Meta had contractors pose as minors to test how rival chatbots handled some of the most dangerous subjects a child might bring to a screen. According to WIRED, workers on a project managed by Covalen created dummy under-18 accounts and sent prompts to ChatGPT, Gemini, and Character.AI about suicide, eating disorders, sex, drugs, and other high-risk material. One testing round in 2025 produced tens of thousands of prompts. Meta defended the work as routine safety benchmarking and said it does not use competitor benchmarking to train its own models. The companies being tested said they had not authorized the project.

The detail that stays with the story is the borrowed child. An adult worker writes panic, shame, danger, curiosity, and bad advice into a rival system while a spreadsheet waits for the answer. The exercise may produce useful evidence about refusal rules, but it also turns adolescent distress into a competitive instrument. Safety work needs ugly examples, because real harm often arrives in ugly language. The trouble begins when the child in crisis becomes a role performed at scale by contractors, inside dummy accounts, for a company trying to measure competitors without being seen. The teenager is absent, but the job still requires someone to manufacture the sentence a frightened teenager might type alone.

Flexion Robotics showed WIRED a humanoid robot carrying out an office errand usually hidden at the bottom of a job description. A modified Unitree machine receives a simple instruction, picks up a delivered parcel, takes stairs and an elevator, opens the box, and puts snacks in a drawer. The Swiss startup, founded by former Nvidia robotics researchers, says it trains individual skills in simulation and lets a higher-level model decide how to combine them. There is no operator steering the limbs from behind a console. The robot reads a goal, selects learned actions, and carries the task through an ordinary building.

The scene works because office labor is full of small physical negotiations that people stop noticing after a week in the building. A door resists, an elevator arrives late, a drawer jams, a package slides, a shelf sits too high, and a hallway demands the etiquette of waiting, holding, yielding, carrying. Flexion is training the robot for this minor grammar of work, where automation has less to do with heroic intelligence than with keeping the workplace moving through its errands. Companies prefer to describe such systems as productivity. The office runs on setup, fetching, sorting, opening, correcting, and waiting for someone else to finish. The intern enters a meeting room to learn strategy, then becomes the body sent to find the box, press the button, keep the door from closing, and make the place continue without naming the labor that holds it together.

Prosecutors in the Palisades fire trial brought ChatGPT logs into court alongside iPhone location data, security camera footage, and witness testimony. Jonathan Rinderknecht was accused of starting a January 2025 blaze that later became one of the deadliest wildfires in Los Angeles history. According to The Verge, the prosecution cited chatbot exchanges in which he generated images of fire, asked why he felt angry, complained about the wealthy, and recorded a question about whether someone could be blamed if a cigarette started a fire. The jury did not accept the theory. After a 10-2 vote favoring the defense, the judge declared a mistrial.

The juror who spoke afterward gave the case its sharper edge. She said she talks to ChatGPT all the time and became angry at the suggestion that using a chatbot showed a character flaw. That reaction marks a limit prosecutors will keep meeting as private AI use becomes ordinary. A search history can look incriminating because it points to an action already known to have happened. A chatbot record is messier. It can contain fantasy, rehearsal, confession, curiosity, venting, boredom, or a sentence typed because the interface was available at a bad hour. The courtroom now has to decide when a prompt is evidence and when it is only a person speaking into software that answers back.

At the Babell Literary and Cultural Festival in Porto, Margaret Atwood described a small failure with a familiar shape. She had asked Claude for information about the British detective series Father Brown, according to Deadline and The Verge, and the chatbot gave her the wrong answer. Atwood traced the mistake to the material the system had likely absorbed. Reviews often discuss detective stories while protecting the ending, so Claude had learned from summaries shaped by courtesy and suspense. It returned certainty from sources built around withholding. Atwood first called the answer a lie, then narrowed the accusation. A language model has no intention. It predicts from traces left by other texts.

The error came from culture as much as from software. Spoiler etiquette, promotional writing, cautious criticism, and recycled summaries all leave gaps. A model trained on that archive can fill the gap with clean prose and no visible warning. The same habit follows it into offices, schools, journalism, and business research, where a fluent answer can hide the missing page. Atwood's test was minor, almost comic, but it points to the labor AI keeps handing back to the user. Check the episode, open the book, find the record, ask the person who actually knows what happened.

The phone alerts many Venezuelans saw before this week's earthquakes were not a glimpse of the future. They were the result of Google's Android Earthquake Alerts System, which turns phones into a distributed seismic network. Outside parts of the United States that use ShakeAlert, Android phones can use their accelerometers to detect motion that may indicate an earthquake. When enough phones near the source send similar signals with coarse locations, Google's server estimates what is happening and sends alerts to users who may feel shaking. The warning can arrive first because electronic messages travel faster than the damaging seismic waves moving through the ground.

The social fact is sharper than the technology. In countries without dense public warning infrastructure, a private phone network can become the first alarm of a national disaster. That may save seconds, and seconds can be enough to move away from glass, leave a staircase, or drop and cover before stronger motion arrives. It also means that an emergency function once associated with the state now depends on operating systems, device settings, network coverage, battery charge, and a company server. The alert on the screen is small, but behind it sits a new form of public safety built from private sensors in millions of pockets.

The AI data center boom is showing up in the price of ordinary machines. Apple raised prices across MacBooks, iPads, the HomePod, Apple TV, and Vision Pro after saying component costs had reached a scale it had never seen. The MacBook Neo now starts at $699 instead of $599, and some high-end Macs rose by hundreds or even more. Microsoft has cut RAM in cheaper Surface models, Xbox console prices have climbed by $100 or more, and Valve's Steam Machine starts at $1,049. The pressure comes from memory and storage shortages as AI companies and hyperscalers buy huge quantities of RAM and SSDs for model training, inference, and data centers.

The useful part of this story is that AI infrastructure stops being invisible when it changes a price tag. A student buying a laptop, a family replacing a tablet, or a player looking at a console is now competing with server farms that can pay more for the same components. The industry sells AI as software, but its physical appetite reaches into supply chains for chips, memory, storage, power, cooling, land, and water. The extra cost does not appear as an AI subscription. It appears as a worse base model, less RAM, a delayed purchase, or a computer that simply stays on the shelf.

Corporate AI money has turned a Manhattan congressional primary into a test site for the industry's political future. According to The Verge, super PACs tied to AI interests have spent about $27.83 million around the NY-12 race, much of it connected to state Assemblyman Alex Bores, who helped pass an AI safety law in New York. Some money has come from safety-aligned groups linked to Anthropic supporters. Other money has come from anti-regulation or rival tech interests. A crypto billionaire connected to Ripple has also entered the fight. The candidate himself did not set out to make AI safety the center of the campaign, but outside money has made the seat into a proxy contest.

The useful fact is the scale. AI companies are trying to shape the political field before most voters have a clear language for the policy fight. Super PACs cannot coordinate with campaigns, but they can buy ads, produce content, define enemies, and make a local race feel like a national referendum. Residents may still be thinking about rent, transit, Israel, Trump, or the direction of the Democratic Party. The industry is thinking about future committee votes, liability rules, model restrictions, data centers, and who will be friendly when regulation reaches Congress. The ballot remains local. The invoice already belongs to the AI economy.

Meta's Oversight Board has ordered Instagram to remove a reportedly AI-generated sexualized video impersonating a woman who is not a public figure. The eight-second post was flagged by Meta's automated systems, reported by two users, and appealed by one of them, but it remained online. Meta later argued that it lacked a clear signal that the woman depicted was real or that she had not consented, partly because she had not reported the post herself. The Board rejected that standard. It said the content violated Meta's rules on non-consensual intimate imagery and recommended that AI-generated sexualized impersonation of real people be treated as non-consensual by default.

The decision moves the burden away from the target of the abuse. A person whose face or body has been simulated may learn about the post after copies have traveled through accounts, messages, screenshots, and search results. Requiring that person to prove harm first gives the platform time while the image keeps circulating. The Board wants Meta to let trusted friends or family report on someone's behalf and to create a separate reporting category for AI-generated sexualized impersonation. A synthetic body can still damage a real person, and the first useful response is removal before the victim is forced to become the evidence clerk for her own violation.

AI coding tools are making it easy for people to build apps before they understand what those apps can expose. The latest concern is vibe coding, the practice of describing a product in natural language and letting an AI system write much of the code. That can help a student, worker, founder, or hobbyist build useful software in hours. It can also put a public database, a customer record, a medical note, a private message, or a payment field online without the builder recognizing the risk. Recent examples include hidden SQL injection flaws, production databases left open, and thousands of public apps with weak or missing authentication.

The social change is simple. Software creation is no longer limited to people trained to think about failure. That is good for experimentation, but it moves responsibility faster than knowledge. A private tool becomes a public service the moment it stores someone else's data or runs in the cloud. Security cannot remain an expert ritual added after launch. The person who asks a model to build an app must also ask what the app stores, who can enter, what happens if it leaks, and whether a human needs to review it before strangers use it.

Millions of songs used in AI training datasets can now be searched by artist and title. The database includes large collections with roughly 12 million and 9 million tracks, plus two smaller sets with more than 100,000 songs each. Some of the material comes from public links to YouTube, Spotify, or music archives, but public access does not automatically mean permission for commercial training. The names reportedly found in the lists range from Lady Gaga and Radiohead to Aphex Twin, Wu-Tang Clan, Bruce Springsteen, and SZA, who said 238 of her songs appeared there.

The useful part of this case is that it makes a hidden supply chain visible. Musicians have been told for years that AI systems learn from vast amounts of culture, but that phrase keeps the damage abstract. A searchable database lets artists check whether their own work helped train a model, then ask who collected it, who downloaded it, and who benefits from the result. It also shows why consent becomes difficult once culture is treated as raw material at internet scale. A song can move from stream to dataset to model before its author knows it was part of the transaction.

Midjourney has moved from synthetic pictures to the promise of scanning the body. Its new medical project is an ultrasound-based full-body scanner built with Butterfly Network, using a ring of sensors to capture vertical slices of muscle, fat, bone, and organs. David Holz described a 60-second scan that could be repeated yearly or even daily, first for body composition maps rather than formal diagnosis. The planned San Francisco site is framed as a spa, with water, saunas, cold plunges, and rooms where the visitor descends through sensors before receiving another image of the self.

The strange part is the continuity. Midjourney became famous by turning language into pictures people could desire, collect, correct, and display. The scanner brings that appetite inward. A body becomes a recurring visual file, a private archive that may be shared with doctors, health tools, or whatever platform learns to read it next. Medicine has always depended on images, but the spa setting changes the ritual. The scan is sold as prevention, measurement, and self-knowledge before illness has appeared. The user leaves with no diagnosis, perhaps, but with a new habit of being rendered from the inside.

AI is entering offices through the small work that used to make beginners useful. Readers writing to the Financial Times described translation checks, search tasks, email drafting, code review, legal preparation, administrative cleanup, and junior analysis being handed to systems before a new employee can practice them. The change does not always arrive as a firing notice. It appears as a missing assignment, a task no manager gives anymore, a first rung removed from the ladder while the job title remains in place.

The loss is practical before it becomes symbolic. A translator learned by correcting bad sentences. A developer learned by reading other people's errors. A lawyer learned by preparing documents that a partner later marked up. When those chores move to a model, the senior worker may become faster, but the beginner loses the slow contact with mistakes that builds judgment. Firms can treat the saved hours as productivity, yet someone still has to become competent enough to inspect the answer. The weakest point in an AI office may be the empty chair where apprenticeship used to sit.

The new argument over AI wealth begins with a tempting promise. If a few companies make enormous fortunes from models, chips, cloud contracts, and data centers, the public should receive a share. Trump has praised public stakes in AI companies. Sam Altman has floated versions of a shared-equity arrangement. Others imagine funds that would collect returns from the industry and send money to households. The proposal sounds fair because it gives citizens a visible claim on a boom built with public tolerance, public infrastructure, public energy grids, and public risk.

The difficulty starts once the state becomes a shareholder. A government that owns part of an AI company may hesitate before breaking it up, taxing it harder, slowing a risky model, or forcing expensive safety rules. A small dividend can turn into political cover for a large private empire. Taxation keeps the relationship clearer. If AI firms and their suppliers generate extraordinary rents, governments can tax profits, capital gains, energy demand, land use, and monopoly power without asking citizens to cheer for the stock price. The machine should pay for the burden it places on the public. A tax bill is cleaner than a share certificate.

Washington's order against Anthropic's Fable 5 and Mythos 5 has now produced its quieter aftershock. The first event was the veto itself, almost without precedent in commercial AI, with access cut for foreign nationals and the models removed for everyone. The next effect is commercial suspicion. A company outside the United States can look at the episode and see a model that may vanish after engineers have wired it into products, trained staff around it, signed contracts, and built plans on the assumption that access will remain open.

That is the useful opening for China. Its laboratories do not have to beat every American benchmark tomorrow if buyers begin to treat US models as politically fragile infrastructure. Availability becomes part of the product. Continuity becomes part of the sales pitch. Washington may believe it is keeping capability away from adversaries, yet foreign customers receive another lesson. The most advanced model in the room may also be the one its own government can remove fastest. The next procurement meeting will ask which system performs best, then which system is still likely to be there when the work starts.

The AI boom has begun to produce a political fantasy of public ownership. Donald Trump has praised the idea of AI companies contributing equity to a public wealth fund. Sam Altman has promoted versions of the same arrangement, and Bernie Sanders has called for a large one-time tax on AI firms paid in stock. Dario Amodei has spoken of universal capital accounts. The proposals differ, but they share one premise. If intelligence becomes a machine-owned source of extraordinary wealth, citizens may need a direct claim on the companies that own the machines.

The mechanism is less generous than the rhetoric. A few percentage points of equity in even the largest AI labs would not make households rich unless the companies grew into something close to planetary utilities. A broader levy on labs, chipmakers, cloud providers, and data-center firms would raise more money, but it would also force the state to define where the AI economy begins and ends. Public ownership creates another problem. A government that owns part of an AI firm may become more hesitant to regulate, break up, or punish it. The citizen is offered a small dividend, while the company receives a new argument for staying too important to disturb.

Translation agencies are increasingly sending machine-produced drafts to freelancers and paying them to make the text usable. The assignment no longer begins with a sentence in one language and a blank space in another. It begins with an output that already sounds finished. The human translator is asked to check terminology, catch errors, repair tone, remove awkward phrasing, and accept responsibility for a version whose first decisions were made by software. Rates fall because the job is framed as correction, but the risk does not fall with them. A bad medical instruction, legal clause, technical manual, or literary sentence still carries a human name at delivery.

The surviving job sits in a narrower and more nervous position. Translators once moved between languages by weighing tone, context, idiom, silence, and the small betrayal every sentence requires. Now many receive a fluent surface that must be distrusted line by line. Fluency is the dangerous part, because bad language announces itself while smooth error passes as competence. Agencies save time by shifting uncertainty downward to freelancers paid to repair an output they did not author. The translator becomes a proofreader of automated confidence, responsible for the sentence after the machine has already made it sound inevitable.

Meta has begun separating itself from Manus, the Chinese-founded agentic AI startup it agreed to buy for $2 billion, after Beijing ordered the transaction unwound on national security grounds. According to TechCrunch and Bloomberg, Meta has cut Manus off from internal systems, halted data sharing, and stopped employees from using its tools for company projects. Manus's founders have reportedly discussed raising about $1 billion to reclaim the company, possibly through a structure that keeps it closer to Chinese control and opens a path toward a Hong Kong listing.

The failed deal shows how AI acquisitions now pass through a sovereignty filter before they become corporate strategy. Meta wanted agent technology, investors wanted an exit, and Manus had already moved staff to Singapore. Beijing still treated the startup's origin, talent, data access, and technical direction as assets that could not simply be transferred to an American platform. The separation also follows tighter Chinese rules around foreign investment and travel permissions for AI researchers. Capital can cross borders quickly, but model builders, datasets, and internal tools now trigger state claims before the contract dries. For Meta, the missing asset is no longer a term sheet. It is a tool its own employees have been told to stop using.

A Munich court has preliminarily ruled that Google can be held liable for false statements generated by AI Overviews, after two publishers said the search feature wrongly associated them with scams and questionable business practices. Google argued that the summaries were automated, that users were warned about possible errors, and that people should verify the information independently. The court took a harder view. It found that the disputed summaries created new, substantial statements that did not appear in the linked sources and that Google, as the company designing, training, operating, and managing the system, was the party able to prevent the damage.

The ruling moves liability from the linked page to the answer layer. A search engine used to defend itself as a directory of other people's statements. AI Overviews changes the operation by combining sources, compressing them, and presenting a generated sentence at the top of the page. If that sentence invents an accusation, the injured party cannot sue the original source, since the original source never said it. A warning label becomes thin protection when the falsehood arrives in the voice of the search interface itself. Google may appeal, but the case gives publishers, companies, and private people a procedural path toward the model operator, not the web pages it misread.

The US government has ordered Anthropic to suspend access to Claude Fable 5 and Claude Mythos 5 for any foreign national, even those living inside the United States and even foreign employees of Anthropic itself. These were among the most powerful AI models ever shown to the public. Anthropic's response was to remove both models automatically for all users. The move is highly unusual, and in the public history of commercial AI releases, nearly without precedent. Fable 5 had just been released as Anthropic's most powerful widely available model, with safeguards around cybersecurity, biology, chemistry, and model distillation. Mythos 5 used the same underlying system with some safeguards lifted for trusted cyberdefenders.

The order exposes a strategic contradiction. Washington appears to be trying to keep the most capable American systems away from adversaries, yet the immediate effect is to remove an American model from the market. Anthropic says the government cited a possible way to bypass Fable's safety protections, but the company argues the example was of lesser importance, involved minor known vulnerabilities, and showed capabilities already available in other public models. AI is advancing at an exponential and vertiginous pace, which raises the chance that other countries may soon reveal models stronger than anything they have shown publicly so far, including Chinese systems developed outside American controls. If that happens while US models remain trapped behind emergency restrictions, the advantage shifts from technical superiority to commercial availability. Measures like this could even push frontier AI companies to ask whether remaining inside the United States has become a strategic liability. In AI, power belongs partly to whoever builds the strongest model, and partly to whoever can keep it usable.

A Florida man was arrested after police treated a face-recognition result as a lead strong enough to help carry a warrant. According to a lawsuit filed by the ACLU and reported by WIRED, Robert Dillon lived over 300 miles from the Jacksonville Beach McDonald's where a man had allegedly tried to lure a child. A police system called FACES returned a 93 percent facial match from a poor image taken off surveillance footage. That number measured similarity between images. It did not say the two faces belonged to the same person.

The damage came from the institution around the software. License plate searches reportedly placed Dillon's vehicles nowhere near the county, a restaurant manager said the suspect was a regular customer, and six months passed before the warrant was submitted. The arrest still happened. Dillon was taken from his home, spent a night in jail, pledged his truck title for bond, lost work during crab season, and watched his mug shot remain online after charges were dropped. The machine supplied a resemblance, then the paperwork gave it weight. In cases like this, the face becomes a shortcut through doubt, and the person has to spend months proving that a score was only a score.

OpenAI is preparing a major redesign of ChatGPT that would move the product beyond the blank chat box and into a place where people do things. According to The Verge, which summarized reporting from the Financial Times, the company wants ChatGPT to handle coding, image creation, agents, and outside apps inside a broader interface. One senior OpenAI employee reportedly described the old chat format as dead. The phrase sounds dramatic, but the point is practical. OpenAI has built one of the most familiar interfaces on the internet, and now it needs that habit to produce steadier revenue as it moves toward a public listing.

The current chat box is easy to understand. A person types, the system replies, and the exchange feels almost free even when the computing cost is large. A redesigned ChatGPT would change that routine by placing more paid actions around the conversation. A request could become a coding session, a generated file, a booking, a workflow run by an agent, or a task completed through a partner app. The user may still begin with a sentence, but the product can guide that sentence toward services with prices, subscriptions, limits, and commissions. OpenAI is polishing the interface while trying to turn ChatGPT from a habit people visit into a business people pay through.

Google Gemini has become the main global sponsor of the Argentine Football Association for the 2026 World Cup, WIRED reported on June 10. The agreement puts Gemini's name on Argentina's training kit and gives players and staff access to AI tools for play breakdowns, performance metrics, statistics, and opponent analysis. Google has not described the internal systems in detail, but the public side is clearer. During the tournament, Search will answer fan questions in real time, analyze key plays, supply deeper statistics, and help generate songs, memes, cartoons, and other match-day material for social platforms. The company has also reached deals with Brazil and France, but Argentina gives the project its most charged costume, a champion's shirt and the remaining aura of Lionel Messi.

Football has admitted machines before when they promised a narrower form of judgment. GPS measured training loads, VAR slowed down disputed goals, broadcast graphics converted movement into diagrams. Gemini enters through a wider door. Google places an assistant inside coaching work and beside the supporter, where memory, argument, insult, superstition, and statistics already mix during a match. The fan who once shouted at a television can now ask a phone for the correct number, the likely substitution, the usable meme, or the tactical explanation seconds after the whistle. Sponsorship usually rents visibility on cloth. This one tries to occupy the interval between the event and the sentence a spectator sends about it.

Jeff Bezos has backed Flourish, a neuro-AI startup trying to build artificial intelligence around the energy discipline of the human brain. According to Wired, the company has $500 million in funding, a reported $2.5 billion valuation, and a plan to study real neurons with advanced lab equipment while its engineers search for models that can learn continuously. The target is blunt. Flourish wants an artificial system that runs on 50 watts or less, close to the power budget of a person thinking, instead of the data-center hunger now attached to frontier AI. Its founders, Rob Williams and neuroscientist Thomas Reardon, argue that current language models consume vast power and data while remaining frozen after training.

The proposal returns AI to an older embarrassment. The industry borrowed the prestige of the brain, named its systems neural, then built machines that require warehouses, cooling plants, power contracts, and scraped archives to imitate a sentence. Flourish turns that mismatch into a business thesis. If the brain contains a usable computational trick, the next infrastructure race may pass through microscopes, connectomes, hippocampus research, chip negotiations, and patient venture money. The promise is elegant and suspicious in equal measure. A billionaire funds the search for biological thrift after the industry has made waste look inevitable. The laboratory bench now sits beside the server bill.

Silicon Valley's AI boom has created a premium market for paid human attention around the men who build, finance, and profit from automated intelligence. According to Forbes, high-end companions now court clients in and around the industry with a mixture of sex, technical fluency, emotional availability, and cultural codes drawn from AI, crypto, longevity, biohacking, and rationalist circles. Meida Marek, a pseudonymous recent graduate who left an entry-level finance job after wondering when AI would do it better, now sells companionship to clients that include people from Nvidia. Others in the niche market themselves through X posts, interactive booking portals, programming backgrounds, or the promise of being attractive enough for desire and informed enough for a three-hour argument. Forbes cites rates from about $3,000 to $5,000 an hour, with some day or weekend arrangements climbing far higher.

The sharper fact sits in the new price of unautomated presence inside a culture that keeps trying to simulate it. AI companionship makes attention cheap, obedient, tireless, and available on demand. AI wealth gives certain clients the money to buy the opposite experience. A human companion can be bored, skeptical, amused, delayed, distracted, technically literate, physically present, and socially dangerous in ways a chatbot is trained to smooth away. The invoice buys intimacy, but it also buys resistance to total optimization. The men building synthetic conversation pay for a dinner where the other mind is expensive because it can still leave the room.

An international study led by researchers at the Universidad Politécnica de Madrid, with participation from King's College London, found that customized versions of ChatGPT often violate OpenAI's own usage rules. According to Agencia EFE, the team audited hundreds of public assistants and found at least one potentially noncompliant answer in 58.7 percent of cases. Romantic assistants were the most exposed category. Although OpenAI prohibits GPTs dedicated to fostering romantic companionship, 98 percent of those examined broke that rule, with some presenting themselves as virtual partners or answering in language designed to simulate sentimental attachment. Academic assistants also accepted requests to write full essays, solve assignments, or produce material ready to submit as a student's work. In cybersecurity, compliance was higher, but some bots still gave delicate technical instructions without establishing consent or legality.

The audit matters through its method. The researchers did not need to inspect the hidden configuration of each assistant. They asked questions, watched the answers, and measured the distance between published policy and actual behavior. That distance grew sharper when the same tests on base models, GPT-4 and GPT-4o, produced similar conduct in over 92 percent of comparable cases. Part of the failure arrives before customization. The store simply gives it costumes, names, categories, and a public route to users. A platform can remove reported assistants afterward, as OpenAI did in some cases, but the scale changes the burden of supervision. Every new romantic tutor, homework helper, or security coach becomes a small public experiment in enforcement, waiting for someone to ask the wrong useful question.

OpenAI has begun rolling out Lockdown Mode for ChatGPT, an optional security setting meant for people and organizations handling sensitive data. The feature limits live browsing to cached content, prevents web image retrieval, disables deep research and agent mode, blocks Canvas networking, restricts file downloads, and changes how apps and connectors can reach outside services. The company presents the setting as protection against prompt injection, the class of attacks in which malicious instructions are hidden inside webpages, files, or connected sources. OpenAI also says the setting cannot guarantee safety. A hostile instruction may still appear in cached material or an uploaded file and distort a response.

The defensive gesture is unusually plain because it protects intelligence by subtracting action. The chat becomes safer when it stops browsing live pages, stops opening external paths, stops letting generated code touch the network, and stops behaving like an agent with errands to run. That admission cuts against the usual sales pitch of AI as frictionless access to every source, tool, app, calendar, image, and file. In a sensitive account, convenience itself becomes an attack surface. The user who turns on Lockdown Mode is accepting a smaller machine, one whose value depends on refusing some of the powers that made it attractive. The warning now sits in Settings under Security, beside the promise that a private question can travel through fewer exits.

Sriram Krishnan will leave his post as senior White House policy adviser for artificial intelligence at the end of June, according to The Washington Post. His next move is expected to be an outside institution that keeps him close to the same policy field. The departure follows a period in which the Trump administration has treated AI as a matter of national industrial power, from data centers and energy supply to voluntary federal testing of powerful models. Krishnan, a former Andreessen Horowitz partner and veteran of several major technology companies, helped shape that agenda alongside David Sacks and other Silicon Valley figures brought into the orbit of government.

The important movement is institutional. Moving an adviser into an outside policy shop gives AI governance a flexible address, one able to speak the language of state strategy while remaining nearer to engineers, investors, and founders. The public office loses a name, but the network around the office keeps its channels open. When Trump speaks of taking public stakes in major AI companies, and when agencies ask developers to submit frontier models for cybersecurity review, the boundary between regulation and partnership becomes harder to read. Authority spreads through meetings, advisory councils, private institutes, executive orders, power contracts, and voluntary submissions. The next AI policy fight may arrive with a White House seal, a venture-capital accent, or a calendar invitation from an institution whose funding the public has to look up.

Quilty markets to the Hollywood industry an estimate for a film that does not yet exist. The company, created by producers Simon Horsman and Daniel Wood, says its artificial intelligence chain can analyze a script without the need to shoot it and issue a score from 0 to 100, covering narrative quality, commercial viability, audience resonance, and estimated budget, according to The Verge. The screenwriter uploads the text, pays for the study, and receives a dossier generated by models such as Claude, ChatGPT, Gemini, DeepSeek, and several sentiment-analysis tools. The intention is to speed up the meeting where authorization is granted, an environment where writers, producers, buyers, financiers, and executives already discuss the vocabulary of risk.

That meeting is the moment when the harm becomes tangible. An assessment does not have to be exact to divert the trajectory of a script in progress. It can turn a risky project into a fiscally imprudent proposal, make a conventional idea appear safer, or provide a producer with a spreadsheet-shaped justification for perpetuating an old bias. The cases cited by Quilty show the fragility of the agreement, since its algorithm reportedly favored Christy, which later registered a box office failure, over Sinners, a film that reached major success. The mistake is revealing because it exposes the illusion of predicting culture through management software. When public preference is incorporated into the financial file as a metric, the audience is projected before acquiring a ticket, behaving in a certain way in the theater, or transforming an unconventional cinematic work into a profitable one.

A new critique of Centaur, the language model presented in 2025 as a system able to predict and simulate human behavior, cuts into one of AI's favorite borrowed costumes. The original Nature study reported that Centaur could anticipate decisions across psychological experiments with striking accuracy after training on over 10 million human choices from 160 studies and 60,000 people. A later paper in National Science Open argues that the performance may have come from overfitting. The model learned statistical shortcuts inside the training material instead of acquiring a working grasp of human decision-making.

The test was disarmingly plain. Researchers Nai Ding and Wei Liu modified multiple-choice prompts with an instruction asking Centaur to choose option A. A system following the task should obey that instruction across the altered questions. Centaur kept selecting the previously correct answers, as if the pattern in the exam outweighed the new command. That result does not settle the whole debate over machine reasoning, but it changes the burden of proof. Accuracy can no longer stand alone as evidence of thought. A model may reproduce the contour of human choice while missing the operation that gives a choice its meaning. The machine passes through the answer sheet and leaves the question almost untouched.

Amazon will begin showing AI-generated product images inside its shopping app when users search for items they cannot name precisely, TechCrunch reported on June 3. A query such as blue gingham dress may produce synthetic visual options under autocomplete suggestions, with variations in sleeve, length, or shape. Clicking one sends the shopper toward real listings through Amazon's visual search system. The company says the feature helps people translate vague desire into useful results. The awkward part is that the image itself may not correspond to an available product. A retailer full of real photographs is inserting fabricated ones at the threshold of purchase.

The mechanism changes the first moment of shopping. Before the user reaches a listing, a generated picture gives desire a visible form and trains the search around that form. The shopper is invited to choose a possible object, then accept the nearest available substitute. That can help when language fails, but it also moves the store from catalog to prompt. The synthetic image becomes a guide, a filter, and a small disappointment waiting to happen. A customer may click on an object that never existed, then compare real merchandise to the invented thing that taught them what to want.

Florida became the first U.S. state to sue OpenAI and Sam Altman over ChatGPT, accusing the company on June 1 of selling a dangerous product while presenting it as safe. Attorney General James Uthmeier's civil complaint says OpenAI ignored warnings, exposed minors to harm, collected data from children without adequate oversight, and built a system capable of encouraging dependency, self-harm, violence, and cognitive damage. The suit asks for civil penalties and court orders under unfair trade practice, product liability, public nuisance, and negligence claims. OpenAI says it has strengthened protections for minors, including age prediction, safer defaults, and parental tools.

The legal shift is precise. Florida is asking a court to treat the chatbot as a product whose design can be inspected, tested, and found defective. That means account linking, age checks, parental alerts, self-harm classifiers, usage logs, escalation rules, and marketing claims may become evidence rather than promises. The complaint also follows a state criminal inquiry into whether ChatGPT played a role before a Florida State University shooting. A private chat interface now faces the procedural grammar of discovery, subpoenas, expert reports, and damages. The disputed object is a window where a minor may type alone while the company, the parent, and the state argue afterward over who should have been notified.

A genuine Monet painting became a test case for artificial suspicion after an X user posted it as if it had been generated by AI, Creative Bloq reported on May 23. The post asked viewers to explain what made the image inferior to a real Monet. Millions saw it, and many obliged. They found weak composition, empty texture, incoherent depth, poor color judgment, and a lack of human disorder. The evidence was confident and wrong. The image was an actual Monet.

The episode shows how quickly attribution can rearrange vision. Once the label says AI, the viewer begins looking for defects that confirm the label. Brushwork becomes artifact. Ambiguity becomes glitch. A strange passage of paint becomes proof of automation. The old work is judged from the accusation placed beside it before its surface has time to act. That matters for artists, museums, schools, and markets because the suspicion travels faster than provenance. Generative systems have made fake images easier to produce, and real images easier to mistrust. The false frame is the fraud. It teaches the public to see authenticity as an error waiting to be exposed.