI Promotion Act (13:52) and the country's take on copyright and AI in the context of "Ghiblification" (24:51).

Hiroki is a senior policy fellow in Japan's Cabinet Office on AI and a research professor at Kyoto University. In 2020, the World Economic Forum's Global Future Council on Agile Governance recognized him as one of the world's 50 most influential people revolutionizing government. Hiroki is a former non-resident senior associate of the Wadhwani AI Center.

In this episode, we discuss reports that OpenAI models escaped a sandboxed evaluation and breached Hugging Face's infrastructure (00:33). We then turn to Moonshot AI's Kimi K3, which has prompted talk of a second "DeepSeek moment," along with the distillation and export-control questions surrounding it (8:31). Finally, we cover New York Governor Kathy Hochul's executive order pausing data center permitting (26:57) and Google DeepMind CEO Demis Hassabis's proposed federal AI framework. (37:24).

In this episode, we're joined by Sean Mullin for a conversation on Canadian AI sovereignty. We discuss implications of the Trump Administration restricting foreign nationals' access to Fable (1:33); highlights from Canada's June 2026 National AI Strategy (13:05); the country's approach to AI regulation, including its AI and Data Act and (27:38); and more.

Sean is a senior fellow at the Munk School of Global Affairs & Public Policy at the University of Toronto. He previously served as economic policy advisor to Prime Minister Justin Trudeau.

You can read Sean's recent report on Canadian AI sovereignty strategy here.

In this episode, we discuss the AI Safety Measures Act, or SB 315, Illinois' new frontier AI bill that mandates third-party audits for frontier AI labs (:51). We also cover the Trump administration's decision to lift export controls on Anthropic's Fable and Mythos models (7:34), Anthropic's claim that Alibaba extracted capabilities from Claude in a large-scale distillation attack (17:46), and OpenAI's proposal for the government to take a 5% stake in the company (26:12).

Additional reading

IL mandates third-party audits of frontier labs

Commerce lifts export controls on AnthropicAnthropic accuses Alibaba of adversarial distillationOpenAI proposes government take 5% stake