D.A.D. joke of the day: I told ChatGPT I've been feeling pretty worthless lately. It said, "You're absolutely right!"
What's New
AI developments from the last 24 hours
AI Was a Ballot Issue for the First Time. It Got Crushed.
Regulation & Policy Chips, Data Centers & Energy
For the first time, AI was something Americans voted on, and the industry had a bad night. Where data centers were on the ballot directly, they appeared to lose every time. Of the 26 data center measures with results early Wednesday, all 15 that would ban or limit data centers were passing or leading, and all 11 that would allow or support them were failing, according to a tracker of official county results compiled by AI policy advocate Brad Carson. Most margins topped 70%. The votes stretched from Ohio, which had the most measures, to Illinois, Kansas, Florida, New Jersey and California. In Wilmington, Ohio, about 80% voted down four rezonings of land tied to a planned Amazon Web Services data center campus, the Wilmington News Journal reported. In Joliet, Illinois, 79% said no to approving AI data centers. Results are unofficial, some measures were advisory, and a court has put one ban, in Emporia, Kansas, on hold. "If this was the data center election, it was a really bad night for the industry," Heatmap's Jael Holzman wrote.
Meanwhile, the party that campaigned against the industry swept into power. Democrats won virtually every race considered a tossup — and many that weren't: in both federal chambers, in statehouses, governorships, and local races, from Texas to Kansas, winning seats they hadn't won in generations. Democratic campaigns and allied super PACs spent heavily on attack ads about data centers, Heatmap reports, tying opponents to the industry through their votes, past support for projects or tech stock holdings. Sen. Jon Ossoff, re-elected in Georgia, used his victory speech to contrast tech billionaires building bunkers with their own warnings that AI could wipe out humanity, his campaign's transcript shows. Texas showed how far the issue reached. Democrat Gina Hinojosa unseated Gov. Greg Abbott after campaigning against data centers, calling for a moratorium on new ones and an end to industry tax breaks. "Elon Musk moving to Texas to get away from Dems, becoming the biggest partisan megadonor in history, then spending around a half-billion dollars to have his own state go blue immediately in the next election — with data centers the top gov issue — is pretty impressive," NBC News' Benjy Sarlin posted on X.
How much did it matter nationally? In the NBC News exit poll, 23% of voters called AI and data centers crucial to their vote for Congress, and they backed Democrats 66% to 31%. Voters for whom it mattered less split almost evenly, 49% to 48%. But among the issues NBC asked about, AI and data centers ranked lowest in importance, a point tech-industry ally Adam Kovacevich stressed on X. Many voters saw the election mainly as a rebuke of President Trump, NBC's poll found. Stock futures fell overnight as the results came in, with fears about AI and data centers weighing on tech, market coverage reported.
What will Democrats do with the power? They may or may not be able to cobble together a majority to pass AI legislation in Washington. They will certainly pass all sorts of bills at the state and local level. That will affect the AI infrastructure buildout and AI guardrails in ways no one can yet foresee. And then there's oversight. Party leaders have discussed creating a House select committee on AI with subpoena power, an idea Reps. Ted Lieu and Bill Foster pitched to Hakeem Jeffries, who is now in line to become speaker, Nextgov reported. Rep. Greg Casar, who leads the Progressive Caucus, wants to bring in AI CEOs and question them about the risks, Politico reported.
Watch who holds the gavels. Rep. Frank Pallone, in line to chair Energy and Commerce, would oversee most tech, kids-safety and energy bills. He has opposed a separate AI committee. Rep. Jamie Raskin would run Judiciary, where Democrats sided with The New York Times last week in its copyright fight with OpenAI. In the Senate, Maria Cantwell, who led the 2025 fight that stopped a ban on state AI laws, is in line to chair the Commerce Committee.
Democrats have signaled four priorities: - Let states keep regulating AI. The Republican push for a single federal rule overriding state AI laws now looks dead. Democrats ran ads against four Republicans who had voted to block state rules, Axios reported. - Electricity bills. Jeffries has said protecting consumers from rising power costs tied to data centers would be a priority. - Kids and chatbots. Bills to restrict AI companions for minors already have support from both parties. - No bailouts. Casar and Rep. Maxine Waters want a House vote declaring that failing AI companies should go bankrupt, not be rescued.
The party isn't united. Progressives including Casar and Sen. Bernie Sanders back a ban on superintelligent AI; Jeffries hasn't endorsed it. Critics say the leaders of Jeffries' own AI commission are too close to Big Tech. The industry was ready for this: OpenAI planned a post-election policy push, and Meta has been courting Democrats, Axios reported.
Why it matters: The AI industry spent heavily, kept quiet about AI in its ads, and now faces a Congress controlled by the party that made data centers and power bills a campaign issue. Expect hearings before laws: executives under oath, documents subpoenaed, and close looks at AI companies' ties to the Trump administration, starting in January. States are likely to keep writing their own AI rules. And the exit poll gives both sides a talking point. AI is not yet a top issue for most voters, but for the quarter who say it is, it pointed one way. That is a warning every campaign will study before 2028.
Sources: NBC News (results) · NBC News (exit poll) · CBS News · CBS News (Kansas) · The Washington Post · CNBC · Nextgov · Politico · Axios (ads) · Axios (OpenAI) · Wilmington News Journal · Ballotpedia · Heatmap · Ossoff campaign · ABC News (Texas) · Benjy Sarlin on X · Adam Kovacevich on X
How Hackers Stole $6.3 Billion From Binance: A Fake Job Interview, a Booby-Trapped AI Download and Binance's Own AI Tool
Cybersecurity Finance & Banking
When we first covered the biggest crypto heist ever, nothing reported said AI played a role (D.A.D., October 30). Binance's own account, released Tuesday, says it did. The FBI has blamed North Korea for the theft of about $6.3 billion, NPR and NBC News report. North Korea denies it, CNBC reports.
Binance's report describes a months-long campaign. In June, an engineer on its wallet team searched online for a popular AI coding tool and downloaded a tampered installer. Binance caught it and replaced his laptop, but the attackers had learned who worked on the wallets. In September, a second engineer got a job approach through a legitimate hiring site. A fake recruiter interviewed him on live video, sharing personal photos and videos on social media to build trust. In a second interview, he was told to download an "assignment" onto his work computer. It carried malware Binance's security tools didn't catch. On October 20, the attackers used that computer and an internal Binance AI tool to take over two more engineers' machines, one holding a key credential. Nine days later, they used it to make Binance's wallet system sign away billions.
Stopping it was hard. Withdrawals were suspended, but the attackers had planted permissions on the blockchain itself that kept money moving. The last transfer went through nearly four hours after the first. Binance says about $2.85 billion has been frozen, it will cover all losses, and it has already replaced 98% of the stolen tokens. It also says it blocked 57 other suspected North Korean attacks this year.
"Tremendous public service from Binance to publish this," Haseeb Qureshi, managing partner at the crypto investment firm Dragonfly, wrote on X. "The Internet is now going to be full of AI-accelerated social engineering toxic sludge, and it's your job to be hypervigilant."
Why it matters: Two lessons reach well beyond crypto. First, AI tools are now a way in: workers hunting for the latest AI app can download a fake, and an in-house AI assistant with wide access can be turned against its owner. Second, the weak points were people: an engineer installing software, and another looking for a job. Qureshi's advice applies to any organization that runs on laptops: assume you'll be breached, watch around the clock, and know how to shut everything down before you need to.
Sources: Binance (on X) · NPR · NBC News · CNBC · Haseeb Qureshi on X
OpenAI Raised Less Than Half Its $30 Billion Goal, Then Cut the Target
OpenAI set out to raise $30 billion. After tepid interest from some large investors, it had secured about $11 billion in commitments, The Wall Street Journal reported Friday. The company has since cut its target to about $20 billion, at a valuation of about $1.25 trillion, the Financial Times reported. The news broke the same day the Nasdaq had its worst day since June, as investors soured on AI stocks (D.A.D., October 31). The costs keep climbing, too: Elon Musk says his Grok Bot costs too much to sustain, and OpenAI admits its Dots agents use a lot of computing power, Fortune reports.
Why it matters: OpenAI's plans depend on raising enormous sums to pay for computing power, and investors just pushed back on the size of the ask. That matters beyond OpenAI. Anthropic is preparing to go public, and a Democratic Congress has promised scrutiny of AI companies' debts and no bailouts if they fail. Money for AI is no longer automatic. Watch whether Anthropic's IPO meets the same caution.
Sources: The Next Web, citing the Financial Times · Fortune
What's Controversial
Stories sparking genuine backlash, policy fights, or heated disagreement in the AI community
Google Lets Schools Turn Off Its AI. The Price: Losing Google Search.
Education Kids, Teens & Mental Health Google
Parents and teachers asked Google for a way to keep its AI away from kids. Google said yes, with a catch. Starting early next year, it will turn off AI tools in its software for K-12 students unless school administrators choose to switch them on, The New York Times reports. But schools that restrict AI entirely will also lose Google search. "Search and A.I. are deeply integrated," said Chris Phillips, Google's vice president for education. If a school has banned AI, he said, students "will not have access to search, and they'll see a message from their school that indicates it's not available."
Schools that allow AI will get two choices for search: the regular version, with AI Overviews and AI chat, or one run through Guided Learning, a student version of Google's Gemini chatbot meant to walk students through a topic rather than hand them answers. Google already offers a links-only "Web" view of its search results, but neither option it's giving schools works that way. Google is also reversing an August change that switched on AI features by default in its Classroom app, which upset some school leaders.
The stakes are high because Google's Gmail, Docs and Classroom apps are everywhere in K-12 schools. And the pushback was already building. In September, New York City and Los Angeles, the nation's two largest school districts and both Google customers, broadly restricted student access to generative AI on school devices. Los Angeles officials had hoped students could keep searching with AI Overviews turned off. A University of Southern California education professor told the Times the move could further energize the anti-tech movement among parents and educators: "This is a particularly clear example of the tech industry going out of its way to force A.I. on us all — even if we have said we don't want it."
Lina Khan, who led the Federal Trade Commission under President Biden, went further. "Google is using its search monopoly to coerce schools into adopting AI," she wrote on X. "This highlights the danger of allowing an unregulated monopolist to control an essential service." Not everyone agrees. "This article is getting posted everywhere this morning, but this decision is being incorrectly framed as punitive," AI writer Andrew Curran posted. "The old Google search is gone."
Why it matters: Google says the reason is technical: AI is now built into search. The business logic points the same way. Google is betting its future on AI search, and running a separate, AI-free version for schools would cut against that bet. Today's students are also tomorrow's users. But the move hands critics a vivid example of a dominant company pushing AI on children, just as lawmakers are acting: New York passed a law last week barring a range of AI features for minors. Kids and screens is one of the few tech issues with support in both parties. Khan's reaction shows where the fight is headed: critics will frame it as a monopoly problem, not just a schools problem. Whoever controls Congress after the midterms, expect it to come up in hearings and state bills.
Sources: The New York Times · TechCrunch · Chalkbeat · Forbes · Lina Khan on X · Andrew Curran on X
What's in the Lab
New announcements from major AI labs
OpenAI Says Two Staffers Designed a Hardware Device With AI in a Week
OpenAI says its models can now help design physical hardware end-to-end. That includes mechanical parts, circuit boards, and firmware, not just code. As a test, two staffers used GPT-6 Astra to design a device for talking to OpenAI's coding assistant Codex in under a week, as a side project. A contractor produced prototypes in under a month. The device still needed several rounds of review because Astra didn't reliably account for manufacturing tolerances and assembly constraints. OpenAI says its hardware teams' model usage has grown 23-fold since February.
Why it matters: If AI can speed up hardware design, not just software, it reaches into manufacturing, robotics, and consumer electronics. OpenAI is inviting engineering firms and design-software vendors to test it.
Chemistry Agent Shrinks Days of Drug-Design Work to an Hour, Google Says
Health & Medicine Science & Discovery AI Agents Google
Google added chemistry-specific tools to its Antigravity AI platform, letting an agent run one drug-design iteration that normally takes days in about an hour. In a demo, the system searched thousands of compounds, analyzed how small molecular tweaks affect drug performance, and proposed new candidate structures with synthesis routes—pairing AI reasoning with established chemistry software and databases like RDKit, ChEMBL, and PubChem rather than guessing at chemistry from scratch.
Why it matters: This is a specialized research tool rather than something most professionals will touch, but it signals AI labs racing to prove their models can speed up real scientific work, not just chatbots and coding.
Landlord Cuts Property Photo Costs 99% Using OpenAI Tools
European residential landlord Heimstaden has used OpenAI tools since June 2025 to automate parts of its property management, from pricing vacant units to generating listing photos. In the Netherlands, the cost of creating a virtually staged property photo dropped from about €50 to under €0.10, potential savings the company estimates at €130,000 annually in that market alone. Heimstaden also used OpenAI's coding tool Codex to analyze over 20,000 property assets, and says half its staff adopted the AI tools within two to three months. Humans still approve final decisions.
Why it matters: It's a concrete example of AI cutting costs in an industry—real estate operations—that isn't typically seen as AI's first battleground, suggesting similar back-office savings are available well beyond tech and finance.
What's in Academe
New papers on AI and its effects from researchers
Warning Labels Barely Protect Users From Politically Slanted Chatbots
A preregistered study of 1,586 people tested whether warning labels that flag political bias actually protect users from being swayed by a politically slanted chatbot. The most detailed disclaimers (naming the bias plus explaining it) did reduce persuasion in some cases, but the effect was inconsistent and faded within a week. User traits predicted persuasion more consistently. Republicans were more persuaded and more trusting than Democrats. Heavy AI users were more susceptible, while people who said they understood AI well were more resistant.
Why it matters: As AI tools get embedded in news, search, and workplace information flows, this suggests slapping a bias warning on a chatbot isn't enough. Who's using it, and how often, matters more than the disclaimer itself.
Proposed Playbook Would Vet Medical AI Beyond Benchmark Scores
A new research framework proposes how hospitals and health systems should actually test general-purpose AI models before trusting them with patient care. Rather than relying on a single benchmark score, the method maps "critical user journeys" (real clinical tasks) across a model's lifecycle. It combines automated scoring, an AI judge grading another AI's answers, and review by human clinicians. Each is evaluated for accuracy, safety, and usefulness.
Why it matters: As hospitals rush to deploy chatbots and clinical AI tools, this is an attempt to build the quality-control playbook that's currently missing, and it signals where regulators and health systems may eventually set the bar for approval.
Same Public Comments, Different Plans: AI Mediators Prove Unstable
A new study tested AI tools that synthesize citizen input into formal proposals assigning tasks, spending, and deadlines. Researchers ran the same thirty opinions through four different AI mediators five times each. In 17% to 20% of five-run batches, some versions included specific commitments (who does what, by when) while others didn't, even with identical input. Simply reordering the opinions shuffled outcomes almost as much as swapping in different opinions entirely. Common fixes, like locking randomness settings, didn't eliminate the instability. Prompts seeking consensus also covered minority views far less often.
Why it matters: As governments and organizations pilot AI to summarize public input into action plans, this suggests a single AI-generated proposal can quietly bury alternative outcomes and minority viewpoints without anyone knowing another version existed.
Synthetic Survey Panels Miss How Context Shapes Behavior, Paper Argues
A new paper challenges a common way of building AI-simulated populations, used to pretest surveys and simulate users. The technique gives chatbots fixed personas (e.g., "32-year-old nurse, conservative, risk-averse") to predict how real groups behave. Under two stated assumptions, the researchers prove that fixed-persona simulations cannot capture behavior driven by context rather than stable traits. Richer personas and bigger models don't fix it. Their alternative approach, tested on Reddit discussions and a clinical-trial recruitment survey, improved on this but still fell short of matching real-world responses.
Why it matters: Companies increasingly use AI persona simulations to cheaply test surveys, messaging, and products. This research suggests those simulations may systematically miss how people actually behave in specific situations. That's a caution worth heeding before trusting them for high-stakes decisions.
Weather Chatbot Without Live Alerts Told Users All Was Calm During 38% of Real Warnings
A study testing a Claude-based weather-alert chatbot found it implied conditions were calm in nearly 38% of cases with an active National Weather Service alert. That happened only when the model lacked access to alert data. Giving it the NWS alert data eliminated the false-reassurance problem entirely. Telling the model to disclose its blind spot didn't help. Data alone wasn't enough either. With a generic prompt, it raised alarms in 87.5% of no-alert cases; the production prompt cut that to zero.
Why it matters: As companies bolt chatbots onto safety-critical information like weather warnings, this shows they need both live data and careful prompts. Missing either produced false calm or false alarms.
What's On The Pod
Some new podcast episodes
AI in Business — What It Actually Takes to Trust an AI System Running on Its Own - with Yogesh Bhagat of CVS Health