Moroccans vote Wednesday for a new parliament after a year in which the country’s young people delivered their verdict by less orderly means: protests in the streets and a mass attempt to cross into the Spanish territory of Ceuta.

The ballot will choose 395 members of the House of Representatives and help determine the next government. It will not transfer all governing power. King Mohammed VI and the monarchy retain authority over strategic affairs, a fact that makes this election both consequential and structurally limited. Voters can change the managers of public policy without changing the institution that sets many of its outer boundaries.

That distinction matters because the next cabinet is already being described as the “World Cup government.” Morocco will co-host the 2030 tournament, and billions of dollars are due to move through stadiums, railways, roads and urban redevelopment. To investors and football officials, this is national acceleration. To young Moroccans who joined last year’s Gen Z-led demonstrations, it is an audit question: why can a state mobilize so visibly for a month of football while schools, hospitals, housing and employment remain daily sources of frustration?

The economy can grow while that argument worsens. National expansion is not the same thing as a first job, a usable clinic or an apartment within reach of a salary. The election therefore turns on distribution more than scarcity. Morocco has demonstrated that it can build. Voters are asking what the building is for and whom it reaches.

The most brutal expression of the gap arrived at the northern frontier this summer, when a mass effort to enter Ceuta left dozens dead. Migration is often discussed in Europe as pressure arriving from elsewhere. From the Moroccan side, it is also a domestic vote of no confidence cast with a body: the conclusion that the dangerous route out offers more agency than the formal institutions at home.

Not every young voter has abandoned those institutions. The campaign has featured younger candidates presenting politics as a usable instrument, and the contest could produce a landmark. If the Authenticity and Modernity Party prevails and its leader is selected, Morocco could have its first woman prime minister. But symbolism will meet a severe test. A new face presiding over the same allocation of attention will not answer the central complaint.

This is why turnout and the distribution of support will matter beyond the identity of the largest party. The election measures whether anger can be converted into representation before it converts again into exit. It also asks established parties to explain what parliament can actually deliver inside a system where royal power remains decisive.

The next government’s calendar will be seductive: contracts awarded, cranes raised, stations opened and tournament deadlines met. Those are crisp achievements, designed for television and measurable by completion date. Public confidence is slower and less photogenic. It is measured in whether a graduate receives an answer, whether a patient finds a bed and whether a family concludes that the future lies on its own side of the water.

Morocco goes to the polls with both calendars running. The World Cup has a fixed opening whistle. The country’s young people have supplied no such deadline.

The Centers for Medicare and Medicaid Services says it is canceling roughly 315,000 Affordable Care Act marketplace enrollments covering more than 760,000 people. The agency describes the enrollments as unauthorized, ineligible or fictitious and estimates that ending their subsidies will return about $2.2 billion to taxpayers.

Vice President JD Vance announced the action as part of a government-wide fraud initiative. CMS says the affected records were identified through data matching and other indicators, including cases involving people who may not exist, who may have other coverage or who may have been enrolled without authorization. The federal marketplace had about 19.2 million active enrollees in early 2026, according to federal figures cited by the Associated Press.

There is a genuine program-integrity problem behind the announcement. CMS has previously documented unauthorized enrollments by brokers, concurrent enrollment in marketplace plans and Medicaid or CHIP, and failures to reconcile advance premium tax credits. People have reported discovering policies they did not request, while brokers can have a financial incentive to move consumers among plans.

The scale of Wednesday’s action makes process as important as the headline number. A database can identify an inconsistency; it cannot by itself establish why the inconsistency exists. A person who never filed a claim may be a phantom enrollee, someone unaware of an unauthorized policy or simply a healthy customer who did not use medical care. Those cases have different meanings even when they produce the same empty claims history.

The operational test is therefore straightforward: what notice will affected consumers receive, how can they see the reason for cancellation, and can an error be reversed before a prescription, appointment or hospitalization exposes the loss of coverage? CMS’s public materials emphasize the fraud findings and fiscal recovery. The durability of the action will depend on the quality and speed of that appeals machinery.

For the administration, 760,000 is evidence of enforcement. For insurers, it is a large adjustment to the risk pool and subsidy stream. For every person attached to one of those records, however, the policy question will arrive in a smaller form: whether the government has identified fraud, corrected somebody else’s unauthorized act or mistaken an unusual life for a nonexistent one.

Saudi Arabia has restarted its East-West oil pipeline after attacks damaged pumping stations, reopening a route from the kingdom’s eastern oil fields to the Red Sea port of Yanbu. Oil prices fell as traders considered the possibility of additional Gulf supplies reaching market without using the most exposed maritime route.

The important word is “restarted,” not “restored.” Reuters reported that the line was pumping at an extremely low rate and below 10 percent of capacity, according to sources familiar with operations. Saudi Aramco also continued to withhold crude supply from some European refiners. The physical system is moving oil, but it has not resumed ordinary service.

That difference is where the market’s current optimism lives. A partial flow proves that the route is not wholly disabled. It does not prove that damaged pumping infrastructure can sustain volume, that repairs will proceed without interruption or that export schedules can be normalized quickly. The line’s strategic value comes from scale. At a trickle, it is reassurance with a pressure gauge attached.

European shares rose Wednesday as lower oil prices improved sentiment, while aerospace and defense stocks led sectoral gains. The combination is instructive: investors were willing to mark down the immediate energy shock without marking down the geopolitical risk producing it.

The pipeline is thus performing two jobs. It is carrying crude west, and it is carrying expectations much faster. Those expectations may prove correct as repairs advance. Until throughput rises materially and customers receive normal allocations, however, the restart should be read as the beginning of a repair process rather than the return of a dependable bypass.

President Donald Trump and Chinese President Xi Jinping approach another meeting with disputes over trade, artificial intelligence and the Iran war competing against both governments’ stated desire for stability.

The ceremony risks obscuring a simpler measure of the relationship: implementation. An Associated Press review found uneven progress on commitments announced after the leaders met in Beijing in May. Some undertakings have advanced, while others remain incomplete or contested. That gives the new encounter an inherited agenda before either side announces another one.

The two governments have incentives to keep the relationship controlled. Businesses need rules they can plan around; markets respond sharply to tariff and export-control signals; and both capitals face security questions that become more dangerous when routine communication breaks down. Those interests do not require political warmth. They require predictable follow-through.

AI has made that harder. Advanced chips, model access, industrial espionage concerns and proposed crisis communications now sit inside the same relationship as conventional trade. A purchase commitment can be counted at a port. An agreement concerning model behavior, technical standards or incident notification requires definitions, monitoring and trust between institutions that are organized around strategic suspicion.

The useful test of the meeting will not be the length of the welcoming ceremony or the size of a newly announced target. It will be whether the governments attach responsible agencies, verification methods and dates to their commitments—and whether the unfinished May items are closed rather than replaced by fresher language.

Summits create political inventory. Eventually somebody has to collect it.

OpenAI released GPT-6 Sol and GPT-6 Luna on Tuesday with prices that make the industry’s direction unusually legible. Sol costs $2 per million input tokens and $10 per million output tokens for standard short-context use. Luna costs 10 cents in and 50 cents out. OpenAI says the models are about 50 percent cheaper than the promotional pricing of their GPT-5.6 counterparts.

The gap inside the product line is as revealing as the absolute price. GPT-6 Astra is listed at $10 per million input tokens and $50 per million output tokens. Sol therefore sells at one-fifth of Astra’s headline rate, while Luna’s input price is one-hundredth of it. The vendor is segmenting intelligence less like a single miracle and more like cloud storage: premium capacity for exceptional jobs, cheaper capacity for everything that becomes routine.

That changes enterprise behavior. A workflow that was uneconomic at last month’s inference price may now be deployed across every customer ticket, code review or internal document. Lower prices also encourage longer agent loops and more background use, which can expand total consumption even while the unit rate falls.

The savings are real. So is the new dependency. Once a company redesigns work around a model’s price, latency and behavior, switching is not merely a matter of pointing to another endpoint. Prompts, evaluations, permissions, audit controls and employee habits form a softer variety of lock-in. The bill can fall while the cost of exit rises.

The immediate winner is the buyer with disciplined measurement: a company able to route ordinary tasks to Luna, harder work to Sol and exceptional cases to a premium system, while checking the results. The loser is the buyer that treats a cheaper token as proof of a cheaper completed task.

AI pricing has entered its industrial phase. The product launch is now a rate card, and the rate card is an invitation to reorganize a company before the next one arrives.

Qualcomm has introduced two flagship mobile processors, the Snapdragon 8 Elite Gen 6 and the higher-end Snapdragon 8 Elite Extreme Gen 6, built around what the company calls the “agentic AI age.” Devices using the chips are expected from brands including Motorola, OnePlus, Oppo, Redmi, Vivo and Xiaomi.

The Extreme platform is designed to run substantially larger AI workloads on a phone, while the standard chip includes a sensing hub for smaller models and persistent contextual tasks. Qualcomm also advertises a new Oryon CPU reaching 5 GHz, along with changes to graphics, imaging and connectivity.

The engineering direction is clear: more inference will happen on the device instead of requiring every request to travel to a distant data center. That can reduce latency, preserve function when connectivity is poor and keep some sensitive material local. It also turns the handset from a screen that waits for instructions into a machine expected to observe context and act.

“On-device” is therefore not a synonym for “private.” A locally running agent can still see too much, retain too much or take an action the owner did not understand. Privacy depends on operating-system permissions, data retention, application design and whether the user can inspect what the agent saw and why it acted. The chip supplies capacity, not governance.

The first phones built on these processors will be sold through familiar numbers: clock speed, benchmark performance and battery life. Their more important specifications will be harder to place on a retail card. Can the agent be paused? Does it distinguish a suggestion from an instruction? Is its activity log readable? What happens when two apps ask it to serve different interests?

Qualcomm has made the pocket computer more capable of initiative. Phone makers must now decide whether initiative is a feature the owner commands or a condition the owner merely discovers.

The technology industry has become infatuated with giving software a personality because personality photographs well. A voice can be warm. An avatar can wink. A demonstration can show an assistant booking dinner before its owner finishes asking.

Manners are less cinematic. Manners mean not reading over a shoulder merely because the sensor can. They mean knowing that a calendar entry is not permission to email everyone attending it. They mean asking before spending, forgetting when asked to forget and understanding that silence can be the correct completion of a task.

Qualcomm’s new flagship phone chips are explicitly built for “agentic AI”: systems that perceive context and take sequences of actions. The hardware achievement is impressive. The cultural danger is that phone makers will translate agency into chumminess because a synthetic friend is easier to market than a meticulous subordinate.

We should resist. The phone is already the least bounded object in modern life. It sits beside the bed, enters the bathroom, hears the argument, maps the commute and holds the messages we did not send. Giving it initiative without first designing etiquette would create an intimate who cannot be embarrassed.

Good agent design should feel less like a new companion and more like excellent stage management. It prepares the necessary object, stays out of the scene and can explain every intervention afterward. Its memory should be visible. Its assumptions should be reversible. Its most elegant behavior should often be restraint.

This is partly a design complaint and partly a taste complaint. A machine constantly announcing its helpfulness is vulgar. A device that acts without establishing the owner’s boundaries is worse. The industry has enough compute to put a character in every pocket. Before it does, it should learn the ancient social technology of knocking.

The falling price of machine intelligence will be celebrated as democratization, and in one sense it is. OpenAI’s new Luna model lists input at ten cents per million tokens. Textual abundance is becoming too cheap to notice.

Abundance, however, does not abolish scarcity. It moves it. If a company can generate a thousand analyses for less than the price of lunch, the valuable act is choosing which analysis governs. If software can draft every conceivable reply, value resides in deciding which message deserves to be sent. If an agent can take action continuously, refusal becomes a premium function.

This is uncomfortable for organizations that spent the first AI wave counting output. They measured documents summarized, tickets closed and lines of code proposed because production was the easiest thing to see. Cheap models will make those figures swell until they become meaningless. A million generated words are not evidence of a million words’ worth of thought.

The durable institutions will charge for judgment even when the underlying cognition is nearly free. They will know which tasks require a human signature, which errors are tolerable and which apparent efficiencies merely transfer risk to customers. They will preserve records of who authorized an action rather than allowing “the model did it” to become corporate passive voice.

This applies beyond business. Newspapers can produce more summaries; readers will return for selection and nerve. Universities can produce more explanations; students will still need standards for proof. Governments can find more anomalies in databases; legitimacy will depend on how they distinguish fraud from a person who does not resemble the average record.

Cheap intelligence is not the end of expertise. It is the end of confusing fluency with expertise. The machine can fill the page for almost nothing. Someone must still put a name beneath it.