Treasury Secretary Scott Bessent emerged from talks with Chinese Vice Premier He Lifeng in New York on Sunday with an idea built for a dangerous technology and an older form of statecraft: Washington and Beijing should warn one another when an artificial-intelligence incident reaches the level of national security.

The proposed “notification mechanism” is not yet a treaty, hotline or published protocol. It has no disclosed threshold, timetable or enforcement provision. Bessent said the countries want a shared view of common goals and threats; Chinese officials showed interest in continuing the discussion, according to people following the talks. Presidents Donald Trump and Xi Jinping are due to meet this week.

Still, the proposal is more consequential than its sparse architecture suggests. Governments create notification systems when they accept that an accident inside one jurisdiction may quickly become a crisis inside another. Nuclear-risk hotlines assume a launch can be mistaken for an attack. Cyber channels assume malicious code will not remain inside the network where it first appeared. An AI channel would acknowledge something neither capital has been eager to advertise: the most strategically important models may produce events their makers cannot contain, interpret or quietly repair.

The first dispute will be vocabulary. Is a model-assisted intrusion into critical infrastructure an AI incident or an ordinary cyberattack with a new tool? Does a laboratory report a model that produces dangerous biological instructions, or only an attempt to use those instructions? Would a financial-market disruption qualify before losses become systemic? Must an incident involve an autonomous system, or merely a human acting at machine speed?

Those questions are not drafting trivia. A threshold set too high produces a ceremonial phone that never rings. One set too low becomes a stream of accusations and commercially sensitive disclosures that neither government will tolerate. The countries also have opposite incentives when attribution is uncertain: the state suffering an incident may want early warning; the state whose company or model is implicated may want proof.

The talks arrive during a remarkable split-screen week. Several leading AI executives have called for coordinated restraint as capabilities accelerate, while other technology leaders oppose new limits. Markets, meanwhile, opened Monday by bidding chip and AI-linked shares higher. Alarm in the conference room and appetite on the trading floor are not contradictory. They are the business model: extraordinary risk justifies extraordinary strategic importance, which justifies extraordinary spending.

A bilateral notification system cannot settle that contradiction. It could do something narrower and genuinely useful. It could force both governments to maintain named contacts, define categories of severe incidents and practice communicating before a real event mixes technical uncertainty with military suspicion. Even a thin channel can reduce the chance that silence is interpreted as concealment or preparation.

But the order matters. The announcement is not the mechanism. A functioning system needs a common incident taxonomy, authentication procedures, protected lines, rules for handling proprietary information, escalation paths and rehearsals. It also needs an answer to the ugliest question: what happens when either government believes the other side’s model caused harm and the other side declines to call?

For now, Washington and Beijing have agreed that an alarm may be necessary. That is the diplomatic achievement. The engineering begins when they decide what noise it makes, who hears it and what each side promises to do before hanging up.

Oil prices dropped more than 2 percent Monday and reached an 11-day low as traders attached value to two hopeful signals: diplomacy around the United Nations meetings in New York and evidence that some Middle Eastern supply was recovering.

The physical picture was less serene. Shipping data cited by Reuters showed only 17 commodity vessels transiting the Strait of Hormuz over the weekend, compared with 37 a week earlier. U.S. Central Command said the volume of crude, cargo and liquefied natural gas moving during the past two weeks exceeded any point in the previous six months. Many escorted ships, however, were sailing with transponders switched off, making the claim difficult for outsiders to verify.

That gap explains Monday’s market better than the headline “oil slides.” Traders were not declaring the Gulf safe. They were reducing the price assigned to its worst immediate outcome. Saudi exports reportedly recovered to just over four million barrels a day so far in September after falling to 2.4 million in August. Additional barrels can soften a disruption without resolving it.

Monday also supplied a reminder that the route remains hazardous. The United Kingdom Maritime Trade Operations center said an unknown projectile hit a tanker entering the strait. The vessel reportedly continued under its own power.

A market price must compress incompatible evidence into one number: fewer visible crossings, more asserted military escorts, improving exports, live negotiations and a tanker struck but not stopped. Brent’s decline means the marginal buyer found de-escalation more plausible than on Friday. It does not mean the waterway reopened in any ordinary sense.

The most important data this week may therefore be mundane: counted hulls, delivered cargoes and insurance terms. Diplomatic language can move a screen immediately. A barrel proves its journey only when somebody receives it.

German Chancellor Friedrich Merz said he intends to remain leader of both the government and his Christian Democratic Union after a punishing sequence of state elections.

The most startling result came in Mecklenburg-Western Pomerania, where the CDU fell below the 5 percent threshold required to retain seats in the regional parliament—the first such exclusion in the 77-year history of Germany’s postwar federal republic. The Alternative for Germany finished first there, its second state-election victory this month. In Berlin, the governing parties also lost ground and the existing coalition lost its majority.

No immediate leadership challenge was visible Monday, according to the Associated Press. That gives Merz time, but not a simple route. Germany’s other major parties continue to refuse coalitions with the AfD. The barrier may keep the party from executive office while making coalition arithmetic progressively harder for everyone else.

The results are regional rather than a federal mandate, and Germany’s next national election is due in 2029. They nevertheless measure a problem the chancellor cannot postpone: voters are connecting national leadership with persistent concerns about the economy and cost of living, while the CDU is losing support in places where it once expected to compete.

Merz’s survival is therefore the least informative part of the morning. The German system is built to absorb bad regional results without converting each one into a national collapse. The question is whether the government uses that stability to change course—or merely discovers how long a political center can remain in office after disappearing from parts of the electoral map.

CNN, MS NOW and Politico notified the Trump administration Monday that they are jointly suing over the loss of White House access, arguing that the government selectively excluded them in violation of the First Amendment.

The White House announced the bans Friday. Reporters for the three organizations were subsequently denied entry and had credentials confiscated over the weekend. The organizations say they are seeking restoration of access.

The case turns a presidential attack on named outlets into a dispute over government process. The Constitution does not guarantee every reporter an interview or a seat in every room. Once the White House establishes a credentialing system and grants access across the press, however, courts have previously examined whether officials may withdraw that access without adequate procedure or on the basis of viewpoint.

The practical consequences extend beyond three newsrooms. Major television networks rotate responsibility for pooled presidential video. Axios reported that the president’s trip to the United Nations would lack a traveling television camera after the networks declined to replace CNN in the rotation. A credential decision can therefore change the material available to outlets that were not banned.

The administration and the plaintiffs will make their arguments in court. The factual question to watch first is narrow: what written criteria and process, if any, governed the exclusions? That record will determine whether the episode remains a dispute at the gate or becomes a durable rule about who may report from behind it.

Anthropic has established a laboratory in the San Francisco Bay Area for physical biology experiments, moving the maker of Claude from advising life-science work toward operating part of the experimental loop itself.

The company’s head of life sciences, Eric Kauderer-Abrams, confirmed the wet lab to Reuters. Anthropic said the facility is not devoted specifically to drug discovery, though it is expanding work in biology while competitors and pharmaceutical companies race to use AI in research.

A wet lab changes the corporate verb. A model can propose an experiment in seconds, but biological progress still depends on samples, instruments, contamination control, replication and results that refuse to conform to the prompt. By housing physical work, Anthropic can feed experimental outcomes back into model development without waiting for an outside customer to translate between software and bench science.

That compression could make the system more useful. It also concentrates responsibility. Model errors that once ended as incorrect text can become poorly chosen experiments; experimental artifacts can return as misleading training signals. The institution operating both halves must decide which findings deserve replication and which capabilities deserve restriction.

Competitors have already demonstrated how long the physical timeline remains. Alphabet’s Isomorphic Labs has spent years pursuing AI-assisted drug development and has pushed its clinic target to the end of 2026. Molecules do not honor software release calendars.

Anthropic’s laboratory is therefore neither a miracle factory nor a decorative side room. It is a test of whether an AI company can develop the disciplines that biology imposes: traceability, patience and the humiliating possibility that nature will return an error message with no explanation.

More than 13 million books were destroyed in Ukraine during two months this summer, according to the country’s Ministry of Culture. The total equals roughly one-third of Ukraine’s annual print run and caused at least 1.75 billion hryvnias, or $39.3 million, in damage.

The losses came from Russian strikes on publishing infrastructure: the warehouses, printing capacity and distribution machinery that sit between a manuscript and its reader. The Kremlin did not respond to an Associated Press request for comment about the attacks.

A destroyed book is unusually easy to misclassify during war. It is neither a person nor a bridge; photographs of blackened paper can look symbolic beside the arithmetic of casualties and power grids. But publishing is an industrial system. When a warehouse containing schoolbooks and new titles burns, the loss propagates through writers’ royalties, booksellers’ inventories, printers’ schedules and children’s classrooms.

Ukraine entered this destruction with demand for Ukrainian-language books rising after Russia’s 2022 invasion. That appetite made publishing part of the country’s cultural independence precisely when its physical base became vulnerable. A reader can choose a language only if the presses can print it and a truck can deliver it.

The financial scale is also deceptive. Ukraine now spends an estimated $190 million for a single day of war, according to the head of parliament’s budget committee. Against that figure, $39.3 million can appear manageable. It is not. A small publisher’s backlist, warehouse and working capital can disappear in the same blast, while reconstruction money naturally flows first to housing, electricity and defense.

Books are memory in a reproducible form. War attacks that reproducibility when it destroys not only individual copies but the machinery capable of replacing them. Thirteen million missing volumes are not simply a cultural elegy. They are an interrupted supply chain for a country’s account of itself.

Britain’s air-traffic controller fixed Monday’s connectivity fault at its Prestwick center and emphasized that safety procedures worked. Traffic was restricted while engineers investigated; aircraft remained protected. That is important. It is also an incomplete definition of success.

This was NATS’ second major technical disruption in two weeks. The earlier failure, attributed to a previously unknown software defect triggered within a millisecond, led to more than 2,000 cancellations and days of disruption. NATS says Monday’s Scottish network problem was unrelated. That defense should make passengers less comfortable, not more. One repeated bug can be patched. Two unrelated failures suggest a system with more than one route to public paralysis.

Safety-critical operators often speak as though preventing catastrophe completes the assignment. It does not. Air-traffic infrastructure is also expected to degrade gracefully: isolate a damaged component, transfer work and continue at reduced capacity without turning a technical exception into thousands of missed connections.

The public rarely sees successful resilience. It looks like nothing: a server fails, traffic moves elsewhere, the departure board stays dull. That invisibility makes redundancy easy to underfund and apologies easy to standardize.

NATS owes regulators and airlines more than a claim that this fault was different. It should publish the dependency chain, explain which fallback operated, identify why capacity still fell and state what independent testing will follow. The standard is not perfection. It is whether an organization entrusted with a national nervous system can absorb ordinary imperfection without exporting it to everyone standing beneath an airport screen.

“Toy Story 5” reaches Disney+ this week, which means a new generation can meet characters originally built around the terror of being left behind. The toys have survived their first child, several endings and the industrial discovery that closure is an underperforming asset.

There is nothing inherently shabby about returning to beloved characters. Folktales live by repetition; actors revisit Shakespeare without anyone demanding that Hamlet protect the integrity of the original franchise. The difference is that contemporary sequels arrive inside platforms designed to eliminate absence. A story no longer returns because an artist found a new necessity. It returns because the interface requires another recognizable tile.

“Toy Story” once made growing up feel consequential. Andy’s attention was finite. Affection moved, bedrooms emptied and loss gave play its value. The franchise economy offers a softer cosmology: nobody truly leaves, the catalog never closes and childhood waits behind a login.

Parents may welcome a familiar film, and children are perfectly capable of loving episode five without reading a balance sheet. The melancholy belongs to the adults. We have built entertainment systems that preserve every childhood object while steadily removing the experience those stories once taught—the clean, frightening dignity of an ending.

Stream it. Enjoy it. Then let the credits finish before the platform begins counting down to whatever refuses to die next.