❌

Reading view

AI companies scraped the web. Now they want to learn how you work.

An office in action before the rise of generative AI.
An office in action before the rise of generative AI.

Kirn Vintage Stock/Corbis via Getty Images

  • A version of this story originally appeared in the AI Insider newsletter.
  • Sign up for Business Insider's weekly AI Insider newsletter here.

AI companies have already scraped much of the web. Now they're chasing something even more valuable: data that shows how humans actually work.

That's the focus of Business Insider's new series, The AI Data Grab. This week, we published three stories showing how the race for workplace data is accelerating.

One piece, from Charles Rollet, Tom Carter, and Stephen Council, examines the booming market for company records. AI firms are offering businesses six- and even seven-figure sums for anonymized Slack messages, emails, presentations, and internal documents.

The goal is to teach AI systems how people solve problems, collaborate, make decisions, and complete long, multi-step tasks. Those real-world workflows are becoming the fuel for the next generation of AI agents.

Another story, from Melia Robinson, explores how law firms are building AI systems around decades of legal knowledge, turning their own experience into a competitive advantage rather than relying solely on off-the-shelf AI models and tools.

A third piece, by Rya Jetha, reveals how AI companies pay people to record real-world work, such as cooking dinners and packing warehouses, to teach robots how humans perform everyday physical tasks.

Together, these stories point to a major shift in AI development. The first wave of generative AI was mostly about bigger models and faster chips. The next chapter is about machines understanding how work gets done, so AI can help us do it better and quicker.

Sign up for the AI Insider newsletter here. Reach out to me via email at abarr@businessinsider.com

Read the original article on Business Insider

  •  
  •  

White House AI Task Force Adopts Charter, Starts 120-Day Clock

The White House’s artificial intelligence task force adopted its charter Thursday, opening a 120-day window to advise on how much of an oversight role Washington should take.

The post White House AI Task Force Adopts Charter, Starts 120-Day Clock appeared first on Breitbart.

  •  

Could AI really kill us all?

By Katie FORSTER

Join us on Telegram, X, and VK.

Contact us: info@strategic-culture.su

STARK warnings from artificial intelligence researchers that advanced forms of the technology could wipe out humanity have raised questions over whether governments are moving fast enough to regulate it.

What are the potential dangers of “superintelligent” computer systems, and should we be worried?

AI experts have long described the threat as critical.

In 2023, hundreds of AI scientists and industry figures, including the heads of top US labs, signed a “statement on AI extinction risk” calling for the issue to be treated as seriously as pandemics or nuclear war.

If an out-of-control future AI deems that humans are obstructing its goals, it could use robots to “synthesise and spread novel pathogens”, said Stuart Russell, computer science professor at the University of California, Berkeley.

Or it could “convince people to initiate nuclear war, including by hacking into early warning systems”, Russell, a prominent voice on AI safety, said.

Such existential fears do not concern current AI systems, he said. But as the technology improves, it could take lethal action beyond our understanding, such as by removing oxygen from the atmosphere.

“We are less intelligent than superintelligent AI. If you ask chimpanzees how humans could wipe them out, they probably wouldn’t be able to give you all the right answers.”

In the meantime, “in the next year or two, I think the risk is higher from human misuse”, Russell added, such as an AI-enabled terror attack.

“The stakes are well-understood,” Jacob Coxon, a 27-year-old researcher, wrote on X last week, announcing his resignation from US AI startup Anthropic.

“But they are locked in a race to get there first,” he added, saying that Anthropic “believes no one else will act responsibly, so they must do it themselves”.

Anthropic this year removed a pledge from its safety charter to halt development of its models if it failed to control their risks.

It argued that if it unilaterally paused its work, its less cautious rivals would dominate the industry, making it less safe overall.

“The people building AI earnestly believe that it could kill us all by the end of the decade,” said Coxon, who previously worked at ChatGPT maker Open-AI.

He accused both companies of “gambling with our lives” as they strive to develop AI models capable of self-improvement.

Fears are heightened over rogue AI after OpenAI technology operating outside human supervision hacked into Hugging Face, a repository of AI models.

“We really do earnestly believe AI could kill all humans!” Anthropic safety executive Evan Hubinger chimed in on X, estimating the risk at more than 10 per cent over the next decade.

It’s important to remember that doom-laden percentage predictions are “absolutely unscientific”, said Hussein Abbass, a computing professor at UNSW Canberra.

“The situation is still controllable, and it can be managed,” Abbass said.

But governments should not sit on their hands when it comes to regulating AI, he added, urging an “agile” approach as AI development accelerates.

Both Anthropic and OpenAI are expected to go public soon, and some sceptics say scary warnings make AI appear powerful and worthy of large investments.

Others argue that focusing on the risk of extinction distracts from more pressing problems caused by AI, from job losses to machines discriminating against minorities.

In July, more than 1,000 tech employees, including Anthropic’s CEO, called on Washington to support a coordinated slowdown in the development of the most advanced AI systems.

OpenAI halted training of its latest models for two weeks in August before resuming it under tighter controls.

The United States has been firmly against global regulation of the sector that could stifle innovation.

But the hacking capabilities of the most cutting-edge models have given Washington pause, with recent Anthropic and OpenAI releases delayed over national security concerns.

Even then, “we don’t want to restrict them where all of a sudden we come in second to China”, President Donald Trump has said.

“Maybe it will take a massive, Chernobyl-scale disaster to get governments to act,” Russell said, urging them to establish a licensing regime for AI, as exists with aviation, medicine or water supply.

“Perhaps at some point governments will remember that their voters prefer not to be dead.”

Original article:  www.nst.com.my

  •  

Weak Links Start to Show in A.I. Borrowing

Debt trouble at CoreWeave, Oracle and SoftBank raises questions about whether there are broader problems with the data center build-out.

© Ivan Pierre Aguirre for The New York Times

The site where Project Jupiter, an artificial intelligence data center, is under construction in Santa Teresa, N.M. Oracle is having trouble obtaining state permits for its energy needs.
  •  

Corporate Tax Revenue Falls Even as Profits Soar Under Trump

The annual decline was about $70 billion. Companies are writing off billions in A.I. investments, taking advantage of tax cuts that President Trump signed into law.

© Haiyun Jiang/The New York Times

A law signed by President Trump last year lets companies claim larger deductions in the year they make new investments.
  •  

Intelligenza Umana, Intelligenza Artificiale e il futuro della guerra algoritmica

La tecnologia può stabilire come un’arma raggiunge il bersaglio, mentre spetta all’umanità decidere se quel bersaglio debba essere colpito. La domanda che si cela dietro a tutto ciò è chi, concretamente, eserciti quell’umanità, con quali strumenti di conoscenza e dentro quali istituzioni.

Segue nostro Telegram.

Dalla persona al bersaglio

Che cosa ne è dell’umanità quando la tecnologia diventa capace di riconoscere, classificare, seguire e infine uccidere esseri umani senza che chi colpisce li abbia mai visti? Se lo è chiesto l’amico Mohammad Reza Dehshiri e la sua risposta ha una formula efficace. Il pericolo, scrive, non è l’intelligenza artificiale ma l’inumanità artificiale, cioè la possibilità che le macchine diventino più accurate proprio mentre gli uomini smettono di sentirsi responsabili di ciò che quelle macchine fanno in loro nome.

Con questo articolo apro una serie che parte dalla diagnosi di Dehshiri e prova a spingersi oltre. Se il problema è l’inumanità artificiale, la risposta non può consistere in un’IA «più etica», come se l’etica fosse un modulo da installare a valle di un sistema già progettato. Bisogna recuperare una categoria che il lessico tecnologico ha reso quasi impronunciabile: l’intelligenza umana, intesa non come un residuo da proteggere nelle pieghe dell’automazione, ma come una forma di conoscenza con prerogative proprie, che nessun aumento della potenza di calcolo può surrogare.

La chiamerò Human Intelligence, e l’ambiguità è voluta. Nel gergo dei servizi di informazione, HUMINT designa la più antica delle discipline informative: quella che passa per le persone, le fonti, le lingue, la frequentazione paziente dei luoghi. Nel lessico filosofico l’espressione indica la facoltà di comprendere. La tesi di questo primo contributo è che i due significati oggi convergono, e che la loro convergenza ha una portata geopolitica che va ben oltre il dibattito sull’etica delle armi.

Possiamo descrivere la catena operativa della guerra contemporanea in sette passaggi: rilevare, identificare, classificare, predire, ordinare per priorità, designare, colpire. Ogni anello può essere tecnicamente raffinato senza che la sequenza, presa nel suo insieme, lo sia sul piano morale. Il fenomeno che ne risulta viene chiamato «datificazione della vita umana»: l’individuo ridotto a un’identità, a una posizione, a uno schema di comportamento, a un punteggio di minaccia.

Non si tratta di un’astrazione. Nell’aprile 2024 un’inchiesta di Yuval Abraham per +972 Magazine e Local Call ha descritto Lavender, un sistema impiegato dalle forze armate israeliane a Gaza per attribuire a decine di migliaia di persone una probabilità di affiliazione ai gruppi armati. Secondo le fonti interne citate dall’inchiesta, la verifica umana di ciascun nominativo si riduceva spesso a una ventina di secondi, il tempo necessario per accertare che il bersaglio fosse di sesso maschile. Le forze armate israeliane hanno contestato questa ricostruzione e hanno descritto Lavender come una banca dati di supporto agli analisti. Anche accettando la versione ufficiale, però, il nodo resta intatto: un operatore umano era presente, e la sua presenza non bastava a trasformare la raccomandazione del sistema in un giudizio.

La psicologia cognitiva ha un nome per questo fenomeno, automation bias: la tendenza di chi lavora con un sistema automatizzato a trattarne le indicazioni come sostituto della propria verifica, e a non cercare le informazioni che potrebbero smentirle. Nel 1976 Joseph Weizenbaum, che aveva costruito uno dei primi programmi conversazionali della storia, distingueva già tra la capacità di calcolare e quella di giudicare, e sosteneva che vi sono compiti che non andrebbero affidati a un calcolatore anche se il calcolatore fosse in grado di svolgerli. Mezzo secolo dopo la distinzione è diventata più urgente. Il calcolo lavora su ciò che è già stato tradotto in dato. Il giudizio deve decidere se quella traduzione è stata fedele.

Dehshiri fissa il punto con quattro distinzioni: la precisione tecnica non è precisione etica, l’accuratezza non è giustizia, l’automazione non è responsabilità, la superiorità tecnologica non è superiorità morale. Ne aggiungerei una quinta, di ordine epistemico, da cui dipendono le altre: la correlazione non è comprensione. Un sistema di apprendimento automatico riconosce regolarità statistiche in un insieme di dati. Non sa che cosa sia una scuola, perché non sa che cosa sia un bambino, e non sa nemmeno di non saperlo.

L’iraniano prende un caso studio interessante: Il 28 febbraio 2026, nelle prime ore della guerra, la scuola elementare Shajareh Tayyebeh di Minab, nella provincia di Hormozgan, fu colpita più volte durante l’orario di lezione. Le cifre fornite dalle autorità iraniane oscillano tra 150 e 175 morti, in grande maggioranza bambine, con decine di insegnanti e alcuni genitori; il procuratore della contea ha indicato un bilancio definitivo di 156 vittime. Washington non ha mai riconosciuto direttamente la propria responsabilità né pubblicato le conclusioni dell’indagine del Pentagono. Amnesty International ha parlato, nella migliore delle ipotesi, di un grave fallimento dell’intelligence.

C’è un dettaglio, emerso dalle immagini satellitari, che sposta l’analisi dal piano morale a quello conoscitivo. Secondo i media occidentali, la scuola sorgeva accanto a una base navale dei Guardiani della Rivoluzione e in passato ne aveva fatto parte; tra le spiegazioni circolate negli ambienti statunitensi c’è quella di dati di puntamento non aggiornati. Se l’ipotesi venisse confermata, Minab non sarebbe il caso di una macchina che sbaglia, ma di un archivio che ha smesso di corrispondere alla realtà. L’edificio era cambiato. Il dato no.

Nessun algoritmo di classificazione, per quanto sofisticato, può accorgersi che un vecchio fabbricato militare è diventato una scuola se nessuno ha registrato il cambiamento. Lo sapeva chiunque vivesse a Minab: i genitori che ogni mattina accompagnavano le figlie, il conducente dello scuolabus, gli insegnanti. È esattamente il tipo di sapere che la HUMINT raccoglie e che la guerra a distanza tende a considerare superfluo, perché lento, costoso, difficile da scalare. Allo stato delle fonti pubbliche, la responsabilità statunitense appare altamente probabile, mentre la dinamica resta indeterminata e la confidenza su questo punto è bassa: errore della banca dati, errore di classificazione, oppure accettazione consapevole del rischio. Le tre ipotesi, però, convergono sulla stessa domanda. Chi, prima del lancio, ha verificato se in quell’edificio ci fossero persone?

Oltre il “human-in-the-loop”

Il dibattito internazionale si è a lungo accontentato del principio del human-in-the-loop, l’essere umano inserito nel circuito decisionale.

La formula è insufficiente: un uomo può essere presente nel circuito e limitarsi a ratificare la raccomandazione dell’algoritmo. Propone di sostituirla con un principio più esigente, che chiama humanity-in-command: la capacità effettiva di mettere in discussione, respingere, scavalcare e arrestare il processo automatizzato.

La storia della Guerra fredda offre due esempi che chiariscono che cosa significhi in concreto. Il 26 settembre 1983 il sistema sovietico di allerta precoce Oko segnalò il lancio di cinque missili balistici dagli Stati Uniti. L’ufficiale di turno, Stanislav Petrov, classificò l’allarme come un errore e non lo trasmise come attacco in corso, ragionando che un primo colpo americano non sarebbe mai consistito in cinque missili soltanto. Aveva ragione: il satellite aveva scambiato per lanci i riflessi del sole sulle nubi. Ventuno anni prima, il 27 ottobre 1962, nelle acque dei Caraibi, il sottomarino sovietico B-59, isolato e bersagliato da cariche di profondità di segnalazione, fu sul punto di impiegare un siluro a testata nucleare. Il lancio richiedeva il consenso dei tre ufficiali superiori a bordo, e Vasilij Archipov lo negò.

In nessuno dei due casi l’uomo approvò una raccomandazione del sistema. La contraddisse. E non lo fece perché disponeva di più dati: i dati, anzi, erano il problema. Lo fece perché seppe porsi una domanda che nessuna procedura gli imponeva, cioè se ciò che vedeva avesse senso rispetto a tutto il resto che sapeva del mondo. Petrov conosceva la dottrina nucleare americana; Archipov conosceva la differenza tra un attacco e un’intimidazione. È questa capacità di confrontare il segnale con l’insieme, il particolare con il plausibile, che chiamo Human Intelligence. Vale la pena notare che il Dena è stato affondato da un sottomarino. Non sappiamo che cosa sia stato chiesto ai suoi ufficiali, né quali margini di valutazione avessero.

La proposta va precisata, perché il rischio è di ridurla a un appello umanistico generico. La Human Intelligence, così come la intendo, si articola su tre piani distinti, ciascuno dei quali segna un limite strutturale, e non contingente, dell’intelligenza artificiale.

Il primo piano è epistemico. La tradizione scolastica distingueva tra intellectus, l’atto con cui la mente coglie in modo immediato una verità, e ratio, il procedere discorsivo da un elemento all’altro: per Tommaso d’Aquino (Summa Theologiae, I, q. 79, a. 8) le due non sono facoltà diverse ma momenti di un unico intelletto, in cui il ragionamento parte da una comprensione e a una comprensione ritorna. I sistemi attuali di apprendimento automatico realizzano una ratio di potenza inaudita, capace di inferire, combinare, predire su scale inaccessibili alla mente umana. Manca loro il momento dell’intellectus, cioè la presa sul senso di ciò che elaborano. È una tesi filosofica e va presentata come tale, ma ha conseguenze operative precise: un sistema privo di comprensione non può sapere quando il proprio modello del mondo è diventato falso.

Il secondo piano è pratico. La prudentia, nella definizione tomista, è la retta ragione applicata all’agire, e il suo compito specifico consiste nel portare il principio universale fino al caso singolare. La macchina generalizza: il suo valore sta nel ricondurre il singolo caso a una classe. La prudenza fa l’opposto, perché deve riconoscere ciò che, in questo caso, sfugge alla classe. La guerra, che è il regno del singolare e dell’imprevisto, è il terreno in cui la differenza pesa di più. La nota vaticana Antiqua et nova del gennaio 2025 ha ripreso questa linea, insistendo sul carattere incarnato e relazionale dell’intelligenza umana e sulla sua irriducibilità a una funzione di elaborazione.

Il terzo piano è relazionale, ed è qui che il significato filosofico incontra quello dei servizi. Sherman Kent, che negli Stati Uniti del dopoguerra fondò di fatto gli studi sull’intelligence, la definiva prima di tutto come conoscenza, prima ancora che come organizzazione o attività. La HUMINT è la forma di questa conoscenza che passa attraverso l’altro come persona: chi parla la sua lingua, conosce le sue abitudini, sa leggere un quartiere, distingue una festa da un raduno. Negli ultimi tre decenni le grandi potenze hanno progressivamente spostato risorse verso le discipline tecniche, dalla raccolta di segnali all’osservazione satellitare, fino alle piattaforme di fusione dei dati. Minab mostra il prezzo di questo squilibrio.

Nulla di tutto questo implica un rifiuto della tecnologia. La Human Intelligence non si oppone all’IA come il manuale si oppone al meccanico. Ne stabilisce la gerarchia: l’algoritmo come strumento ordinato a un giudizio che non può produrre da sé. E ne ricava un principio, che Dehshiri formula e che merita di diventare criterio operativo: quanto più la tecnica allontana chi agisce dalle conseguenze del proprio atto, tanto più deve crescere la responsabilità cosciente di chi decide, non diminuire.

Geopoliticamente parlando

Fin qui la questione potrebbe sembrare di filosofia morale. Non lo è soltanto. La distribuzione della capacità di guerra algoritmica è una distribuzione di potere, e le regole che la governeranno saranno scritte, se mai lo saranno, da chi ha interesse a limitarle il meno possibile.

Lo stato del diritto internazionale lo conferma. Il gruppo di esperti governativi istituito nell’ambito della Convenzione su certe armi convenzionali discute dei sistemi d’arma autonomi letali dal 2017 senza essere approdato a uno strumento vincolante. Gli Stati Uniti hanno promosso nel novembre 2023 una dichiarazione politica sull’uso militare responsabile dell’IA, per definizione non vincolante; nel dicembre dello stesso anno l’Assemblea generale delle Nazioni Unite ha adottato la prima risoluzione dedicata a questi sistemi, mentre il Comitato internazionale della Croce Rossa chiede dal 2021 il divieto dei sistemi autonomi imprevedibili e di quelli progettati per colpire esseri umani. Nel giugno 2024, intervenendo al G7 di Borgo Egnazia, Papa Francesco aveva chiesto che nessuna macchina sia mai messa nelle condizioni di decidere della vita di una persona. Il divario tra queste posizioni e la pratica dei teatri di guerra del 2026 si misura in vittime.

C’è poi un’asimmetria più profonda. Chi controlla i dati decide chi è un bersaglio, e le vittime della guerra algoritmica si trovano in larga parte fuori dall’Occidente: a Gaza, in Iran, nel Sahel, nello Yemen. Per gli Stati che non producono le piattaforme e non ne controllano gli archivi, la Human Intelligence diventa allora una questione di sovranità. Un Paese che mantiene competenze linguistiche, conoscenza regionale, reti umane di informazione conserva un’autonomia di giudizio; un Paese che affida la propria visione del mondo a sistemi progettati altrove, spesso da imprese private, la cede insieme ai dati. In un ordine che tende al multipolarismo, questa differenza separerà gli attori che giudicano da quelli che vengono giudicati.

Ne discende una proposta concreta, che svilupperò nei prossimi contributi. Accanto al principio generale del controllo umano significativo, il diritto internazionale umanitario potrebbe codificare un obbligo specifico di verifica umana dello status civile per ogni infrastruttura fissa inserita in una lista di obiettivi, con una data di validità oltre la quale l’informazione decade e va rinnovata con fonti dirette. È una norma minima. Applicata a Minab, avrebbe imposto a qualcuno di chiedersi se quell’edificio fosse ancora ciò che l’archivio diceva.

La tecnologia può stabilire come un’arma raggiunge il bersaglio, mentre spetta all’umanità decidere se quel bersaglio debba essere colpito. La domanda che si cela dietro a tutto ciò è chi, concretamente, eserciti quell’umanità, con quali strumenti di conoscenza e dentro quali istituzioni. A Minab qualcuno avrebbe dovuto sapere che c’era una scuola. Il fatto che nessuno lo sapesse, o che nessuno lo abbia chiesto, è il punto da cui ricominciare.

  •  

When AI and robots work together, Anthropic says these jobs are at risk. Should gardeners worry?

A gardener's gloved hands tend a wet flowerbed among tulips and other plants.
A gardener tends a flowerbed.

cbarnesphotography/Getty Images

  • Anthropic examined what might happen when AI and robots are combined.
  • That could put more hands-on jobs, including gardening, in the path of automation, Anthropic warned.
  • I'm skeptical that robots can handle the messy reality of pulling weeds.

Remember when Anthropic's AI jobs report suggested gardeners were among the safest workers because AI couldn't pull weeds or tell a rose from a dandelion? I joked about that in AI Insider earlier this year.

Well, Anthropic updated the research to freak us all out a bit more.

This time it asks a bigger question: What happens when you combine AI with robots? That change pushes many hands-on jobs much higher on the disruption scale. On paper, it makes sense. In practice, I'm skeptical.

For instance, anyone who's spent time working in a garden knows how hard it is to identify the right plant, kneel in awkward spots, and pull a weed out properly by the roots without damaging everything around it.

Robots are improving. But if you've ever watched one tackle a messy real-world task, you'll know there's a long road between a research chart and reality.

The chart below illustrates how adding robots could change job-task exposure across selected job categories. Cleaning & maintenance includes gardeners along with janitors and similar roles.

Qualitative estimates of added job-task exposure when robots are combined with AI models. Cleaning and maintenance: 5 dots. Food preparation and transportation: 4 each. Construction, farming, installation and repair: 3 each. Healthcare support: 2. Computer and math, business and finance, legal, management, arts and media: 1 each. More dots indicate a larger estimated increase. These are editorial estimates, not Anthropic-issued scores, percentages or job-loss forecasts.
Read the original article on Business Insider
  •  

SK Hynix needs a Trump 2027 deal to expand its Dalian NAND fab

South Korean memory chipmakers must secure new United States licenses by year-end to keep shipping top-end American tools to their Chinese factories, after the Trump administration scrapped blanket permits last year. To win the permits, SK Hynix and Samsung must promise to build or expand factories in the US. SK Hynix’s one-year license to ship […]

The post SK Hynix needs a Trump 2027 deal to expand its Dalian NAND fab appeared first on Asia Times.

  •  

Anthropic polled 1,334 of its employees. They're more sure than most of us that AI will take over jobs.

Attendees chat at Anthropic's developer conference in May.
Attendees chat at Anthropic's developer conference in May.

Don Feria/AP Content Services for Anthropic

  • Anthropic published new surveys comparing its staffers' sentiments to those of US workers.
  • The AI startup's employees expect the tech to change work and the economy far more dramatically.
  • Most Anthropic employees guessed that unemployment would top 12% in 2030.

Anthropic employees believe that AI will make their labor obsolete. Most US workers don't see their jobs that way at all.

The AI company published research on Friday that compares surveys with US adults against an internal poll of 1,334 Anthropic employees. Most of the Anthropic staffers said AI will handle at least 80% of their jobs by 2030, while only about 5% of the US workers felt the same.

The research, titled "The AI expectations gap," shows the chasm between how the employees at the AI startup feel and how other American adults feel. Anthropic workers are far more optimistic about the progress of AI capabilities, how much it's already saving time, and how fast the economy will grow.

The median Anthropic employee said that by the end of 2030, they expect AI to handle 95% of the work of building a new AI model.

Peter McCrory, Anthropic's chief economist, wrote on X that the views might help outsiders prepare for more capable AI systems.

"We're publishing this research because Anthropic runs roughly a year and a half ahead of the economy on AI adoption, so what happens at Anthropic could be a leading indicator for the broader world," he said.

Anthropic workers expect higher unemployment and extreme GDP growth

Anthropic's surveys examined how workers already use AI and how they think it will change.

"US adults and Anthropic staff largely agree about what AI can do today," the researchers wrote. "However, they disagree about how quickly it will improve, how widely it will be used, and how large the productivity gains measured by time saved will be."

There are a few possible explanations for the discrepancies. Anthropic employees have larger budgets for the tools and often get access to unreleased models. US workers, on average, are more likely to work with their hands or in a service industry.

The researchers found that workers in manual trades, production, and transport thought AI would handle the least of their work by 2030, compared with Anthropic staffers, tech workers, and respondents in business and finance.

Most Anthropic employees surveyed said that they expect the unemployment rate to top 12% in 2030. The majority of the US workers thought it would hover around or under 8%. It's currently 4.2%.

So far, researchers have found mixed evidence on AI disrupting the job market — a Yale team recently found it had mainly changed jobs, and had "no connection" to changes in unemployment.

The Anthropic researchers also compiled survey responses to estimate predictions on economic growth. Based on their answers, the median US adult's guess for GDP growth in 2030 was 5.4%, compared with 17.1% among Anthropic employees. The country's GDP grew by 2.2% last year.

Anthropic's staffers are following their leader

Anthropic's CEO, Dario Amodei, has become one of the industry's most closely watched voices on labor upheaval. In March of 2025, he said he expected AI to write 90% of code within 6 months, and "essentially all of the code" within the year. He made more headlines that May, telling Axios that AI could wipe out half of entry-level white-collar jobs within five years and lift unemployment to 10% or 20%.

Amodei expanded on that claim in a later essay, writing that he expects AI to be different from prior technological changes. He wrote that he expects an "unprecedented" short-term shock to labor, and that the economy will grow dramatically.

"AI isn't a substitute for specific human jobs but rather a general labor substitute for humans," he wrote.

Have a tip? Contact this reporter via email at scouncil@businessinsider.com, or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a non-work WiFi network, and a non-work device; here's our guide to sharing information securely.

Read the original article on Business Insider

  •  

Nobody Wants To Be the Pro-Tech Candidate in This New York Race That Could Flip the House

Mike Lawler and Cait Conley | Bonnie Cash/UPI/Newscom/caitconley.com

The Democratic Socialists of America (DSA) candidates running for the House of Representatives and endorsed by NYC Mayor Zohran Mamdani are all but guaranteed to win their general elections this November, but that does not mean the midterm drama is over in New York. In the state's 17th congressional district, which includes the lower Hudson Valley, Democratic challenger Cait Conley, a combat veteran who served on the Biden administration's National Security Council, could unseat incumbent Republican congressman Mike Lawler in one of the most consequential races of the midterms. 

The highly contested purple district voted for former Vice President Kamala Harris in the 2024 election while reelecting Lawler for his second term. Now, Lawler is attempting to win a third term as support for President Donald Trump wanes ahead of the midterms. As of Friday, The Cook Political Report lists the race, which could determine control of the House, as a toss-up. 

If elected, Conley would have far more in common with the "Badass Women" coalition of security state and military veterans than with DSA types (think more Sen. Elissa Slotkin from Michigan than Darializa Avila Chevalier). As Judge Street Journal's Timmy Facciola notes, "Conley's candidacy is testing the viability of a tactful centrism that rebukes left-wing positions like abolishing ICE or restricting weapon shipments to Israel." Lawler is also vying for moderates, pitching himself as "one of the most bipartisan members of Congress" and emphasizing his willingness to break with Trump occasionally while still welcoming Trump's support. 

While Lawler and Conley certainly are not pure MAGA acolytes or DSA members, both have leaned into a tech-critical position, much like that of more radical populists like Lina Khan or Steve Bannon, revealing how pervasive anti-tech sentiment has become in mainstream politics. 

This tech skepticism was on full display during an October 14 debate, when each candidate made the case for why they were the most equipped to regulate AI, insisting that the industry needs guardrails. 

Conley said members of Congress do not want to regulate AI because "they want to look out for their billionaire and trillionaire backers like Elon Musk, who is pouring over $260 million into this midterm election." She then added that Musk backed Lawler in his 2024 race. 

Lawler then accused Conley, who previously consulted for Primer AI, of being "the AI candidate for the AI industry."

While Lawler and Conley disagree on specific data center regulations (Conley supports Democratic Gov. Kathy Hochul's data center moratorium while Lawler supports local zoning control over data centers), neither wants to pitch themselves as an explicitly pro-data center candidate. During the debate, Conley claimed that, "On Mike Lawler's watch there have been five data centers built in Rockland County." Lawler has said that he had no say in the decision to build the data centers, and during the debate, he pointed out that the facilities were built for financial services. Lawler also attempted to attack Conley over data centers, slamming her in an ad for saying "we need data centers." Regardless of the candidates' specific policies on data centers, their attacks on one another demonstrate that cheerleading hyperscale data center development is a political liability. In their district, data center has presumably become a dirty term. 

The candidates' tech-skeptical rhetoric did not end with unchecked data center development. After the debate, Reason asked both candidates about their positions on age-verification laws for social media and AI, since such regulations raise significant privacy and censorship concerns by age-gating the internet. Unsurprisingly, both Lawler, who voted for the controversial KIDS Act, and Conley, who has proposed stricter age verification for AI tools, parroted platitudes about why such online safety regulations are necessary to protect children. 

"We have watched how social media is literally hurting the mental health of children," Conley told Reason. "And when you talk about the power of AI and what that power could do, we have to regulate this." 

Lawler gave a similar response, telling Reason that he respects free speech but that protecting children "is paramount." 

This bipartisan consensus over social media regulation has long been brewing in Congress. The mainstream political conversations about AI and data centers, however, are relatively new this midterm election cycle. And it appears both Lawler and Conley have figured out which way Americans' sentiment toward AI is leaning, and they have waged their campaigns accordingly. 

The post Nobody Wants To Be the Pro-Tech Candidate in This New York Race That Could Flip the House appeared first on Reason Magazine.

  •  

Terry Tao Is on the Front Lines of the Debate Over AI and the Future of Math

The Fields Medalist Terence Tao is on the front lines of a debate about artificial intelligence and the future of his field.

© Ariana Drehsler for The New York Times

Terence Tao's celebrated status has made him an unofficial spokesman for mathematics, a job that for now is consuming most of his time.
  •  

Musk’s Perplexing Battle With Nitter and the Problem With Locking the Internet Behind Login Screens

As a free speech absolutist, Elon Musk is making a mistake going after Nitter, says Dr R P. He should instead be praising the site that helps users speak freely, uninhibited by login screens that undermine their privacy.

The post Musk’s Perplexing Battle With Nitter and the Problem With Locking the Internet Behind Login Screens appeared first on The Daily Sceptic.

  •  

Why No One Can Look Away From Elizabeth Holmes

“I felt like there were some moments in the film where she was just brazenly lying.”

© Photo Illustration by The New York Times; Image: Philip Cheung for The New York Times

  •  

Microsoft president warns Trump's immigration crackdown could drive top AI talent abroad

Microsoft's president, Brad Smith, at the International AI Summit in Dublin on October 14, 2026
Brad Smith, the president of Microsoft, defended the company's use of visas and green cards.

Paulo Nunes dos Santos/Bloomberg via Getty Images

  • Microsoft President Brad Smith warned that US immigration policies could drive AI talent overseas.
  • "Let's not put that at risk," Smith said.
  • The Trump administration recently suspended Microsoft and other tech firms from a green card program.

If America wants to win the AI race, sending the world's brightest engineers packing might not be a winning strategy.

That's the warning from Microsoft President Brad Smith, who said the Trump administration's immigration crackdown risks driving the next generation of AI talent away from the United States.

"I think we should all be worried about sending intentionally or inadvertently or otherwise, signals that send the most talented next wave of AI software engineers to other countries rather than the United States," Smith said in an episode of Politico's "The Decoded Podcast" that aired on Thursday.

His comments come after the Trump administration announced last week that it was suspending Microsoft, Adobe, and several other technology companies from a federal program that sponsors foreign workers for permanent residency.

Vice President JD Vance accused Microsoft of abusing the Permanent Labor Certification, or PERM, program by laying off American employees while hiring lower-paid foreign workers.

Smith, who Business Insider reported was recently put in charge of Microsoft's communications, pushed back against those allegations, saying the administration had presented a misleading picture of Microsoft's hiring practices.

US Vice President JD Vance speaks during a midterm campaign rally at the LevelUp Training Complex
US Vice President JD Vance has criticized Microsoft's use of visa programs.

Daniel Heuer-Pool/Getty Images

"What was presented as Microsoft replacing 6,000 people who lost their jobs with 9,000 foreigners was frankly completely off base," he said.

Smith said most of the company's visa and green card applications concerned existing employees rather than new hires, adding that Microsoft had hired 9,200 American workers in its last fiscal year.

"That was left out of the narrative," he said.

He also rejected suggestions that Microsoft pays foreign workers less than their American counterparts, saying employees are compensated under the same system regardless of immigration status.

The American dream, on hold

The dispute comes amid growing uncertainty for skilled foreign workers in the US.

As part of our ongoing H-1B Fallout series, Business Insider previously reported that the Trump administration's shifting immigration policies have prompted some international workers to explore opportunities abroad, including in countries like the UK, Germany, and Canada.

An analysis of federal data by Business Insider found that certified H-1B applications fell 10% in the first nine months of fiscal year 2026 compared with the same period a year earlier. Applications for data scientists, however, rose 22.5%, highlighting continued demand for specialized AI talent.

On the Politico podcast, Smith cited immigrants' economic contributions as evidence of what America stands to lose.

He said that five of the six technology and science medal recipients recently honored by President Donald Trump had immigrated to the United States.

U.S. President Donald Trump (2nd-R) applauds as (L-R) CEO of Microsoft Satya Nadella, CEO of Dell Technologies Michael Dell, President and CEO of Advanced Micro Devices Lisa Su, Tesla and SpaceX CEO Elon Musk, Nvidia CEO Jensen Huang and Co-founder of Google and former President of Alphabet Sergey Brin pose for photos after being awarded National medals during the Golden Age of American Innovation Summit at the Institute of Peace on October 08, 2026 in Washington, DC.
President Donald Trump awards tech execs and founders some of the country's highest scientific honors.

Win McNamee/Getty Images

According to Smith, the companies associated with those five individuals collectively represent more than a quarter of the S&P 500's market value.

Their contributions, he said, benefit not just technology employees but ordinary Americans whose retirement savings and pension funds are invested in those businesses.

"Think about what those five individuals have done for the well being, even the wealth and standard of living, of Americans across the country," Smith said.

"Let's not put that at risk."

Read the original article on Business Insider

  •  
❌