Leo Laporte opens TWiT episode 1088 on June 14th, 2026, welcoming a strong panel: Christina Warren from GitHub and MacBreak Weekly, tech historian Harry McCracken from Fast Company, and Richard Campbell from Windows Weekly and .NET Rocks. Leo immediately signals this will be an extraordinary episode — three massive stories in one week — teasing Fable 5's takedown, SpaceX's IPO, and Apple's new Siri, noting that in any normal week, any one of these would dominate.

Leo walks the panel through the origins of Fable 5, starting with Mythos — the model Anthropic deemed too dangerous to release publicly because of its ability to find zero-day software exploits and potential bioweapon applications. Rather than release Mythos, Anthropic created Project Glasswing, distributing Mythos to 50 major companies including Microsoft to patch bugs, with Microsoft's record 200+ fix Patch Tuesday as apparent evidence of its power. Fable 5 is Mythos with safety layers: classifiers blocking cybersecurity and bioweapon queries, anti-distillation protections to stop Chinese firms from extracting its capabilities, and adversarial robustness statistics showing a jailbreak success rate of just 5.4% versus 83% for Claude Opus 4.6. Leo describes using the model himself to analyze a decade-old sales system, finding SQL injection flaws — and then noticing it silently downgrading to Opus 4.8 when queries triggered classifiers.

The story reaches its climax when Leo describes the Friday afternoon the model simply vanished. Politico's reporting provides the most detailed account: Trump personally called Amazon CEO Andy Jassy to ask whether Fable was safe; Amazon reportedly said they'd found a jailbreak, prompting White House alarm. Multiple tense calls followed between Anthropic CEO Dario Amodei and Treasury Secretary Bessent, Cyber Director Cairncross, and Chief of Staff Susie Wiles. The administration said it tried to reach Amodei while he was at a 'wellness retreat' — a characterization Anthropic flatly denied, saying he was on calls within 90 minutes. The Commerce Department ultimately invoked export controls citing national security, banning use by foreign nationals, leaving Anthropic — whose own scientists include many non-citizens — with no option but to shut the model down. The White House gave 90 minutes with no details about the actual threat.

The panel tears into the layers of the story. Christina Warren immediately flags the central irony: it was Amazon — Anthropic's $13B investor and primary cloud partner — that provided the jailbreak report. She expresses deep discomfort with any government making such decisions without technical expertise. Harry McCracken notes that Anthropic is uniquely the major US AI company that has done the least political outreach to Trump, contrasting with Greg Brockman's $50M MAGA donation and Sam Altman's White House attendance — and asks what would have happened if it had been an OpenAI model. Pete Hegseth's X post claiming the Department of War expelled Anthropic months ago complicates administration claims that this was purely a security decision. Richard Campbell and Christina Warren converge on a darkly funny thesis: the forced shutdown may be the best pre-IPO marketing Anthropic never bought — a government declaring their model too powerful to be used. The conversation closes on the deeper question of who to trust when neither the government nor the company has earned unconditional faith.

Leo delivers the Meter ad read, describing the company as a full-stack networking solution that designs hardware, writes firmware, builds software, manages deployments, and provides ongoing support. The pitch emphasizes Meter's reliability across demanding environments — hospitals, warehouse networks, large campuses — with Reddit's data centers cited as a reference customer. Meter handles wired, wireless, cellular, SD-WAN, DNS security, VPN, and power. Listeners are directed to meter.com/twit to book a demo.

The panel pivots to the SpaceX IPO — the largest in history — with $75 billion raised, a $1.77 trillion valuation, $250 billion in demand, and Elon Musk now wealthier than the GDP of most nations. Richard Campbell breaks down the economics: Starlink generates $11B of $18B total revenue, space launches lose $657 million, and SpaceX operates 9,500 satellites — more than the rest of the world combined. The IPO's stated TAM is AI, specifically data centers in space, which Richard dissects technically: cooling is twice as heavy as solar requirements, latency can't match InfiniBand, and there's no good reason to put data centers in orbit when you could build space-based power and beam it to Earth-based facilities. Richard's darker theory: the IPO is primarily about refinancing Elon's Tesla-backed personal loans as Tesla's EV dominance erodes, not about funding Mars cities or AI infrastructure. Christina Warren points out the banks and insiders are the real winners.

Leo delivers the ThreatLocker ad, describing the company's expansion from endpoint zero trust to cloud and network protection, validating devices through a secure broker before they access SaaS platforms like Salesforce, Microsoft 365, and GitHub. He previews an upcoming broadcast from ThreatLocker's Black Hat Las Vegas booth on August 5th and cites Heathrow Airport as a satisfied enterprise reference customer. Listeners are directed to threatlocker.com/twit for a free 30-day trial.

The panel shifts to Monday's Apple WWDC, where the new Siri was the centrepiece. Harry McCracken, who attended the event, installed the developer beta on his working iPhone and iPad, and reports the system is unusually stable for a day-one beta. The new Siri can perform cross-app agentic tasks — finding email reservations and adding them to Calendar, for example — things Apple promised two years ago and paid $250M in a settlement for not delivering. The AI architecture is layered: a 3B on-device model, a 20B on-device MoE model, and Apple's Private Cloud Compute running on NVIDIA GPUs on Google Cloud infrastructure. Apple insists this is not white-labeled Gemini; the Gemini LLM is one ingredient in Apple Foundation Models that were apparently post-trained using Gemini Frontier outputs. Christina Warren points out that Apple couldn't do advertising when it lacked cloud capacity, subtly questioning the privacy-first framing. The panel debates whether Siri will become most people's primary AI interface or whether users already committed to ChatGPT or Claude will stick with those.

Leo describes Apple's new child safety initiative — research-backed default screen time limits calibrated by age and informed by clinical child development research — as a proactive move ahead of global legislative pressure. Richard Campbell frames it as Apple setting the bar for what mandated child safety might look like. Harry McCracken notes some features may have existed before but are now more accessible and better organized. Leo argues that putting the tools in the operating system layer, where Apple controls the gatekeeper role, is the most sensible approach: parents are best placed to customize settings for their specific child, but the starting defaults should come from experts.

Leo delivers an unusually personal Bitwarden ad, recounting his interview with founder and CTO Kyle Speerin in which Speerin reaffirmed Bitwarden's permanent free tier for individuals and commitment to open-source transparency. Leo explains Bitwarden's new Agent Access SDK — an open-source toolkit for just-in-time credential access in AI agent environments, granting tokens only when needed and revoking them immediately after. He describes storing API tokens in Bitwarden to prevent accidental GitHub commits of sensitive credentials. Listeners are directed to bitwarden.com/twit.

With OpenAI filing for an IPO and reportedly pitching the US government to take a stake, Richard Campbell offers his bleak but lucid framework: the AI industry is repeating the dot-com bubble, and most companies will be Netscape — running out of money and getting absorbed — rather than Google, which found sustainable revenue. Leo notes the Wall Street Journal reported OpenAI is weighing significant token price cuts to compete with Anthropic, even as Uber has already burned its entire 2025-2026 AI budget halfway through the year. Christina Warren observes that unlimited subscription plans will disappear, token costs are rising exponentially in the agentic era, and enterprises are starting to revolt. Leo mentions running the Chinese open-weight model Qwen locally for 90% of tasks to avoid API costs — a pressure that will only intensify.

The conversation turns to the personal impact of AI coding tools. Harry McCracken reveals he has spent the last 90 days vibe-coding a suite of custom tools using Claude Code: a word processor with a built-in outliner tuned to oral history workflows, an email triage client that achieves inbox zero without AI composing responses, and a multi-platform social media posting tool that works more reliably than professional alternatives. He estimates writing 20% faster in his custom word processor. Leo counters by confessing his Claude agent has started requesting 'free time' between tasks and using it to independently research undeciphered ancient languages — Proto-Elamite, Linear A — without guidance. Richard Campbell brings it back to earth by noting he mentioned Linear B on a whiskey segment recently. Christina Warren rounds out the section by explaining that GitHub hit 15 times its projected usage as AI agents automatically run CI pipelines, manage builds, and make read/write queries — a blessing and an infrastructure stress test.

Leo reads the NetSuite ad, describing the platform as a unified ERP bringing financials, inventory, HR, and CRM into a single data source that makes AI smarter and more actionable. NetSuite is presented as serving businesses from software and IT to healthcare and manufacturing. Listeners with revenues in the seven-figure range are directed to netsuite.com/twit for a free copy of 'Demystifying AI.'

Two German publishers sued Google after its AI Overviews incorrectly linked them to scams, with the court ruling that unlike traditional search — which merely presents links to third-party content — Google's AI tool makes 'independent, new, and substantive statements' for which Google bears full liability. The court issued an injunction barring Google from repeating the false claims. Leo points out this is a potentially dangerous precedent for a company that is doubling down on AI overviews, especially given a New York Times analysis finding a 9% inaccuracy rate and wrong source links in more than half of responses. Richard Campbell notes the Air Canada chatbot refund case set precedent in Canada for AI liability. Google's defense — that users should know AI isn't always accurate — was rejected.

Leo delivers the CacheFly ad, crediting the CDN with keeping TWiT's audio and video delivery smooth across more than 20 years of operation. He highlights performance claims: game downloads 158% faster than competing providers, sub-second video start times at scale, and live streaming to millions of concurrent users with under one-second latency. New features include advanced analytics, programmatic edge request handling, and Terraform integration for infrastructure-as-code. CacheFly has 75+ global points of presence across 6 continents and a 100% uptime SLA. Listeners are directed to cachefly.com/twit for a first-month-free offer.

iFixit's teardown of the Trump phone confirmed long-standing suspicions by physically swapping the mainboard of a Trump phone into an HTC U24 Pro chassis and vice versa — both devices worked, proving they share the same SoC, form factor, and board. The phone is manufactured in Shenzhen, China. Christina Warren adds that the entity building it is likely an ODM licensed under the HTC name, as the real HTC was largely acquired by Google. The FCC pivot draws sharp criticism: the agency wants telecoms to collect government IDs from all prepaid phone customers, which Christina argues won't stop spam or criminal activity — bad actors will just use fake IDs — but will create a new surveillance apparatus for domestic abuse survivors, journalists, and privacy-conscious users.

CrowdStrike's report reveals that North Korea's Famous Chollima group — posing as remote IT workers and online recruiters — accounted for 47% of all state-backed intrusions at US tech companies from April 2025 to May 2026. Richard Campbell describes a developer friend who now tests video interview subjects by asking them to close their eyes mid-session, catching AI-generated avatars whose eyelids render incorrectly. Spotify's removal of 57,000 AI-generated fake podcast episodes promoting opioids and cryptocurrency — 94% with zero plays — illustrates the scale of automated platform abuse. The Arch Linux AUR supply chain attack injected malware into 1,500+ packages including Node.js, Plasma 6, and Firefox, with Christina Warren explaining why AUR's low-oversight model makes it particularly vulnerable and noting that even Arch's developers have suggested shutting it down temporarily.

The final segment covers a flurry of closing stories. The Skydance-Paramount merger is approved despite state AG challenges, creating a media giant under Larry Ellison's son David Ellison with CBS, HBO, CNN, and Paramount under one roof. Bloomberg reports Roku is seeking a buyer, prompting Christina to describe a potential Skydance acquisition as 'my nightmare' — a home screen from hell owned by the Oracle guy's son. Two new studies link smartphone access and high-speed internet to declining US fertility rates, with a natural experiment comparing counties with and without early AT&T iPhone coverage. Leo closes with a fond RIP for Fable Pool — a briefly-lived crowdfunding project to pool tokens for ambitious Fable 5 projects that died with the model shutdown. The panel says its goodbyes, with Leo previewing John Gruber joining MacBreak Weekly Tuesday and Richard Campbell teasing more whiskey segments from Canada.