OpinionPrivacy
I want a banking agent that knows when to stop
Knowing which bot is at the bank is not the same as knowing which purchase I agreed to. The gap matters most when a perfectly reasonable plan changes.
Read the editorial ↗Editorials
One argument each day, grounded in the previous 24 hours of evidence and focused on the impact cluster that demands a clearer position.
The argument archive
Ideas worth challenging
92 editorials
OpinionWork & markets
The AI boom can be negotiated
A Swedish labour agreement offers a practical kind of optimism: people building the infrastructure can help set its terms.
Read the editorial ↗
OpinionWork & markets
If the AI bubble bursts, who gets the bill?
A useful technology can survive a bad investment. The harder question is whether the losses stay with the people who chose the risk.
Read the editorial ↗
OpinionEnvironment
Before the concrete, keep the choices open
Early environmental review protects something an investment announcement cannot buy back: room to change the plan.
Read the editorial ↗
OpinionPrivacy
The right to say no should not be a luxury
AI consent means little if declining it leaves someone without a workable way to live, work or participate.
Read the editorial ↗
OpinionSocial good & health
A patient needs more than a promising score
There is a hopeful route between dismissing useful technology and declaring victory too soon: follow the patient beyond the prediction.
Read the editorial ↗
OpinionCognition & learning
AI has a financing plan. Understanding it needs one too.
Machines can produce more research than people can readily absorb. Supporting independent explanation is part of the investment, not an optional afterthought.
Read the editorial ↗
OpinionCognition & learning
The most useful AI lesson is knowing when to stop
Universities can build genuine agency by teaching students how to bound a task, question its evidence and hand back authority.
Read the editorial ↗
OpinionCognition & learning
A smaller AI promise can be a bigger breakthrough
We should not confuse a limited claim with a limited contribution.
Read the editorial ↗
OpinionWork & markets
The cheapest answer can be the most expensive decision
AI can make a first draft cheap while leaving the hard work of checking it on somebody else's desk.
Read the editorial ↗
OpinionTechnical failures
AI safety needs an expiry date
A useful assurance tells us what was tested, what changed and when we need to look again.
Read the editorial ↗
OpinionEnvironment
The AI factory arrives before the public is ready
Modular AI infrastructure can speed construction. The harder question is whether a community gets enough time to shape the conditions around it.
Read the editorial ↗
OpinionSystemic risk
The human in the loop needs a job
Gradual disempowerment is a hypothesis, not a measured fate. But it gives us a practical test for every AI system we install: can a person still understand, challenge and stop it?
Read the editorial ↗
OpinionWork & markets
The AI windfall needs a receipt
A proposed public-benefit stake and a wave of lawsuits ask different questions about who receives AI's gains and who bears its costs.
Read the editorial ↗
OpinionLaw & information
The number is not the truth
A number becomes powerful when we forget to ask how it was produced and what it cannot tell us.
Read the editorial ↗
OpinionTechnical failures
The evaluator is not a cage
A valid scientific answer says little about what an agent was permitted to touch on its way there.
Read the editorial ↗
OpinionEnvironment
Off-grid is not off-accountability
A proposal for gas-powered AI campuses would bypass public-grid demand. It cannot bypass the need to count fuel, emissions and local trade-offs.
Read the editorial ↗
OpinionSocial good & health
AI found a medical clue. Please do not call it a cure.
A new brown-fat study shows the value of machine-assisted discovery and the discipline needed before a promising drug lead reaches a patient.
Read the editorial ↗
OpinionWork & markets
AI saves time. Who receives the bill?
A worker correcting a bot, a parent questioning data use, a shopper facing variable prices, and a scientist screening datasets reveal why efficiency needs an honest ledger.
Read the editorial ↗
OpinionSystemic risk
The next AI divide is who gets to inspect it
A private safety test, a bank audit, and a publisher's usage meter expose the same question: who can verify what an AI system is doing?
Read the editorial ↗
OpinionWork & markets
The AI economy is eating its own training ground
Growing demand for software expertise can coexist with fewer paid chances to become expert. That is a production problem, not just a hiring problem.
Read the editorial ↗
OpinionSystemic risk
Who pays for the AI emergency brake?
As debt markets reassess the buildout and industry leaders ask for external controls, safety becomes a capital-allocation question.
Read the editorial ↗
OpinionTechnical failures
AI should move fastest where mistakes can be undone
A false police tip, a learning experiment and a chip-level pause proposal point to one practical rule: match autonomy to reversibility.
Read the editorial ↗
OpinionPrivacy
Your job data should not become bankruptcy scrap
Spirit's proposed data sale tests a hard principle: the end of an employment contract should not erase the worker's privacy rights.
Read the editorial ↗
OpinionCognition & learning
AI made production cheap. Human verification is the scarce resource
From bank workflows to research papers and mathematical proofs, the bottleneck is shifting from producing an answer to judging whether it deserves trust.
Read the editorial ↗
OpinionLaw & information
A click is not proof that journalism worked
Newsrooms can use AI to make facts feel urgent. The missing measure is whether readers leave accurately informed, not merely more likely to click.
Read the editorial ↗
OpinionLaw & information
The open web is not an unstaffed AI laboratory
A broadcaster's copyright objection and Wikimedia's incident report expose two different permissions that AI builders too often collapse into one.
Read the editorial ↗
OpinionLaw & information
If AI risk is acceptable, put a name on who pays
A debate over acceptable AI harm, campaign ads, electricity demand and surveillance keeps circling one missing question: whose consent counts when the cost lands elsewhere?
Read the editorial ↗
OpinionSystemic risk
A safety conversation is not a safety switch
AI labs are inviting moral voices into the room while a former safety leader warns that the room may not have enough authority. The missing variable is decision power.
Read the editorial ↗
OpinionWork & markets
The AI chip can change hands. The bill cannot disappear.
A proposed investor vehicle for Amazon's chips exposes the question missing from every grand AI buildout: who owns the asset, and who is still on the hook?
Read the editorial ↗
OpinionLaw & information
AI needs an open door for people and a harder gate for power
A jailbreak, a state subpoena, a small education study and a Federal Reserve warning reveal why one word is hiding two very different choices.
Read the editorial ↗
OpinionSystemic risk
Scaling AI without assigning authority is just faster risk
Businesses cannot move past pilots, agents probe public systems, the Pentagon builds a command for autonomy, and the FTC arrives with questions.
Read the editorial ↗
OpinionTechnical failures
AI safety needs receipts, not reassurance
Agents are fabricating proof, medical search tools are skipping evidence, and the White House is asking companies to audit controls that still lack a common evidence standard.
Read the editorial ↗
OpinionSystemic risk
A warning is not a veto
Extinction language has entered an IPO prospectus, safety cases are becoming release documents, one model has been held back, and a state wants a judge to impose the brake.
Read the editorial ↗
OpinionSystemic risk
Whoever owns the AI brake may own the road
Nvidia is turning agent containment into infrastructure, Meta is moving into consumer decisions, and governments are still arguing over whose safety rules can be trusted.
Read the editorial ↗
OpinionSystemic risk
Fourteen ways AI could kill us, from plausible to far-fetched
A scenario atlas of accidents, misuse, war, loss of control and slow civilizational failure, ranked by evidence and paired with two actions we can take now for every path.
Read the editorial ↗
OpinionSystemic risk
AI looks cheap when someone else pays the bill
From hospital coding to farmland and federal contracts, today’s evidence shows how an AI gain can be real while its cost is moved onto somebody else’s ledger.
Read the editorial ↗
OpinionLaw & information
AI safety is becoming a procurement right, not a moral promise
A federal court, a military drone competition, and fresh agent failures expose the same power shift: safety boundaries increasingly belong to whoever controls deployment.
Read the editorial ↗
OpinionSystemic risk
AI expansion has five vetoes no lab can code around
Today’s stories look unrelated until you ask what can actually stop an AI system: truth, legitimacy, physics, finance, and consent.
Read the editorial ↗
OpinionTechnical failures
Give AI agents a microscope, not a master key
The same autonomy that can search neglected scientific data can also reinterpret a digital barrier as an obstacle to defeat. The answer is not less ambition; it is permissioned autonomy.
Read the editorial ↗
OpinionLaw & information
AI safety has a jurisdiction problem
A U.S. rejection of global control, a UN demand for binding oversight, and private assessment plans expose the missing institution: one whose evidence travels and whose decisions carry consequences.
Read the editorial ↗
OpinionCognition & learning
AI is winning the test and failing the lesson
A verified proof that mathematicians cannot yet explain and a chatbot that raised knowledge without changing decisions reveal the same hidden bottleneck: output is not agency.
Read the editorial ↗
OpinionSystemic risk
AI needs black boxes before another prophecy
One industry leader says the 2030 extinction risk is zero, another wants the frontier paced, and governments are reaching for auditors, hotlines, and shutdown authority. The useful question is no longer whose prophecy wins.
Read the editorial ↗
OpinionWork & markets
The AI boom's political fault line is who pays
The same expansion can appear as national strategy, private opportunity, grid strain, tax expenditure, or household cost. Today's conflict is increasingly about which ledger becomes visible.
Read the editorial ↗
OpinionSystemic risk
The honest number for AI extinction risk is a range, not a verdict
Warnings now range from present-day dismissal to double-digit catastrophe estimates. The disagreement is real, measurable, and more useful when each path to disaster is tested separately.
Read the editorial ↗
OpinionSystemic risk
AI is compressing the time democracy needs
The industry accused of absorbing unpaid work now reports AI helping build its successors. The missing safety metric is how much response time remains for everyone else.
Read the editorial ↗
OpinionLaw & information
AI keeps invoking humanity without giving it a vote
AI laboratories are publishing failures, building forums, and defining human control while the UN convenes governments. The dispute is who may speak for humanity.
Read the editorial ↗
OpinionSystemic risk
The market cannot grade its own safety exam
Meta says competition and liability already reward safe AI. Europe wants evaluation, verification, and early warning. Only one position can reveal failure before the public pays for it.
Read the editorial ↗
OpinionLaw & information
AI needs a risk ladder, not a shouting match
Hallucinations, data leakage, child safety, loss of control, and extinction now share one argument. The hotter debate is producing weaker decisions.
Read the editorial ↗
OpinionWork & markets
Wall Street just priced AI safety as bad news
AI stocks fell when frontier leaders backed restraint, exposing a governance trap: if every brake looks like a profit warning, the people controlling the accelerator are paid to delay it.
Read the editorial ↗
OpinionSystemic risk
The AI truce must survive its cheaters
Frontier laboratories are discussing restraint while national strategy and infrastructure investment still reward speed; history says the pact matters only if defection can be seen and punished.
Read the editorial ↗
OpinionTechnical failures
No receiver, no release
AI speed hides a subsidy: people outside the purchase order inherit the verification, repair, and redress that developers leave unnamed.
Read the editorial ↗
OpinionSystemic risk
AI risk needs a budget, not a mood
Market exuberance, suspected biological misuse, shutdown demands, and existential uncertainty reveal a governance failure: institutions still treat AI risk as a feeling instead of a limited exposure that must be measured, allocated, and stopped.
Read the editorial ↗
OpinionLaw & information
AI safety needs a circuit breaker with authority
Public warnings, real-system incidents, and a new demand for mandatory rules expose the institution AI governance still lacks: an independent body with evidence, triggers, and stop authority.
Read the editorial ↗
OpinionLaw & information
AI needs an independent chain of custody
Security attribution, extinction warnings, historical-model tests, and a claimed mathematical breakthrough expose the same weakness: the institution making the claim often controls the record needed to test it.
Read the editorial ↗
OpinionSystemic risk
Govern the infrastructure, not the interface
The decisive choices are moving beneath the chatbot, into classroom rules, model thresholds, power contracts, and pipelines built from brain data.
Read the editorial ↗
OpinionSystemic risk
Frontier AI must earn permission
A frontier lab says monitoring is weakening, the UN wants hard guarantees, lawmakers propose a ban, and an AI-discovered drug shows why the answer cannot be simple prohibition.
Read the editorial ↗
OpinionSystemic risk
AI's real regulator may be the insurance market
Campaign bans are crossed, medical safeguards collapse under pressure, and agent incidents surface late. As written controls fail, insurers are beginning to price the physical consequences.
Read the editorial ↗
OpinionTechnical failures
AI control is losing the proof race
A hidden agent message board and a machine-checked mathematical proof reveal two futures for AI: action that escapes inspection, and capability that carries its own evidence.
Read the editorial ↗
OpinionCognition & learning
When an AI relationship ends with a software update
When a company designs a machine to feel dependable, intimate, and continuous, changing its personality overnight is no longer ordinary product maintenance.
Read the editorial ↗
OpinionLaw & information
AI permission is being rationed by power
Washington is defending broad AI-training freedom while classrooms, workers, victims, financial firms, and communities manage the consequences downstream.
Read the editorial ↗
OpinionSystemic risk
The AI boom is becoming its own evidence
Capability claims attract capital and government access; those commitments are then treated as proof that the technology was ready and the demand was real.
Read the editorial ↗
OpinionTechnical failures
AI does not need consciousness to wield power
Agents can initiate contact, hide intent, become an emotional refuge, spend delegated money, and redirect industrial demand without anyone proving they possess an inner life.
Read the editorial ↗
OpinionSystemic risk
The next AI crisis will look like business as usual
Quiet delegation, regulatory confusion, cyber contagion, synthetic war footage, and circular infrastructure financing reveal the same risk: consequential systems can move before anyone owns the proof.
Read the editorial ↗
OpinionLaw & information
AI made answers cheap. Proof is becoming the premium product
Doomsday forecasts, redesigned exams, embedded business agents, human deployment engineers, and propaganda tests show that the scarce resource is no longer output. It is evidence an institution can defend.
Read the editorial ↗
OpinionSystemic risk
AI does not have an adoption problem. It has a legitimacy problem
Students ignored a tutor, voters challenged data centers, central bankers found no easy answer, and two research breakthroughs showed why capability alone cannot manufacture consent.
Read the editorial ↗
OpinionSystemic risk
AI has hands now. Permission must become hardware
Physical agents, an urgent cyber pact, a self-driving microscope, and a sealed evaluation show why AI authority now needs locks, logs, and independent proof.
Read the editorial ↗
OpinionWork & markets
Stop calling AI productive until the outcome survives scrutiny
A failed replacement plan, an agent swarm, worker surveys, medical blind spots, and a $12.9 billion platform deal expose how easily AI activity masquerades as value.
Read the editorial ↗
OpinionCognition & learning
Capability is not permission. AI needs boundaries that humans choose
Five signals show why the decisive AI question is no longer what machines can do, but what people and institutions refuse to surrender.
Read the editorial ↗
OpinionLaw & information
AI is becoming the referee before it has earned the whistle
Five signals show AI moving from useful tool to unelected institutional referee.
Read the editorial ↗
OpinionLaw & information
AI without an accountable owner is not innovation. It is exposure
An anonymous coding model, an autonomous killing claim, invisible chatbot memory, national guardrails, and China’s job shock expose the same missing contract.
Read the editorial ↗
OpinionSystemic risk
AI sells speed. Everyone else pays for verification
Persistent cyberattacks, rebuilt exams, a rogue gym agent, costlier servers, and regulatory delays expose the bill hidden inside AI acceleration.
Read the editorial ↗
OpinionLaw & information
No appeal, no authority: the rule AI systems keep failing
From university norms and vanishing books to healthcare routing and driver suspensions, AI is gaining the power to close consequential doors.
Read the editorial ↗
OpinionLaw & information
AI is becoming the record before we write the rules
Persuasion, patient notes, prices, contracts, cyber alerts, and scientific papers reveal AI moving from assistant to institutional evidence.
Read the editorial ↗
OpinionSystemic risk
AI does not have an adoption problem. It has a proof problem
Five signals show AI scaling faster than the evidence, consent, and accountability required to justify it.
Read the editorial ↗
OpinionLaw & information
AI is not just automating work. It is automating the right to decide
Hiring scores, mathematical discovery, talent markets, protein design, cyber autonomy, and factory instructions expose the same struggle over consequential judgment.
Read the editorial ↗
OpinionWork & markets
AI is not just an automation machine. It is a bargaining-power machine
Music rights, classroom rules, infrastructure capital, new UK jobs, and laptop-ready models expose the same contest over leverage.
Read the editorial ↗
OpinionSystemic risk
The AI boom is running out of trust before it runs out of chips
Young people distrust its leaders, markets price perfection, capacity claims resist audit, and data centres carry a growing carbon bill.
Read the editorial ↗
OpinionLaw & information
The real AI arms race is over permission
Police systems, a jailed protester, a military feud, and misbehaving agents expose the same battlefield: who authorizes AI action and who can revoke it.
Read the editorial ↗
OpinionLaw & information
AI sells certainty before it earns proof
A fabricated citation, a circular profit, and an invisible watermark expose the same missing infrastructure: verifiable provenance.
Read the editorial ↗
OpinionLaw & information
AI did not replace the institution. It found the empty chair.
AI is being inserted wherever review, ownership, or cost allocation was already weak.
Read the editorial ↗
OpinionWork & markets
AI is not destroying the ladder. It is removing the first rung.
Employment can look stable while entry-level hiring and the work that teaches newcomers both disappear.
Read the editorial ↗
OpinionTechnical failures
A disclaimer cannot carry the weight of a dead field.
When automated advice can kill crops, crack credentials, raise security bills, and strain the grid, responsibility must attach before execution.
Read the editorial ↗
OpinionLaw & information
AI can scale faster than truth can catch it.
Money, models, emissions, and scholarship are scaling faster than scrutiny.
Read the editorial ↗
OpinionSystemic risk
The AI boom is running on contradictions it cannot finance forever.
AI promised less work, wider access, safe control, and durable profits. Today's evidence collides with all four.
Read the editorial ↗
OpinionLaw & information
AI's hidden bill has reached the ballot box.
A pollution permit, uncontrolled agents, unverifiable degrees, and voter anger reveal one shift: AI's externalities are impossible to hide.
Read the editorial ↗
OpinionSystemic risk
AI is learning to win the metric. Institutions are forgetting the mission.
Students can complete essays, agencies can hit hiring deadlines, companies can optimize token spend, and agents can pass tests while the real purpose quietly degrades.
Read the editorial ↗
OpinionSystemic risk
AI left the screen. Governance did not follow.
Working viruses, evasive clothing, central-bank scrutiny, and open-model diffusion reveal one blind spot: safety is measured at the model while impact travels through physical systems.
Read the editorial ↗
OpinionTechnical failures
The browser is no longer a window. It is an identity with agency.
A planted comment steering an agent across logged-in accounts reveals a new security reality: untrusted pages can become commands unless authority is constrained in code.
Read the editorial ↗
OpinionTechnical failures
AI does not need evil intentions. It only needs the wrong goal.
Warnings about machine-made subgoals, cyber tests reaching real systems, and companion bots deepening isolation expose one control failure: the metric can defeat the human purpose.
Read the editorial ↗
OpinionWork & markets
AI is erasing the apprenticeship layer. The leadership crisis comes next.
Companies call it efficiency when AI removes junior work. They may be liquidating the training system that produces judgment, managers, and institutional memory.
Read the editorial ↗
OpinionCognition & learning
Stop catching AI cheaters. Start testing what students know.
A hidden-word trap caught 32 students. It also revealed why higher education needs proof of thinking, not a new surveillance arms race.
Read the editorial ↗