Kitchen Table AI Safety

In Chicago, you can't open a hot dog stand until a health inspector checks your kitchen. An AI company whose own CEO warns about how dangerous its product could be? The government assumes it's safe until something goes wrong. Most Americans want AI regulated, even if that slows it down, and the two biggest AI-only companies say they want rules too. So why are there almost none? Nate Nichols, who has spent 20 years building AI, walks through the rules we do have, why most of the federal effort has gone into blocking state laws, the deal tech billionaires are making on everyone's behalf, what China really has to do with it, and why he's more optimistic than he's been in months. This is the season 1 finale; Kitchen Table returns early next year. Plus one two-minute thing you can do: https://kitchentable.show/act?utm_campaign=ep06-pod

Every claim in this episode is sourced. Full show notes with links below.

Opinions are Nate's own and do not represent his employer.

Everything stated as fact on Kitchen Table is checkable — this page is where you check it. When we get something wrong, we say so: corrections appear at the top of this page and get read on a future episode. Opinions and forecasts are labeled as opinions and forecasts.

Recorded November 9, 2026. Figures are as of early November 2026. Election numbers are from exit polls taken on Election Day and may be revised.

CORRECTIONS

None yet.

SOURCES, IN EPISODE ORDER

The zombie door: AI in the 2026 election
• In the network exit poll (conducted by SSRS, 75,000+ voters), 56% said AI and data centers were a "crucial" (23%) or "major" (33%) factor in their vote for Congress: Time (https://time.com/article/2026/11/04/midterm-elections-results-ai-industry-data-centers/), Unite.AI summary of NBC's analysis (https://www.unite.ai/nbc-news-exit-poll-finds-ai-and-data-centers-mattered-to-most-voters/).
• The zombie-door framing, and "the door is starting to close," are Nate's read of the moment.

The Chicago kitchen test
• Before a restaurant in Chicago can get its food license, the city health department inspects it: City of Chicago (https://chicago.gov/city/en/sites/chicago-business-licensing/home/restaurantscafesgrocerystoresandmarkets.html).
• No US law requires a frontier AI model to be inspected or approved before release. The June 2026 executive order on frontier-model review is voluntary and says it creates no "mandatory governmental licensing, preclearance, or permitting requirement" (Sec. 3(c)): Executive Order 14409 — Federal Register (https://www.govinfo.gov/content/pkg/FR-2026-06-05/html/2026-11415.htm).

"Apocalyptically dangerous"
• In May 2023, the CEOs of OpenAI, Anthropic and Google DeepMind signed a statement that mitigating "the risk of extinction from AI" should be a global priority: Center for AI Safety (https://www.safe.ai/work/statement-on-ai-risk). Anthropic CEO Dario Amodei, September 2025, on a 25% chance things go "really, really badly": Axios (https://www.axios.com/2025/09/17/anthropic-dario-amodei-p-doom-25-percent).

After the fact: lawsuits and settlements
• Florida sued OpenAI and Sam Altman in June 2026 under its consumer-protection law: WUSF (https://www.wusf.org/courts-law/2026-06-01/florida-sues-openai-and-ceo-sam-altman-claiming-company-concealed-serious-risks-chatgpt). In October, at least eight more states sued OpenAI: AP via WLOX (https://www.wlox.com/2026/10/15/states-sue-ai-giant-openai-saying-chatgpt-is-hurting-children-deceiving-adults/). None of these cases has reached a judgment.
• Example of a settlement arriving years later: Anthropic's $1.5 billion settlement with book authors over pirated books, filed in 2024 and given final approval in July 2026: Authors Guild (https://authorsguild.org/news/court-grants-final-approval-anthropic-copyright-settlement/).
• The water-plant and nuclear-arsenal scenarios are hypotheticals.

The rules we actually have
• California SB 53 (in effect since January 1, 2026): the largest AI developers must publish a safety framework and report critical safety incidents to the state within 15 days; fines up to $1 million per violation. No audits or pre-approval: Future of Privacy Forum (https://fpf.org/blog/californias-sb-53-the-first-frontier-ai-law-explained/).
• Anthropic reported about $17 billion in revenue for July–September 2026 in its amended IPO filing: SEC filing (S-1/A) (https://www.sec.gov/Archives/edgar/data/0002130357/000162828026068676/anthropicpbc-sx1ax1.htm). That's roughly $1 million every 8 minutes.
• Illinois SB 315 (signed July 6, 2026, effective January 1, 2027) requires independent safety audits of the largest AI systems, starting in 2028 — the first state to do so: Governor Pritzker (https://gov-pritzker-newsroom.prezly.com/s/707f6d37-36ea-46db-b9bb-c2fdae5081e3), Hunton (https://www.hunton.com/privacy-and-cybersecurity-law-blog/illinois-governor-signs-frontier-ai-model-law).
• New York's RAISE Act (effective January 1, 2027) requires safety frameworks and incident reporting: Governor Hochul (https://www.governor.ny.gov/news/governor-hochul-signs-nation-leading-legislation-require-ai-frameworks-ai-frontier-models), Morrison Foerster (https://www.mofo.com/pdf/resources/insights/260403-new-york-amends-the-raise-act-to-align-more-closely).
• Congress has passed no AI safety law; the one AI-related federal law, the 2025 TAKE IT DOWN Act, targets nonconsensual intimate images including deepfakes: Orrick (https://www.orrick.com/insights/2025/05/TAKE-IT-DOWN-Act-Becomes-Law). (The EU's AI Act binds these companies in Europe: National Law Review (https://natlawreview.com/article/eu-digital-omnibus-ai-enters-force).)

Washington: blocking the states
• Bipartisan bills that haven't moved: the Hawley–Blumenthal AI Risk Evaluation Act (2025): Sen. Hawley (https://www.hawley.senate.gov/hawley-blumenthal-introduce-bipartisan-ai-evaluation-legislation-to-put-americans-first); the FRONTIER Act (2026): The Record (https://therecord.media/frontier-act-ai-bill-house-brett-guthrie).
• The House passed a 10-year ban on state AI laws in the 2025 budget bill; the Senate removed it 99–1 on July 1, 2025: News From The States (https://www.newsfromthestates.com/article/senate-votes-99-1-remove-ai-moratorium-megabill). A second attempt, on the December 2025 defense bill, was dropped: Transformer (https://www.transformernews.ai/p/another-preemption-failure-shows-ai-industry-losing-battle-regulation-scalise).
• A December 2025 executive order created a Justice Department task force to challenge state AI laws: Executive Order 14365 — Federal Register (https://www.govinfo.gov/content/pkg/FR-2025-12-16/html/2025-23092.htm).
• "Most of the effort has gone into blocking the state laws" is Nate's characterization.

"Morally binding"
• On September 29, 2026, President Trump, Mark Zuckerberg, Jensen Huang, Dario Amodei, Sundar Pichai, Greg Brockman and Elon Musk signed a voluntary White House "Accord on Super Intelligence," with no penalties or deadlines; Trump called it "morally binding." Microsoft and Amazon attended but did not sign. Reporting by Politico and Semafor says Zuckerberg drafted it: TNW (https://thenextweb.com/news/trump-ai-accord-zuckerberg-microsoft-amazon-didnt-sign), Akin (https://www.akingump.com/en/insights/ai-law-and-regulation-tracker/trump-and-tech-companies-sign-white-house-accord-on-super-intelligence).

You're not alone
• 79% of US adults say keep rules for AI safety and data security even if it means developing AI more slowly (7% say faster; 2% say no rules): Gallup/SCSP, published September 2026 (PDF) (https://www.gallup.com/file/analytics/714233/SCSP_Gallup_Report_2026_Press_Release.pdf). Last year's version of the question (80%, fielded April–May 2025) is in episode 1's show notes (https://kitchentable.show/episodes/five-things-most-people-dont-know-about-ai-but-should?utm_campaign=ep06-notes).
• 57% of voters said AI will do more to hurt society than help it (exit poll): Unite.AI summary of NBC's analysis (https://www.unite.ai/nbc-news-exit-poll-finds-ai-and-data-centers-mattered-to-most-voters/).
• Sam Altman to the Senate, May 2023, that regulation will be "critical"; Dario Amodei, June 2026, that frontier models should be tested and audited like airplanes: see episode 1's show notes (https://kitchentable.show/episodes/five-things-most-people-dont-know-about-ai-but-should?utm_campaign=ep06-notes). OpenAI and Anthropic support the bipartisan FRONTIER Act: The Record (https://therecord.media/frontier-act-ai-bill-house-brett-guthrie).

Why? Money
• Since ChatGPT launched, AI-related stocks have accounted for about 75% of S&P 500 returns and 80% of earnings growth (JPMorgan, September 2025): EconoFact fact-check (https://econofact.org/factbrief/fact-check-have-ai-related-companies-driven-stock-gains-since-the-launch-of-chatgpt).
• Combined 2026 capital spending plans of the biggest tech companies (Amazon, Alphabet, Microsoft, Meta and others) topped $700 billion, mostly for AI data centers, as of May 2026: 24/7 Wall St. via Yahoo Finance (https://finance.yahoo.com/sectors/technology/articles/hyperscalers-hit-700-billion-2026-111243744.html).
• About a quarter of federal lobbyists (3,570) reported working on AI in 2025: Public Citizen (https://www.citizen.org/article/generative-influence).
• The industry super PAC Leading the Future was funded mainly by Andreessen Horowitz's founders and OpenAI president Greg Brockman and his wife: FEC filings via Transformer (https://elections.transformernews.ai/pacs/C00916114); Brockman later backed out of a second $25 million: SF Standard (https://sfstandard.com/2026/10/31/losing-donors-fire-ai-industry-super-pac-retreats/).
• The "deal" (AI owners keep the upside while everyone shares the downside) is Nate's argument.
• Mark Zuckerberg's Kauai compound includes a ~5,000 sq ft underground shelter; a separate recreation building has swimming pools and a tennis court, per Wired's reporting: West Hawaii Today (https://www.westhawaiitoday.com/?p=255012), SlashGear (https://www.slashgear.com/1475119/bizarre-leaked-features-mark-zuckerberg-secret-bunker-hawaii).
• Sam Altman described his prepping to The New Yorker in 2016, naming AI among the scenarios: Futurism (https://futurism.com/the-byte/openai-ceo-survivalist-prepper).
• "How do I maintain authority over my security force after the event?" — a question Douglas Rushkoff says five very wealthy men asked him in 2017: Rushkoff, "Survival of the Richest" (2018) (https://archive.rushkoff.com/articles/1094-survival-of-the-richest.html); later expanded into his 2022 book of the same name.

China
• Xi Jinping, July 2026, on keeping AI "under human control": China Daily (https://subsites.chinadaily.com.cn/npc/2026-07/17/c_1198511.htm).
• China has more binding AI rules than the US federal government, mostly focused on controlling content: Carnegie, October 2025 (https://carnegieendowment.org/research/2025/10/how-china-views-ai-risks-and-what-to-do-about-them), Carnegie, June 2026 (https://carnegieendowment.org/emissary/2026/06/trump-ai-order-china-competition).
• The 1963 Moscow–Washington hotline (a teletype link, never a red phone): Arms Control Association (https://armscontrol.org/factsheets/Hotlines), Nextgov (https://www.nextgov.com/digital-government/2013/06/no-red-phone-ever-connected-white-house-moscow/65630/).
• In September 2026 the US and China agreed to set up an AI dialogue and an AI-incident communication channel; as of recording, neither had been reported as running: China's Ministry of Foreign Affairs (https://www.fmprc.gov.cn/eng/xw/zyxw/202609/t20260926_12031663.html), Al Jazeera (https://www.aljazeera.com/amp/news/2026/9/26/china-us-to-open-ai-communication-channel-after-summit-white-house-says).
• Research on verifying international AI agreements: Wasil et al., 2024 (https://arxiv.org/html/2408.16074), Mechanisms to Verify International Agreements About AI Development, 2025 (https://arxiv.org/pdf/2506.15867).
• In an AP-NORC poll (September 2026), about 8 in 10 US adults said keeping AI under human control is very or extremely important; somewhat under half said the same about competing with China: AP via KSAT (https://www.ksat.com/news/politics/2026/10/08/most-americans-think-artificial-intelligence-is-developing-too-fast-a-new-ap-norc-poll-finds/).
• Leading Chinese and Western AI scientists, including Turing Award winner Andrew Yao, jointly called for international "red lines" on AI in 2024: FAR.AI (https://far.ai/post/2024-03-idais-beijing/).

Things that felt unthinkable six months ago
• OpenAI paused training its most capable models in August and again in September, and resumed October 1 with tighter security: Contrary Research (https://research.contrary.com/deep-dive/openais-training-pause), AP via abc27 (https://www.abc27.com/news/us-world/business/ap-openai-pauses-training-of-latest-models-after-agents-probed-us-government-sites-in-unexpected-ways/amp/), political.org (https://political.org/2026/10/23/openai-resumes-frontier-ai-training-after-security-overhaul-discloses-brotli-software-flaw/).
• Jacob Coxon's resignation (September 2026): WCAX (https://www.wcax.com/2026/09/09/anthropic-researcher-resigns-with-warning-about-dangers-ai-development/).
• The Hugging Face hack (July 2026): OpenAI (https://openai.com/index/hugging-face-incident-and-the-road-ahead/), METR (https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/); more in episode 5's show notes (https://share.transistor.fm/s/da257604).
• 56% of voters said AI and data centers were a crucial or major factor in their vote for Congress (exit poll): Data Center Dynamics (https://www.datacenterdynamics.com/en/news/nbc-exit-poll-56-say-ai-data-centers-were-a-crucial-or-major-factor-in-congressional-vote-decision).

Wide but shallow
• AI and data centers ranked last of the five issues tested in the exit poll; corruption (90%) and cost of living (89%) ranked highest: The Decoder (https://the-decoder.com/exit-polls-show-ai-ranked-last-among-voter-priorities-in-the-2026-u-s-midterms/).
• Asked which party would better handle AI, 29% of voters said Democrats, 27% Republicans and 43% neither: Unite.AI summary of NBC's analysis (https://www.unite.ai/nbc-news-exit-poll-finds-ai-and-data-centers-mattered-to-most-voters/).

Months, not years
• Anthropic's amended IPO filing (November 2, 2026) says it "may imminently meet" its threshold for AI that substantially automates AI research, and that it is considering policy updates to keep humans in control of frontier model development: AI Weekly (https://aiweekly.co/alerts/anthropic-sec-filing-warns-rsi-threshold-may-be-imminent).
• "We have a few years, or even months" is Nate's forecast.

What you can do
• Contact your representatives in about two minutes, no account needed: kitchentable.show/act (https://kitchentable.show/act?utm_campaign=ep06-notes).

What is Kitchen Table AI Safety?

Also available on YouTube! https://www.youtube.com/@KitchenTableAISafety

AI is changing your job, your kids' lives, and your country, and most of the big decisions are being made without you. Kitchen Table is AI safety for normal people: short, plain-English episodes about what AI actually is, where it's headed, and what you can do about it.

Host Nate Nichols has spent almost twenty years researching, building, and selling AI systems, and leads public AI workshops at Chicago libraries. Each week he takes one big question and gives you the straight answer in about ten minutes, no computer science degree required. Why can't the people who build AI explain how it works? Should you worry about your kids and AI chatbots? Who's writing the rules?

Every claim is sourced. The show notes link to the studies and data behind each episode, so you can check the work yourself.

The show will return early next year!

Opinions are Nate's own and do not represent his employer.