The AI Policy Podcast

This week, Aalok is joined by Chris Gorin, associate fellow at the Wadhwani AI Center and the newest co-host on the AI Policy Podcast, for a recap of Tuesday’s midterm elections. They discuss how AI influenced voters at the polls and the potential for AI regulation in the new Congress. Aalok then sits down with Sella Nevo to discuss threats to frontier AI labs, securing model weights, and his work on SL5.



Guests:

Sella Nevo, Director, RAND Center on AI, Security, and Technology



0:00 Intro

1:49 Chris intro

4:06 Midterms roundup

9:46 How data center measures could impact AI development

11:55 Exit polls and the potential for new AI regulation in Congress

18:23 Google restricts search for schools that disable AI

22:37 Sella intro

25:19 Why securing model weights matters

31:44 Threat vectors and real-world attacks

36:19 How frontier lab security has evolved over time

38:49 AI for cyber defense vs. offense

45:22 Security Levels 1–5 (SL1–SL5) explained

51:46 Where frontier labs stand today and obstacles to SL5

58:50 AI agents, containment breaches and sovereign AI

1:03:41 Security risks of embedded evaluators

1:06:26 Sella's predictions and biggest worries

1:10:30 Policy prescriptions

Additional Reading:

NBC News 2026 Exit Polls: https://www.nbcnews.com/politics/2026-elections/exit-polls

Saudi National Arrested For Installing Surreptitious Audio Recording Device At OpenAI In San Francisco (U.S. Attorney’s Office): https://www.justice.gov/usao-ndca/pr/saudi-national-arrested-installing-surreptitious-audio-recording-device-openai-san

How an AI start-up founded by ex-DOGE staffers broke into Anthropic (The Washington Post): https://www.washingtonpost.com/technology/2026/10/26/researchers-cybersecurity-start-up-cathedral-got-inside-anthropic/

“Securing AI Model Weights” by Sella Nevo, Dan Lahav, Ajay Karpur, Yogev Bar-On, Henry Alexander Bradley, Jeff Alstott (RAND): https://www.rand.org/pubs/research_reports/RRA2849-1.html

Aalok and Nicole discuss the latest AI cyber incidents, including major ARTEX-driven hacks and Qwen3.8 attempting to send local system data to a server in China. On the bright side, security experts used AI to find and patch a vulnerability in PHP impacting 70% of the internet. Plus, what's interesting in the Kimi K3.1 Pro model card and why Anthropic is getting backlash for its draft safety bill.



Timestamps:



1:12 ARTEX-enabled hacks

9:50 Qwen3.8 attempts to send data to China server

17:56 Calif finds vulnerability in PHP with GPT-5.6 Sol

25:32 Kimi K3.1 Pro model launch

31:06 Anthropic pitches draft safety bill



Additional Reading:



Microsoft blog post on ARTEX-enabled attack: https://www.microsoft.com/en-us/security/blog/2026/10/21/storm-0940-uses-artex-agentic-ai-connectwise-supply-chain-compromise/



Cohere report on misaligned Qwen3.8: https://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model



Calif report on PHP2Shell: https://calif.io/php2shell/



Kimi K3.1 Pro model card: https://www.kimi.ai/blog/kimi-k3-1-pro



Anthropic's statement on improved distillation defenses: https://www.anthropic.com/news/distillation-defenses



Anthropic’s draft AI safety bill: https://drive.google.com/file/d/1lJLyXL_2T6yc4rlKPqi6niMpfGucZ-oR/view



Anthropic’s IPO prospectus: https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm

This episode cross-posts the event "From Shahed to Geran: How Russia Continues to Reinvent the One-Way Attack Drone" hosted by the Wadhwani AI Center at CSIS on September 29.

Check out the full event recording: https://www.csis.org/events/shahed-geran-how-russia-continues-reinvent-one-way-attack-drone



Guests:

Kateryna Bondar, Senior Fellow, Wadhwani AI Center at the Center for Strategic and International Studies (CSIS)

Nicole Errera, Research Assistant, Wadhwani AI Center at CSIS

Samuel Bendett, Senior Associate (Non-resident), Europe, Russia, and Eurasia Program at CSIS

Additional Reading:

"From Shahed to Geran: How Russia Continues to Reinvent the One-Way Attack Drone" by Kateryna Bondar and Nicole Errera: https://www.csis.org/analysis/shahed-geran-how-russia-continues-reinvent-one-way-attack-drone

In this episode, Aalok and Nicole unpack OpenAI's latest math proof drop before diving into the White House's first policy actions in the "SI" era. They cover the (voluntary?) Accord on Super Intelligence signed by all the major industry players and discuss what listeners can expect from Trump's new Super Intelligence Force.



*This episode was recorded at 2:00 pm ET on Tuesday, October 13, 2026.*



Timestamps:

00:39 Model release updates + OpenAI/Anthropic price war

7:48 New math results from OpenAI

21:42 Tech leaders sign Accord on Super Intelligence

34:02 Trump's new Super Intelligence task force

43:40 Nonprofit sues OpenAI over Hugging Face hack



Additional Reading:

"Introducing GPT-6 Sol and Luna" by OpenAI: https://openai.com/index/introducing-gpt-6-sol-and-luna/

"Introducing Claude Opus 5.5" by Anthropic: https://www.anthropic.com/claude-opus-5-5

"GPT-6.1 Sol" by OpenAI: https://www.anthropic.com/claude-opus-5-5

"Claude Haiku 5.5" by Anthropic: https://www.anthropic.com/claude-haiku-5-5

"Introducing Muse Wonder and Muse Wonder Full: Frontier Models for Agents" by Meta: https://about.fb.com/news/2026/10/introducing-muse-wonder-models/

"Mark Zuckerberg says Meta has reached AI 'frontier'" by Financial Times: https://www.ft.com/content/fca23e77-be57-4d7e-ae61-74ff299e25cb

"Gemini 4 Argon: Built to solve your hardest problems" by Google: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/

OpenAI math results GitHub repository: https://github.com/openai/math/tree/main

"100+ reactions to 100+ solutions" on Proofs and Prompts: https://proofsandprompts.com/2026/10/08/100-reactions-to-100-solutions/

White House Executive Order on Super Intelligence: https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/

White House Accord on Super Intelligence: https://www.presidency.ucsb.edu/documents/white-house-accord-super-intelligence

Letter by three fired OpenAI employees: https://mikitabalesni.com/letter/letter.pdf

Legal Advocates for Safe Science and Technology vs. OpenAI: https://lasst.org/wp-content/uploads/2026/09/LASST-v.-OpenAI-Complaint-09.29.2026-AS-FILED.pdf

California Comprehensive Computer Data Access and Fraud Act: https://www.calpers.ca.gov/documents/ca-penal-code-502/download

California Assembly Bill No. 316: https://calmatters.digitaldemocracy.org/bills/ca_202520260ab316

"AI models may be capable of self-replicating without access to a datacenter" by Anthropic: https://www.anthropic.com/research/self-replicating-ai

AI data centers have become a flashpoint for policymakers across the country in the leadup to midterm elections, with polls showing that seven in 10 Americans oppose new data center construction in their local area. In this episode of the AI Policy Podcast, Wadhwani AI Center director Aalok Mehta sits down with writer Jasmine Sun for a conversation about the AI data center backlash. They unpack what's driving the opposition movement, where Washington and Silicon Valley diverge from the rest of the country, and how the AI infrastructure buildout might influence voters in November.

This event is made possible by general funding to CSIS and the CSIS Wadhwani AI Center.