Executive Summary
The United States and Israel expanded a joint air campaign across Iran on Saturday and Sunday, with Iranian state media reporting that Supreme Leader Ayatollah Ali Khamenei was killed and Tehran responding with missiles and drones aimed at Israel and multiple American bases in the Gulf. The strikes, framed by President Trump as “major combat operations” and openly linked to regime change, set off emergency diplomacy at the United Nations and a fast-moving contest over casualty claims, including reports of a school strike that could not be independently verified amid an internet blackout. In Washington, the Trump administration ordered federal agencies to stop using Anthropic’s A.I. tools and the Pentagon labeled the company a “supply chain risk,” while OpenAI said it had secured a deal to deploy its models on classified military networks with what it described as strict prohibitions on mass surveillance and autonomous lethal use. Separately, Pakistan and the Afghan Taliban traded airstrikes and cross-border attacks, a second crisis that regional officials warned could spiral as attention and resources are pulled toward the Gulf.
AI & Technology
Trump Orders Federal Phase-Out of Anthropic After Pentagon Dispute
The Trump administration on Friday directed federal agencies to cease using Anthropic’s A.I. technology and began a six-month phase-out, escalating a dispute that has fused procurement policy with the industry’s most contested ethical lines: surveillance, targeting and autonomy in warfare. Defense Secretary Pete Hegseth designated Anthropic a Pentagon “supply chain risk,” a label typically associated with foreign influence and security vulnerabilities, effectively shutting the company out of defense contracting and pressuring prime contractors to follow suit.
Anthropic’s chief executive, Dario Amodei, said the company would challenge the designation in court and framed its refusal to accept broader Pentagon terms as a matter of principle. “We cannot in good conscience accede” to uses that would enable mass domestic surveillance or fully autonomous weapons, he said, describing the company’s “red lines” as rooted in “American values.” President Trump, in a Truth Social post, dismissed Anthropic’s leadership as “Leftwing nut jobs” and argued that decisions about military operations belong to the commander in chief, accusing the company of trying to “strong-arm” what he called the “Department of War.”
The confrontation is particularly disruptive because Anthropic had become unusually embedded in sensitive government work. The company, founded by former OpenAI researchers, has received up to $200 million in Pentagon contracts, and analysts had viewed its “Claude” models as among the first advanced generative systems integrated into classified environments. Anthropic has said it previously agreed to guardrails for government deployments; the company and people familiar with the negotiations described the current breakdown as stemming from a Defense Department push for “all lawful use,” language Anthropic saw as a blank check that contradicted earlier assurances.
Critics of the A.I. industry argued the dispute also exposed the limits of voluntary governance. Max Tegmark, the physicist and A.I. safety advocate, said labs that resisted binding regulation in favor of self-policing had left themselves vulnerable when political winds shifted. What remains unclear is how quickly agencies can replace Anthropic systems already tied into workflows, and whether the administration’s “supply chain” rationale can survive judicial scrutiny if the fight becomes a public test of what ethical constraints, if any, contractors may impose on national security customers.
OpenAI Moves Into Classified Networks as Rival Is Pushed Out
OpenAI said on Friday that it had reached an agreement with the Department of Defense to deploy its models on classified military networks, a rapid advance that came as the government was forcing out one of its most prominent rivals. Sam Altman, OpenAI’s chief executive, said the agreement included safeguards against uses like mass domestic surveillance and “autonomous weapons” that can kill without human involvement, and he urged the government to offer similar terms broadly rather than turning disputes into legal warfare.
The announcement arrived in the slipstream of Anthropic’s ban, and the timing raised questions about how the Pentagon is defining acceptable guardrails. In public, Pentagon officials have said they have no interest in mass surveillance of Americans or in delegating life-and-death decisions to machines. Yet Anthropic contended the department’s insistence on “unrestricted” or “all lawful” use made prior promises unenforceable, effectively eliminating the company’s ability to police downstream behavior once its model was installed in sensitive systems.
OpenAI, seeking to differentiate its approach, emphasized a “multi-layered” safety posture: retaining control of its safety stack, using cleared personnel and cloud infrastructure, and relying on contract language that bars certain applications. The company’s framing also carried an implicit critique of competitors, suggesting some labs had relaxed policies to win national security business. Still, it is unclear how compliance would be monitored inside classified environments where independent oversight is limited and definitions of “decision support” versus “decision-making” can blur in operational settings.
For the defense A.I. market, the twin developments signal a narrowing field shaped as much by political alignment and contractual posture as by technical performance. Anthropic, once considered the only frontier lab with deep classified integration, is now fighting for access in court; OpenAI is positioning itself as a partner that can satisfy Pentagon timelines while advertising firm limits. The immediate test will come in the contract details: whether the same “red lines” that triggered a ban for one company are enforceable for another, and whether they hold when operations accelerate.
Geopolitics & Security
Iran Says Khamenei Killed as U.S. and Israel Expand Air Campaign
Joint U.S. and Israeli strikes hit targets across Iran beginning Saturday, and Iranian state media on Sunday said the attacks had killed Supreme Leader Ayatollah Ali Khamenei, the 86-year-old cleric who has ruled since 1989. President Trump announced the operation as “major combat operations” and described it as aimed at eliminating “imminent threats,” dismantling Iran’s nuclear and missile capabilities and, in unusually explicit terms, encouraging Iranians to overthrow their government. Prime Minister Benjamin Netanyahu said Israel was acting to remove what he called an “existential threat” and described the campaign, named by Israel “Operation Lion’s Roar,” as necessary to prevent Iran from obtaining nuclear weapons.
Even as Iranian outlets reported Khamenei’s death, the fog of war remained thick. In the first hours after the strikes, Iranian officials issued conflicting signals, with some state media suggesting Khamenei was still directing affairs and other reports pointing to a leadership compound in Tehran that appeared heavily damaged in satellite imagery obtained by The New York Times. Iranian authorities also imposed severe communications restrictions, including mobile disruptions and an internet blackout that made on-the-ground confirmation difficult. U.S. and Israeli officials, speaking anonymously to several news organizations, said they believed Khamenei had been killed; Iran’s later confirmation, carried by outlets including Tasnim and Press TV, sought to rebut claims that he was in hiding.
Iran retaliated with missiles and drones aimed at Israel and at American facilities in Iraq and the Gulf, including Bahrain, Qatar, the United Arab Emirates and other regional sites cited in early reports. The scale of damage at U.S. installations was not immediately clear, and initial claims from Iranian-aligned media and regional outlets varied widely. Iranian officials vowed a sustained response, while Mr. Trump warned Iran against further retaliation, threatening “force that has never been seen before,” language that suggested Washington is preparing for a prolonged exchange rather than a single night of strikes.
The U.S. Central Command released videos showing elements of the operation, including cruise missiles and aircraft launches, and American officials described a large regional buildup that included carrier strike groups and air defenses, signaling capacity for weeks of operations. Senator Tom Cotton, the chairman of the Senate Intelligence Committee, said “joint efforts” by the United States, Israel and Arab partners could extend “weeks, not days.” Yet the administration has offered limited public detail on target sets and legal authority, and the absence of congressional approval has already drawn private concerns among some U.S. allies, according to diplomats.
Khamenei’s reported death opens a succession crisis inside a system designed for continuity but vulnerable to factional capture. Under Iran’s constitution, a temporary leadership arrangement can pass to a council including the president, the head of the judiciary and a senior cleric, while the 88-member Assembly of Experts chooses a successor. In practice, analysts have long argued that the Islamic Revolutionary Guard Corps, already the country’s most powerful security institution, is positioned to shape — or dominate — the transition. Whether the IRGC consolidates control, whether clerical institutions assert themselves, or whether street unrest fractures the regime is uncertain, particularly as the strikes reportedly targeted senior commanders and command-and-control nodes that could disrupt coherence just as the system tries to reconstitute itself.
Civilian Casualty Claims, a School Strike Report, and an Information Blackout
Iranian officials said the U.S.-Israeli campaign caused extensive civilian harm, with the Red Crescent and government spokesmen reporting more than 200 deaths and hundreds of injuries across multiple cities. Among the most inflammatory claims were reports that a strike hit a girls’ school in Minab, with Iranian authorities alleging that more than 100 people — many of them schoolchildren — were killed. President Masoud Pezeshkian denounced the incident as “barbaric,” and Iranian officials called it a “war crime,” statements that were quickly amplified across regional media.
American officials acknowledged awareness of civilian harm reports but did not confirm the incident. Central Command said it was investigating and emphasized that precautions were taken to minimize unintended casualties. The competing narratives have been difficult to arbitrate because international news organizations face heavy restrictions inside Iran, and the new communications blackout has narrowed avenues for verification. Analysts cautioned that casualty figures offered by governments at war can be shaped by propaganda incentives, while also noting that large-scale strikes near military installations in densely populated areas carry an obvious risk of civilian death even when targets are nominally military.
The information environment has been further distorted by a flood of disinformation on social media, including recycled footage and A.I.-generated images that purported to show strikes, casualties or celebrations. Researchers tracking viral posts said many originated from high-reach accounts with paid verification badges, a dynamic that has repeatedly complicated crisis reporting. X did not respond to requests for comment in some reports, and the company’s handling of false content — particularly under the pressure of fast-breaking war footage — has remained a point of contention among policymakers and civil-society groups.
Within Iran, scattered videos showed both panic and celebration, depending on neighborhood and political constituency, and some reports described residents fleeing Tehran. But such clips are especially hard to authenticate under blackout conditions, and opposition groups have their own incentives to portray momentum for regime collapse. Mr. Trump, seizing on the chaos, called Khamenei’s reported death “the single greatest chance for the Iranian people to take back their Country,” language that echoed Israeli statements hinting the strikes could create an opening for regime change. Whether that rhetoric accelerates internal rupture or hardens nationalist backlash is one of the central unanswered questions of the coming days.
U.N. Emergency Diplomacy Exposes Divisions Among U.S. Allies
The strikes prompted an emergency session of the United Nations Security Council and cautious statements from world leaders, many of whom tried to balance condemnation of force with longstanding concerns about Iran’s nuclear program and regional proxy networks. António Guterres, the U.N. secretary general, denounced “the use of force” by both sides and warned that the escalation was undermining international peace and security, calling for an immediate cessation of hostilities.
European leaders urged restraint but also criticized Tehran. Ursula von der Leyen, the president of the European Commission, emphasized the imperative of nuclear safety and nonproliferation, while Kaja Kallas, the European Union’s foreign policy chief, called the situation “perilous” and noted Iran’s support for militant groups. France, Germany and Britain issued a joint statement calling on Iran to end its nuclear program and destabilizing actions, while urging a negotiated solution — a formulation that effectively acknowledged Iran’s role in the crisis even as it resisted endorsing Washington’s decision to strike.
Other U.S. partners were more supportive. Canada and Australia expressed backing for the American operation in statements cited in reports, and Australia’s prime minister, Anthony Albanese, said he stood “with the brave people of Iran in their struggle against oppression,” language that aligned closely with the Trump administration’s regime-change framing. Russia and China condemned the strikes more sharply, though their specific actions at the Security Council were still emerging as diplomats worked through draft statements and procedural fights.
The diplomatic divide is not merely rhetorical. If the conflict expands — through Iranian proxy attacks, strikes on Gulf infrastructure, or disruptions to shipping — European and Asian governments may face pressure at home over energy costs and the safety of expatriates and troops deployed in the region. For Washington, whose military posture now spans multiple Gulf states, the question is whether allied basing rights and overflight permissions remain stable if civilian casualty claims gain independent corroboration, or if the campaign drags into a longer war with uncertain objectives.
Regional Developments
Pakistan and the Afghan Taliban Trade Airstrikes as Border Fighting Intensifies
Pakistan and the Afghan Taliban sharply escalated cross-border fighting over the weekend, with each side claiming heavy casualties and strikes deep into the other’s territory. Pakistani officials said their forces hit 37 locations inside Afghanistan and described the offensive, named “Operation Ghazab Lil Haq,” as targeting militant facilities; Pakistan’s information minister, Attaullah Tarar, claimed more than 331 Taliban fighters were killed and more than 500 wounded, and that Pakistani forces destroyed 104 Afghan posts and captured 22. Kabul disputed the figures and accused Islamabad of attacking Afghan cities, including Kabul and Kandahar, claims that could not be independently verified.
The Taliban said it retaliated with attacks on Pakistani military positions, including strikes on bases in Miranshah and Spin Wam, and some reports claimed Afghan forces shot down a Pakistani fighter jet over Jalalabad and captured the pilot. Pakistan has not publicly confirmed that account in the reporting provided, and such claims are often contested in the immediate aftermath of air combat. But even the suggestion of a downed aircraft marks a dangerous threshold, intensifying domestic pressure on leaders in both capitals to respond forcefully.
Islamabad framed the conflict through its longstanding accusation that Afghan territory is used by the Pakistani Taliban, the T.T.P., to stage attacks, insisting there would be “no dialogue” until Kabul stopped harboring what Pakistan calls terrorism — an allegation the Taliban deny. The Taliban government, for its part, said it remained open to negotiations even as it vowed retaliation, a posture that suggests internal debate about whether it can absorb sustained Pakistani strikes without losing face among commanders and supporters.
International actors, including the European Union, Russia, Iran, Jordan, the United Arab Emirates and the U.N. secretary general, urged de-escalation and offered mediation. Yet the timing is grim: Pakistan now faces an intensifying border crisis as the Gulf region edges toward a wider war, and Pakistani leaders must also calculate how instability to the west interacts with domestic economic fragility and internal militant threats. With Washington publicly expressing support for Pakistan’s right to self-defense in earlier statements while focusing militarily on Iran, it is unclear what leverage outside powers can bring to bear if both sides decide they are already in “open war.”
From the Timeline
Middle East escalation, regime-change chatter, and information-warfare as “content”
A cluster of accounts amplified fast-moving claims about decapitation strikes and leadership losses, with outlets like @zerohedge boosting reports of senior Iranian figures being “eliminated” while also circulating a framing that modern wars are only “easy” when incentives aren’t distorted.
“I dunno man seems like wars are super easy when the objective is to win and not launder a trillion dollars to your frie…”
— @zerohedge
On the “street-level” narrative, @tobi and @tobi leaned into viral clips of celebration/dancing as symbolic proof that public sentiment is breaking against the regime (while the provenance and representativeness of such footage goes mostly unexamined on-timeline). @garrytan highlighted a more tactical thread—claims that Israel turned a mass-used Iranian prayer app into a broadcast channel—underscoring how “civilian” software surfaces are treated as contested terrain. Meanwhile, @wolfejosh amplified a finance-and-politics wing arguing over US leaders’ posture toward Iran’s regime, and @levelsio echoed a familiar critique of Europe moving slowly (“on Monday”) as events accelerate.
Platform dynamics: filtering ragebait while the news cycle drives record attention
In the middle of war-saturated feeds, @elonmusk pitched new topic-selection controls as a way to opt out of political ragebait—implicitly conceding that the default product pulls users toward it. In parallel, @Noahpinion amplified a media-critique framing that legacy outlets are being “forced” to cover pro-Trump street sentiment, reflecting how the attention economy is increasingly narrated as a contest between platforms, TV, and what’s “allowed” to be visible.
Defense AI partnerships and the “ToS as national security policy” dispute
OpenAI’s defense relationship became a flashpoint not just about capability, but governance: @sama described a classified-network deployment deal with explicit limits (no domestic mass surveillance; humans responsible for force decisions) and framed it as a template others should accept. Counterpoint energy came from @chamath, who argued that shifting terms-of-service can make model vendors strategically unreliable for governments and major enterprises, and predicted the “least restrictive” ToS providers could win adoption on pure risk management.
“My hunch is that the company that embraces the “no holds barred” ToS will win because it’s the least risky to adopt wrt long term risk of getting rug-pulled.”
— @chamath
On the startup operator side, @paulg pushed back on blanket vendor blacklisting: early-stage founders should use whichever model performs best now, even if defense sales are a future consideration—separating product iteration from later procurement politics.
Trust shifts toward open models, amid a realism that “principles are tribal” and states are reorganizing for AI
A recurring undercurrent is that institutional trust is eroding on all sides, pushing interest toward decentralization: @balajis argued the equilibrium outcome is open-source models as the only broadly “trusted” option as US AI firms collide with partisan politics, military tensions, and China-related security fears. At the same time, @fchollet emphasized uncertainty at the technical layer—scaling produces unpredictable emergent behavior—supporting the idea that governance can’t rely on static assumptions about what models will become. Internationally, @hardmaru amplified Japan’s framing of domestic AI capability as urgent national-security infrastructure, signaling that “sovereign AI” isn’t only a US/China story. And @EMostaque took the argument to its endpoint: if AI inevitably runs government functions, the real question is who it is aligned to—citizens, companies, or nobody.
Privacy backlash: chatbots as surveillance surfaces by default
The privacy conversation sharpened into a direct accusation that the major AI providers treat user chats as training fuel: @chamath highlighted a Stanford review alleging default data collection/training behavior across top US AI companies, with especially negative attention on vague opt-outs and the possibility of cross-product data merging. @sama tried to pre-empt adjacent fears on the defense side by explicitly calling out prohibitions on domestic mass surveillance in OpenAI’s agreement—an implicit acknowledgment that privacy skepticism is now inseparable from both consumer AI and national-security deployments.
Agentic coding accelerates—debate shifts from “can it build?” to “can it be trusted?” (and whether it’s addictive)
Developers and protocol builders are increasingly treating AI as a force multiplier, but the timeline shows anxiety about the failure modes. @VitalikButerin described AI “vibe-coding” as legitimately compressing roadmaps—while warning that speed has to be reinvested into verification, testing, and multiple implementations to avoid brittle systems shipping faster. In the tooling trenches, @dhh shared a practical workflow that pairs multiple coding agents (fast first pass, then “second opinion”), reflecting a new norm: orchestration matters as much as model choice. Meanwhile, @EMostaque injected a more psychological critique, characterizing “vibe code” as variable-reward skinner-box behavior—less about engineering excellence, more about compulsion loops.
Crypto rails for agents and users: account abstraction meets HTTP-native payments (with geopolitics bleeding into onchain flows)
On the protocol side, @VitalikButerin argued Ethereum’s account abstraction is finally consolidating into a clean, general “frame transactions” approach, aiming to make multisig/key changes, paymasters, and batch operations first-class without intermediaries. In payments/product land, @brian_armstrong pointed to x402 as a “checkout proof of concept for the internet,” positioning multi-chain payments as something software agents will natively negotiate. Bridging the two worlds, @paulg highlighted onchain wallets allegedly profiting from bets tied to US strike rumors—an example of how crypto markets operationalize news faster than traditional compliance regimes can comfortably explain.
“Pure software” drawdown vibes: multiples compress as compute and policy risk dominate
Investment chatter showed a defensiveness about software’s standalone appeal: @naval flatly claimed “pure software is rapidly becoming un-investable,” echoing a broader sense that distribution moats and margins are eroding. @DavidSacks amplified a discussion agenda focused on imploding software stocks and datacenter opposition—suggesting the market’s center of gravity is shifting from SaaS narratives to compute, infrastructure bottlenecks, and regulatory friction.