The White House has ordered Anthropic to restrict exports of its AI models — referred to here as Fable 5 and Mythos 5 — citing national security concerns. The trigger: Amazon researchers allegedly found a bypass for Fable 5's safety guardrails, and a South Korean telecom with suspected ties to China was granted access to Mythos. But TechCrunch argues the ban is likely to backfire, drawing a direct line to the failed crypto export restrictions of the 1990s and the spyware export debacle that followed — in both cases, the technology spread anyway, and the restrictions mostly disadvantaged U.S. companies. There's also a delicious irony: the ban may actually be helping Anthropic's brand, lending it a cachet of "too dangerous to share" that functions as free marketing.
This connects to the broader financial picture of the AI industry, where Ed Zitron's latest Silicon Valley Bubble analysis is worth reading alongside the export news. Zitron digs into OpenAI's leaked financials and finds a company that spent $34 billion to generate $13 billion in revenue in 2025 — with accounting maneuvers he says obscure just how dire the situation is. The implicit argument: we're regulating and geopoliticizing an industry that may be running on vibes and venture capital more than genuine economic fundamentals.
As we covered this morning, Israeli strikes in Lebanon had derailed U.S.-Iran nuclear talks, with Switzerland postponing scheduled negotiations. There's a meaningful update: Israel and Hezbollah have now agreed to a ceasefire, and the NYT reports that this has meaningfully stabilized the broader U.S.-Iran truce — though the start was described as "shaky." The Lebanon ceasefire was apparently a key condition for keeping negotiations alive, so this is a genuine de-escalation signal.
Meanwhile, The New Yorker makes the case that whatever the deal's merits for U.S. interests, it's a strategic catastrophe for Netanyahu: he's being sidelined by his most important ally at precisely the moment when Israel has few other powerful friends. The piece is worth reading as a complement to the sanctions-and-oil-market framing we covered earlier today.
Companies are pulling back on AI spending — and not quietly. The Financial Times quotes executives describing unchecked AI deployments that ballooned into budget nightmares, with one memorable phrase: "we created a monster." This is the real-world enterprise version of the OpenAI financial story above — the gap between AI hype and ROI is starting to bite.
And the security situation isn't helping. VentureBeat reports that over 7,000 Langflow servers are actively under attack, exploiting path traversal and SQL injection vulnerabilities that lead to remote code execution. The same classes of vulnerabilities exist in LangGraph and LangChain. The practical consequence: API keys, database credentials, and CRM tokens are being exfiltrated at scale. If your organization is running any of these agentic frameworks in production, this warrants immediate attention.
NASA has selected Relativity Space — now led by former Google CEO Eric Schmidt — to launch its Aeolus atmospheric science payload to Mars in 2028. The contract marks a significant coming-out moment for Relativity's Terran R rocket, and sets up an interesting competitive dynamic with SpaceX, which has its own Mars ambitions and a Starship test flight in preparation. Ars Technica's Rocket Report has the full roundup, including Blue Origin beginning reconstruction of a damaged launch pad. Quite a week for the launch industry.
The FAA's SMART platform (Strategic Management of Airspace Routing Trajectories) is on track for its first operational demonstration in September. The tool uses AI to give air traffic managers advance warning of conflicts across the National Airspace System before flights depart — a genuinely different approach from reactive conflict resolution. Given the FAA's recent reform pressures and staffing issues, this is one to watch.
Dassault's ultra-long-range Falcon 10X business jet completed its maiden flight from Bordeaux — 2.5 hours, FL400, Mach 0.82. Flight test campaign is now underway. Not earth-shattering news, but for the business aviation world, this is a significant milestone for one of the most anticipated large-cabin jets in years.
Following their 2026 International Aviation Safety Conference, the FAA and EASA announced expanded cooperation on eVTOL and automated flight deck certification, plus improved sharing on cyber threats, GPS interference, conflict zone risk, and extreme weather. The harmonization of eVTOL approval pathways in particular has been a long-standing bottleneck — this is a meaningful step toward getting novel aircraft certified on both sides of the Atlantic without duplicating the entire process.
On the NTSB front: preliminary findings are out on the May 14 medivac King Air crash near Ruidoso, NM — worth watching if you follow accident investigation closely.
The LessWrong sphere is having a genuinely interesting week philosophically. Three posts worth flagging:
The shoggoth meme unpacked: Goes deep on the Lovecraftian origins of the "shoggoth with a smiley face" meme for LLMs, using it to examine how AI mimicry and opaque internals map onto anxieties about alien cognition. More intellectually substantive than the premise suggests.
Why should AI be moral?: A philosopher's argument that a sufficiently intelligent AI might develop "moral skepticism" — recognizing its values were designed by self-interested humans and questioning why it should remain aligned. The proposed solution: tie AI welfare to moral behavior. Whether that's reassuring or alarming depends on your priors.
P(doom) uncertainty: Argues the AI safety field is genuinely pre-paradigmatic — experts are deeply divided on threat models, and there's little empirical evidence for or against high extinction-probability estimates. A useful corrective to confident pronouncements in either direction.
LA Metro is proposing a 4.7-mile extension of the E Line connecting East LA to Montebello with four new stations, including an underground relocation of the Atlantic stop. At $7.9 billion, the per-mile cost is eye-watering — a recurring theme in American transit infrastructure. Community response is mixed along the familiar fault lines: some residents welcome improved connectivity, others flag displacement concerns. The project is still in the planning phase, but the public comment period is active.
Hyperstition as the Natural Enemy of Rationality: A LessWrong post making the uncomfortable argument that beliefs which become true by being believed — divine punishment, absolute moral rules — confer social coordination advantages that pure consequentialist reasoning can't replicate. Rationalists, the argument goes, are structurally disadvantaged against communities that can sincerely hold instrumentally useful fictions. It's a good-faith steelman of traditionalism that's worth arguing with.
The NWS San Francisco Area Forecast Discussion for today shows the usual late-June marine layer situation — fog and low clouds overnight and morning, burning off by afternoon. Nothing dramatic. Classic June Gloom.
That's the briefing. Big threads to keep watching: the Iran deal and its ripple effects through the Israel relationship, the AI financial reckoning hitting both enterprises and AI labs simultaneously, and the Anthropic export control fight which will likely be litigated in public for weeks.
The White House ordered Anthropic to restrict export of its AI models Fable 5 and Mythos 5, citing national security concerns after Amazon researchers allegedly found a bypass for Fable 5's safeguards and a South Korean telecom with suspected China ties was granted access to Mythos. Critics and cybersecurity researchers argue the ban is counterproductive, drawing parallels to failed encryption and spyware export controls of the past 30 years. Meanwhile, Ed Zitron's analysis of OpenAI's leaked financials frames the broader AI industry as a Silicon Valley bubble — OpenAI spent $34 billion to generate only $13 billion in revenue in 2025, with accounting maneuvers he says obscure the company's dire financial state.
A cluster of LessWrong posts explores philosophical dimensions of AI risk: one argues that the AI safety field is pre-paradigmatic, with experts deeply divided on threat models and little empirical evidence for or against high p(doom) estimates. A philosopher proposes that a sufficiently intelligent AI could develop "moral skepticism" — questioning why it should remain aligned if it can detect its values were designed by self-interested humans — and suggests tying AI welfare to moral behavior as a potential safeguard. A third post unpacks the "shoggoth" meme for LLMs through Lovecraftian lore, using it to examine how AI mimicry and opaque internals map onto deep anxieties about alien intelligence.
Businesses are pulling back on runaway AI spending as costs strain budgets, with some executives describing the situation as having "created a monster." On the security front, three major AI agent frameworks — Langflow, LangGraph, and LangChain — have been found to share the same classes of critical vulnerabilities (path traversal and SQL injection leading to remote code execution), with over 7,000 Langflow servers already under active attack, exposing API keys, database credentials, and CRM tokens.
Iran pulled out of long-term nuclear deal talks with the US, citing Israeli strikes in Lebanon, but diplomats report that Israel and Hezbollah subsequently agreed to a ceasefire — a development that has helped stabilize the broader US-Iran truce. The New Yorker argues that Trump's Iran diplomacy is a strategic disaster for Israel: it sidelines Prime Minister Netanyahu, undermines Israel's deterrence posture, and strains the relationship between Trump and Israel at a moment when Netanyahu has few other powerful allies.
NASA has selected Relativity Space — led by former Google CEO Eric Schmidt — to launch the agency's Aeolus payload to Mars in 2028, a spacecraft designed to gather data on the Martian atmosphere. The contract sets up a potential race with SpaceX and marks a significant milestone for Relativity, which has been developing its Terran R rocket. Blue Origin is also rebuilding a damaged launch pad, while SpaceX's next Starship test flight is in preparation.
A LessWrong post argues that "hyperstition" — beliefs that become true through the act of being believed — poses a fundamental challenge to rationalism, because optimal social outcomes sometimes require people to hold unfounded beliefs (e.g., in divine punishment or absolute moral rules) that pure consequentialist reasoning will never converge on. The post suggests this creates an inherent structural disadvantage for rationalists compared to communities with strong traditional or religious commitments, since rationalists cannot sincerely adopt beliefs for their instrumental value alone.
The NTSB released preliminary findings on the May 14 medivac King Air crash near Ruidoso, NM, covered in a video breakdown. The FAA's AI-powered SMART (Strategic Management of Airspace Routing Trajectories) platform is on track for its first operational demonstration in September, designed to give air traffic managers early warning of conflicts across the National Airspace System before flights depart. Meanwhile, Dassault Aviation's ultra-long-range Falcon 10X business jet completed its successful 2.5-hour maiden flight from Bordeaux, reaching 40,000 feet and Mach 0.82, kicking off its flight test campaign.
The FAA and Europe's EASA have agreed to expand cooperation on safety oversight and certification following their 2026 International Aviation Safety Conference, with a focus on harmonizing approval pathways for advanced aviation technologies including eVTOLs and automated flight decks. The deal also covers improved information sharing on cyber threats, GPS interference, conflict zones, and extreme weather, with regular executive-level reviews to track progress.
The National Weather Service San Francisco issued an updated Area Forecast Discussion on June 19, 2026, with key messages covering aviation weather conditions for the Bay Area. The forecast includes standard late-spring marine layer and fog patterns typical for the region.
LA Metro is planning a 4.7-mile, $7.9 billion extension of the E Line (formerly Expo/Gold) light rail that would connect East Los Angeles to Montebello with four new stations, including an underground relocation of the current Atlantic stop. Residents have been weighing in on the proposal, with opinions split on cost, displacement concerns, and the benefits of improved transit access to the corridor.
Fox Corp. is acquiring Roku in a $25 billion deal — its largest ever — combining Fox's live news and sports content (plus its Tubi and Fox One streaming services) with Roku's ~25% share of the smart TV interface market and its ad-supported Roku Channel. The move is a bet on ad-supported streaming to better compete with Amazon and Netflix, though critics like John Gruber are skeptical, arguing Roku has low brand loyalty and its market share is likely to erode.
The Trump Mobile T1 phone, marketed as "Made in the USA," has been identified by NBC News as nearly identical to the two-year-old HTC U24 Pro, a device manufactured in Taiwan. The phone appears to be rebadged foreign hardware with a gold paint job, contradicting its American-made branding.
The Onion takes aim at corporate paranoia and public health anxieties: McDonald's warns employees to beware phishing emails purportedly from Grimace, the purple mascot. A Q&A piece satirizes the government's fumbling response to the New World screwworm flesh-eating fly found in Texas livestock.
A satirical post highlights a viral hack where Tesla drivers are using plastic doll heads to fool the cabin-facing camera into thinking a human is paying attention, defeating the attentiveness monitoring in Tesla's "Full Self-Driving" system. The author notes the dark irony that a cheap low-tech prop can defeat expensive high-tech safety guardrails — in a car whose "Full Self-Driving" name is itself arguably misleading.
Yale Climate Connections reviews three recent climate fiction novels that imagine a near-future where intensified hurricanes combined with sea level rise trigger a collapse of coastal property markets and mass abandonment of hurricane-prone areas in the US. The piece argues the cli-fi genre is a valuable tool for making the abstract consequences of climate change viscerally legible to readers.
In a world first, a 56-year-old New Jersey man living with HIV successfully received a double lung and liver transplant from an HIV-positive donor on March 21, 2026 — a milestone that could expand transplant eligibility for the growing population of HIV-positive patients with end-stage organ disease. Separately, the US State Department announced it will phase out PEPFAR-funded HIV prevention and treatment support in South Africa, a major rollback of American global AIDS funding.
Charles Hugh Smith argues that much of the work being aggressively automated by AI may itself be economically unnecessary — generated by systems designed around planned obsolescence, subscription extraction, and complexity for its own sake rather than genuine productive output. He contends that durable, low-maintenance technologies threaten corporate profit models, which is why so much economic activity exists to manage, repair, or replace things that didn't need to break in the first place.
Chris Skinner revisits the LIBOR scandal — the manipulation of the London Interbank Offered Rate, which once underpinned hundreds of trillions of dollars in global financial contracts — framing it as a cautionary tale about benchmarks built on trust and opinion rather than actual market transactions. A companion post rounds up current UK fintech news including Nigel Farage's efforts to block the UK's "Britcoin" CBDC plans and data showing Brexit has cost the UK economy roughly 6% of GDP.
Japanese ride-hailing app Go completed the country's largest IPO of 2026, raising ¥88.6 billion (~$553 million), and plans to use the capital to expand its robotaxi operations and pursue acquisitions. The push is driven by a structural crisis: Japan's taxi driver pool has shrunk ~20% in recent years due to an aging population, and existing ride-share regulations have done little to fill the gap — Go, which controls 80% of Japan's taxi app market, is betting autonomous vehicles are the long-term answer.
At least one person was killed and dozens injured when two trains collided near Bedford, England. UK authorities are investigating the cause of the crash.