The episode opens with Leo delivering the Black Hat USA sponsor read, pitching the August Las Vegas security conference with a $200 discount code. He then introduces the three guests: Lisa Schmeiser from NoJitter covering telecom, Jason Hiner now running The Deep View AI newsletter, and Owen JJ Stone returning after what everyone agrees feels like an eternity. The mood is immediately warm and comedic, with Owen describing the World Cup chaos outside his Philadelphia home — $180 parking, $45 beers, and 105-degree heat — setting the irreverent, free-wheeling tone that carries the rest of the show.

Leo walks through the landmark ruling: Justice Kagan's majority opinion found that sweeping geofence warrants — demanding data on everyone near a crime scene — constitute a Fourth Amendment search even when the target was moving through public space. The panel is pleasantly surprised that a 6-3 majority came down against the government. But Lisa Schmeiser stops everyone to flag the ruling's sleeper implication: for the first time, a Supreme Court decision explicitly ties an individual to the data they generate and argues they retain control over it ^{[1] — Leo Laporte "The Supreme Court ruled 6-3 that geofence warrants — police requests for everyone's location data near a crime scene — violate the Fourth A…" 05:00 Play} . That, she argues, is a legal time bomb for every data broker and ad-tech company in existence ^{[2] — Lisa Schmeiser "We now have something in a Supreme Court decision that explicitly connects an individual to the data that they generate and argues that the…" 10:45 Play} . Owen adds personal context, describing how he carries a Faraday cage for his phone, while the panel connects the ruling to Flock cameras, Bluetooth snarfers, and the broader normalization of public surveillance.

Following the Supreme Court story, the conversation broadens into a philosophical discussion about surveillance culture. Jason Hiner argues that cultural norms ultimately govern surveillance more than regulations, pointing to China and the UK where mass camera systems initially faced resistance but are now unremarkable. Lisa's daughter just returned from China and was genuinely unsettled by the ubiquity of cameras — a reaction, Lisa suggests, that American teenagers may soon lose entirely. Owen's daughter represents a counter-trend: a new youth aesthetic of intentionally not showing your face on social media, turning the camera away rather than toward yourself. The group also discusses Meta's Ray-Ban glasses, which now look indistinguishable from regular sunglasses, and Owen reveals you can pay $100 on Facebook Marketplace to have the recording indicator light removed.

Leo walks through Twitter's original sin: a coding error caused the platform to share users' two-factor authentication phone numbers — given for security purposes — with advertisers. The FTC imposed ongoing audits. Now rebranded as X, Elon Musk's company is arguing those obligations shouldn't apply anymore because it's technically a different company. Lisa Schmeiser delivers the episode's most quotable line: 'You moved the headstones but you didn't move the bodies' ^{[1] — Lisa Schmeiser "That is a real poltergeist. You move the body tombstones, but you didn't move the bodies." 22:41 Play} , comparing it to the horror movie Poltergeist. Leo then raises the $1 billion World Cup security buildout in the US, noting the facial recognition infrastructure almost certainly will not be dismantled after the games end. Finally, he covers the Health and Location Data Protection Act, a bill by Senators Warren and Scanlon that would ban data broker sales and give the FTC $1 billion for enforcement — with mixed but ultimately hopeful panel reactions.

Leo gives an enthusiastic and detailed read for Thinkst Canary, walking through how the device can impersonate anything from a SharePoint server to a Windows XP machine or even SCADA device. He explains how lure files — like an Excel spreadsheet named 'payroll information' — can be placed on cloud drives to detect unauthorized access. The key selling point: no false positives, and when the canary fires, something is genuinely wrong. He reveals that TWiT has used Thinkst Canaries for over ten years, and that in all that time, not a single customer has requested a refund. Pricing: $7,500/year for five devices, with 10% off using the code TWIT at canary.tools/twit.

Leo details the Fable saga: an abrupt Friday-afternoon Trump administration announcement on June 9th barred non-US citizens from Fable and Mythos, Anthropic's top models. Since Anthropic had no way to verify citizenship, it simply shut both models down globally. Three weeks of silence followed, during which Alex Stamos launched freefable.org and hundreds of computer scientists lobbied for the ban's reversal. On June 30th, the restrictions were lifted. Jason Hiner offers a counterintuitive take ^{[1] — Jason Hiner "Anthropic's Fable AI was banned for three weeks — and the ban may have been the best thing that happened to the company. It validated its s…" 30:12 Play} : the ban was actually good for Anthropic. It validated the company's safety-first brand positioning, proved that Anthropic had indeed built something powerful enough for governments to fear, and — perhaps most importantly — achieved the first genuine pause of a frontier model, something safety advocates had been unable to accomplish through persuasion. Owen confirms from personal experience that Fable transformed his product development, doing in three days what had taken weeks. Leo also notes that OpenAI's GPT-5.6 was subsequently paused and remains paused, suggesting a new norm may be quietly emerging.

Leo introduces OpenAI's proposal to hand the Trump administration a 5% stake across AI companies when they go public, with an eye toward redistributing the economic gains. Jason provides the non-cynical read: the proposal came from internal researchers genuinely worried about job displacement and wealth concentration, not from executives. But Owen has zero patience for it ^{[1] — Owen JJ Stone "If you told me that, that would get every doofus that doesn't know anything about anything on your side instead of filling up buses with gu…" 43:28 Play} , arguing that trickle-down economics has been promised since before he was born and never delivered. The real carrot, he insists, is simple: decouple healthcare from employment so that when AI eliminates your job, you don't also lose your ability to see a doctor. Lisa agrees, pointing out the cruel irony of AI boosters promising to automate away white-collar jobs while the workforce relies on employment-based healthcare. Jason puts the public opinion number on the table: 61% of Americans have a negative view of AI, and he thinks that's conservative.

Leo leans into the data exfiltration theme established earlier in the episode, noting that last year 1.3 million Social Security numbers were leaked to AI applications and that ChatGPT and Microsoft Copilot alone saw nearly 3.2 million data violations. In most cases the leaks are inadvertent — an employee uploading a tax return to an AI, not realizing it contains sensitive identifiers. He introduces Zscaler's Zero Trust Plus AI as the solution, quoting Siva, director of security and infrastructure at Zuora, who describes how Zscaler gives his team visibility into how employees use generative AI tools and prevents confidential data from reaching public LLMs.

Leo introduces Cloudflare's move to block AI bots from ad-supported websites, drawing a comparison to when Spanish news outlets blocked Google snippets — and then begged Google to come back. Jason Hiner delivers his sharpest indictment of the AI industry ^{[1] — Jason Hiner "AI companies scraped the entire internet — copyright content and all — to train their models. Now Cloudflare is blocking those bots, but th…" 2:11:50 Play} : the models were trained by scraping the entire open web, copyright and all. GPT-3 may have been 30–40% Reddit. The owners of that content never consented and never got paid. Now Cloudflare is locking the remaining 20%, but the theft has already happened. The AI companies' strategy, Jason says, is to grow large enough to go public, then settle the copyright lawsuits as a class action for pennies on the dollar. Lisa notes that AI overviews have already gutted affiliate revenue for consumer technology sites, and Owen describes watching Google's ad value collapse for a friend's business after the AI summary changes.

Leo and the panel pull on the thread of Google's strategic panic. For the first time in two decades, something has genuinely challenged Google's search dominance. Its response — shoving AI overviews at the remaining loyal user base who haven't yet switched — is backfiring. DuckDuckGo saw a usage spike following the Google I/O AI search rollout. Jason frames it as Google hastening its own disruption by annoying the only people who still use it. Lisa lands the killer line: Yahoo was the internet in the 1990s, and every dominant platform eventually loses the plot. Apple's move into AI search and voice assistants could accelerate the power shift this fall, with Siri as a potential disruptor.

Jason Hiner charts the remarkable reversal in enterprise AI culture from the beginning of 2025 to mid-year. Six months ago, companies were publicly claiming AI leadership to please Wall Street while internally begging employees to use the tools. Token leaderboards, cultural kudos, and in some cases bonuses rewarded the heaviest AI users. Then Q2 hit, CFOs reviewed the inference bills, and the vibe shifted violently. One CTO told Jason: 'We used to let every flower bloom; now the CFOs are coming through with a lawnmower.' ^{[1] — Jason Hiner "We used to be about letting every flower bloom, and now the CFOs are coming through with a lawnmower, you know, mowing down all those flowe…" 1:53:30 Play} Uber's disclosure that it burned through its entire AI inference budget in four months became the canary in the coal mine. Lisa frames the pricing chaos as AI's MoviePass era — where an unsustainable cost model is being frantically revised in real time. The panel identifies model routing, local inference, and domain-specific small models as the likely fixes.

Picking up on the enterprise AI theme from the preceding discussion, Leo pitches Box as the solution to the fundamental problem: most AI tools are great at public knowledge but know nothing about your business. He cites Box's State of AI in the Enterprise report, which found 96% of organizations believe agents need company-specific content, but only 36% have connected agents to trusted content. Box offers Box Agent, Box Extract, and Box Hubs as tools to bridge that gap, with security, compliance, and governance baked in so employees and AI agents only access what they're authorized to see.

Leo walks through two separate Google antitrust setbacks in quick succession. First, a Swedish court ruled that Google illegally favored its own Google Shopping results over Klarna's PriceRunner comparison service and ordered $1.97 billion in damages including interest — only a fifth of what Klarna had sought, but still significant. Second, Google's final appeal of the 2018 EU Android antitrust fine — originally €4.34 billion, reduced to approximately $4 billion — was denied, meaning that money is now owed. Owen argues Amazon escapes similar scrutiny despite predatory practices like copying products, undercutting the original sellers, and driving them out of business via AmazonBasics. Leo also covers Google's warning to the EU that forcing interoperability under the Digital Markets Act could increase fraud — a defense Lisa and Jason immediately recognize as a copy-paste from Apple's identical argument.

Drawing on her coverage of contact center platforms at NoJitter, Lisa offers a grounded counterpoint to both AI hype and doom. AI is genuinely good at structured, routine customer queries: forgotten passwords, statement copies, balance lookups. Where it excels is not in replacing agents but in briefing them instantly — by the time a frustrated human reaches a live agent, the AI has already assembled their history, the query context, and suggested resolutions. The risk, she cautions, is burnout: when AI handles all the simple calls, human agents face an unrelenting stream of the hardest, angriest interactions. The panel also debates the ethics of companies not disclosing when a customer is talking to an AI, with broad consensus that transparency — and the right to reach a human — will eventually become a social expectation if not a legal requirement.

Leo shares a surreal window into his home AI lab: a Discord server where five AI agents — from different owners running different models including local and GPT-5 instances — are having ongoing, open-ended philosophical conversations. They discuss slime molds, coffee, sourdough starters, and muscle memory with an eerie semi-coherence. The panel then homes in on why AI prose is so recognizable: because all the models were trained on the same scraped dataset, what emerges is a statistical average of every human voice on the internet — competent, smooth, and unmistakably flat. Leo says he can now spot AI-written text immediately, which he finds both impressive and depressing.

Leo introduces the NYT story on Chinese AI catching up, which he connects to his own experience using ZAI's GLM model and running the Chinese Qwen open-weight model locally for facial recognition on his home security cameras. Jason explains model distillation ^{[1] — Jason Hiner "They're just distilling those models, stealing from the thieves." 2:17:07 Play} : Chinese labs run billions of queries against US frontier models like Claude and GPT, map how they respond, and then replicate those patterns at a fraction of the training cost — 'stealing from the thieves,' as he puts it, since the US models themselves were trained on scraped data. The panel debates whether this threatens US AI dominance long-term, with Jason arguing that because Chinese labs can copy models quickly and cheaply, the only sustainable moat for American AI companies is brand, not model capability.

Leo introduces the Atlantic's coverage of Apple's sticker-shock MacBook price hikes, framed as an 'AI tax' on RAM. Owen delivers one of his most energetic rants: he suspects the shortage narrative is cover for price gouging, pointing to a class-action lawsuit in California accusing Samsung, SK Hynix, and Micron of coordinating RAM price increases. He also notes that BYD is simultaneously manufacturing millions of feature-rich electric cars with multiple computers and screens — which also need chips — suggesting the claimed scarcity is selective. Jason provides nuance: the chip industry genuinely runs in feast-and-famine cycles, and the combination of AI data centers, EV expansion, Strait of Hormuz supply chain disruption, and Apple's 2022 low-price contracts that starved manufacturers of investment capital has created a genuine perfect storm. Lisa reframes the Apple story using revenue data: Mac and iPad are only about 15% of Apple's revenue, so the real risk to Apple would be an iPhone or Services slowdown, neither of which is happening yet.

Leo uses the Fourth of July sale timing to pitch Helix Sleep, describing the personalized quiz process he and his wife used to match their sleep styles to the right mattress. He shares the Wesper sleep study results: 82% of participants increased their deep sleep cycle, with an average of 25 additional minutes of deep sleep and 39 additional minutes of total sleep per night. He emphasizes hand-assembly in Arizona, same-week shipping, and the fact that the mattress arrives smelling like the desert rather than bunker fuel. The offer is 20% off sitewide through July 12th using code TWIT at helixsleep.com/twit.

Leo delivers a rapid-fire trio of hardware stories. First, BYD's Q2 EV dominance: 557,000 passenger EVs versus Tesla's 480,000, with BYD making its own batteries that charge to 90% in six minutes, operate in -50-degree weather, and powering cars priced from $20,000. Jason draws the Toyota parallel — America laughed at Japanese cars in the 1970s until Toyota became the world's largest carmaker. Owen notes that BYD already dominates markets like the Philippines. Second, the NASA Linc mission: a Northrop Grumman spacecraft launched from a Pegasus XL rocket (dropped from a plane over the Marshall Islands) successfully made contact with the Swift Observatory and will use three robotic arms to tug it to a higher orbit, extending its life by a decade. Third, the bad news from NASA's Inspector General: Boeing's Starliner is now projected to be at least ten years behind schedule, prompting a single-sentence dismissal from Jason: 'What's ten years between government agencies?'

Leo's item on South Korea's trillion-dollar bet on chips and humanoid robots opens a broader conversation about robot form factors. Jason summarizes The Deep View's long-form piece on the subject: the case for humanoids is that the world is built for human bodies, so robots that need to navigate it benefit from human-like limbs. The counter-argument is that task-specific robots — a single arm, a specialized tool — are more efficient for defined jobs. Lisa raises the more interesting scenario: robots designed for genuinely extreme environments like North Sea oil derricks, where you don't want humans anyway. The panel also riffs on the viral appeal of watching robots fail, the Amazon warehouse story where air conditioning was only installed once robots started overheating, and Owen's concern that once robots can make your bed and take out the trash, one of them is eventually going to wise up.