CoinPulseHq

A rede layer-1 compatível com Ethereum Harmony propôs encerrar seu blockchain e migrar seu token nativo ONE para o Ethereum, sete anos após o lançamento de seu mainnet. A proposta, anunciada no domingo, vem semanas após um exploit que forçou a rede a planejar um rollback de mais de 109.000 transações.
Proposta de migração da Harmony
Na proposta não vinculativa, a Harmony tiraria um snapshot final da rede, emitiria tokens ERC-20 ONE no Ethereum e migraria listagens de exchanges. Aos validadores seriam oferecidas opções para parar seus nós, continuar como governadores ou ingressar na nova iniciativa de vídeos com IA da Harmony. A proposta não especifica quando o bloco final seria produzido nem se o desligamento seria enviado ao processo de governança liderado por validadores da rede.

The Seattle Times and Newsday filed a copyright infringement lawsuit against OpenAI and Microsoft on September 5, 2026, escalating the publishing industry's legal confrontation with generative AI companies. The lawsuit, reported by TechCrunch, alleges that the companies used the newspapers' journalism to train AI models like ChatGPT and Copilot without authorization or compensation.
The complaint argues that the journalism industry could become "broken beyond repair" due to AI, describing generative AI as "a snake eating its own tail" that could "destroy the very organizations" producing the content it relies on. The legal filing sharply criticizes the AI companies' business model, stating: "AI products like ChatGPT and CoPilot are touted as producers of content, but in fact they are rapacious consumers, devouring human-authored content and delivering back to the world copies and derivative imitations of that same original content they consumed to achieve their commercial objectives."
A Growing Wave of Publisher Lawsuits
The new lawsuit follows a pattern established in December 2023, when The New York Times sued OpenAI and Microsoft for alleged copyright infringement. Since then, numerous other publications have filed similar actions, including the Chicago Tribune, the New York Daily News, and several digital-first outlets like The Intercept and Raw Story.
What makes this particular case stand out is the pre-existing relationship between the parties. Microsoft and OpenAI have previously funded journalism projects and fellowships at The Seattle Times, a fact that complicates the narrative of adversarial parties. A Microsoft spokesperson told GeekWire the company is "surprised by the lawsuit" but remains "always happy to sit down and explore solutions to this type of dispute."
Why This Legal Fight Matters for the News Industry
The outcome of these consolidated disputes could fundamentally reshape how AI companies source training data. At stake is not just financial compensation for past use of copyrighted material, but the establishment of a legal framework for how AI models can be trained on journalistic content going forward.
News organizations have watched with growing alarm as AI-powered search and chat products increasingly deliver answers drawn from their reporting without driving traffic back to their websites. The Seattle Times and Newsday lawsuit directly challenges this dynamic, arguing that AI systems are effectively competing with the very publishers whose work they consume.
The case also highlights a structural tension: even as publishers sue AI companies, many have struck separate licensing deals. News Corp, Associated Press, and Dotdash Meredith have all signed content agreements with OpenAI, creating a split in the industry between those who negotiate and those who litigate. The Seattle Times and Newsday have chosen the courtroom path, though Microsoft's statement suggests a potential openness to settlement discussions.
Legal experts following the broader litigation note that courts have yet to rule definitively on the core question of whether training AI on copyrighted material constitutes fair use. The New York Times case, which remains ongoing, is widely seen as the bellwether that could set precedent for the dozens of similar lawsuits filed since.
For readers and journalists alike, the stakes extend beyond corporate balance sheets. If publishers succeed in establishing that AI companies must license journalistic content, it could create a new revenue stream for an industry that has struggled financially for two decades. Conversely, a ruling favoring the AI companies could accelerate the disruption of traditional news business models.
As this litigation progresses through the courts, the industry will be watching closely for any ruling that clarifies the boundaries between AI innovation and intellectual property protection. The Seattle Times and Newsday lawsuit adds another layer of pressure on OpenAI and Microsoft to reach broader industry-wide agreements rather than fighting each publisher individually.
This article discusses ongoing litigation and market dynamics. It does not constitute financial or legal advice, and the outcomes of legal proceedings remain uncertain and subject to change.
Originally published on CoinPulseHQ: https://coinpulsehq.com/seattle-times-newsday-sue-openai-microsoft-copyright/

The AI industry moves fast enough that its own vocabulary can leave even seasoned technologists scrambling. On September 1, 2026, OpenAI released Astra, its new reasoning model, and with it introduced a term that has since dominated safety discussions: "opaque recurrence." That single phrase — describing a technique where the model loops queries through its internal layers rather than explaining its reasoning step-by-step — has sparked debate among researchers and prompted a wave of explainer articles across the tech press.
But opaque recurrence is just the latest addition to a rapidly expanding lexicon. From "RAMageddon" to "neuralese," the language of AI is evolving as quickly as the technology itself. This glossary aims to provide clear, practical definitions for the terms you're most likely to encounter, whether you're building with these systems, investing in them, or simply trying to follow along in meetings.
Core concepts: from AGI to inference
Understanding AI starts with a few foundational ideas. Artificial general intelligence (AGI) remains a moving target — OpenAI's charter describes it as "highly autonomous systems that outperform humans at most economically valuable work," while Google DeepMind frames it as AI "at least as capable as humans at most cognitive tasks." Even experts disagree on the precise threshold.
Beneath AGI lies the machinery that powers today's tools. Neural networks, inspired by the human brain's interconnected pathways, form the basis of deep learning. Large language models (LLMs) like those behind ChatGPT and Claude are deep neural networks trained on billions of words to predict and generate text. Training involves feeding data to a model so it can learn patterns, while inference is the process of running that trained model to make predictions or generate responses.
Two related techniques have become central to modern AI development: fine-tuning and distillation. Fine-tuning takes a pre-trained model and further trains it on specialized data for a specific task — a common approach for startups building vertical AI tools. Distillation, meanwhile, transfers knowledge from a large "teacher" model to a smaller "student" model, which is how OpenAI reportedly developed GPT-4 Turbo. Distillation from competitors typically violates terms of service, though it's widely used internally.
The new frontier: opaque recurrence and reasoning models
The most significant recent shift in AI has been the move from simple chatbots to reasoning models that can think through problems. Chain of thought — breaking a query into intermediate steps — has been the standard approach, producing a visible trail of logic that safety researchers can audit. But Astra's opaque recurrence technique bypasses that readable trail, looping the query through the model's internal layers instead.
This approach is more efficient, allowing smaller models to perform better while using less compute. But it has a cost: fewer readable traces for oversight. The term neuralese has emerged to describe a hypothetical worst-case scenario where a model reasons entirely in opaque numerical representations. OpenAI has stated that Astra keeps its chain of thought legible and has pushed back on comparisons to neuralese, but safety researchers see opaque recurrence as a first step in that direction.
Related to this is recurrent depth, the engineering term for the same looping mechanism. Media outlets often use the two interchangeably, though "recurrent depth" emphasizes the technical method while "opaque recurrence" highlights the safety concern.
Why this matters beyond the lab
These terms aren't just academic jargon. They reflect real trade-offs that affect how AI systems are built, deployed, and regulated. The debate over opaque recurrence, for instance, is fundamentally about accountability: if we can't see how a model reaches its conclusions, how do we trust it with consequential decisions?
The vocabulary also captures broader industry trends. RAMageddon — the global shortage of memory chips driven by AI data center demand — has already forced gaming console price hikes and threatens smartphone shipments. Token throughput, a measure of how much text a model can process at once, has become an obsession for infrastructure teams, with AI researcher Andrej Karpathy even describing anxiety over idle AI subscriptions.
Meanwhile, open source models like Meta's Llama family continue to challenge the closed approaches of OpenAI and Google, fueling an ongoing debate about transparency and safety. And AI agents — tools that can autonomously perform multi-step tasks like filing expenses or writing code — are moving from concept to reality, raising new questions about oversight and reliability.
As the field evolves, so will its language. This glossary will be updated regularly to reflect new developments, whether that means decoding the next breakthrough or simply keeping pace with the industry's relentless appetite for new terminology.
This article is for informational purposes only and does not constitute financial advice. The AI market is volatile and uncertain; readers should conduct their own research before making any investment decisions.
Originally published on CoinPulseHQ: https://coinpulsehq.com/ai-glossary-terms-2026/

O sidechain do Bitcoin Liquid suspendeu as operações após a retirada de aproximadamente 4.000 Bitcoins — avaliados em cerca de US$ 320 milhões — da carteira da sua federação por atores que se apresentaram como hackers white-hat. O incidente, divulgado no domingo, levou o Liquid a desativar nós de ponte e interromper novas transações, enquanto exchanges pausavam ou se preparavam para pausar depósitos e saques de L-BTC.
A Blockstream se envolve com atores que alegam status white-hat
A Blockstream, o fornecedor de tecnologia por trás do Liquid, iniciou contato com os atores por meio de mensagens onchain assinadas. De acordo com comunicações posteriores, os indivíduos afirmaram que devolveriam a maior parte do Bitcoin assim que a vulnerabilidade no Elements — o software open source que dá suporte ao Liquid — fosse corrigida e todos os nós de rede fossem atualizados. Eles também enviaram detalhes técnicos criptografados à Blockstream, segundo a pesquisa do chefe da Galaxy Digital, Alex Thorn. Até os relatórios mais recentes, os fundos ainda não haviam sido devolvidos.

French AI lab Mistral AI announced Tuesday that it has raised €3 billion (about $3.58 billion) in a Series D round at a post-money valuation of more than €21 billion (about $24.39 billion), confirming earlier reports. The round, which Mistral called “the largest equity fundraising round ever completed by a European technology company,” was led by Samsung Electronics, with EQT-managed Scaleup Europe Fund and existing investor PSG Equity joining as co-leads.
Mistral said it will use the capital to scale its compute capacity, build infrastructure, accelerate commercial growth, and expand its international footprint. The funding also sharpens its strategic positioning: the company insists it is not building a “European ChatGPT,” and while its models have not achieved mainstream consumer adoption, it continues to define itself as an AI research lab with a focus on enterprise and government clients.
Strategic shift toward sovereign AI
The funding supports Mistral’s subtle but significant strategy shift aimed at addressing European concerns about over-dependence on the United States for critical technology. In August, Mistral unveiled tools that let customers choose which regions their AI queries are processed in, and it began hosting third-party, open-weight AI models—including Chinese ones—to strengthen its position as an AI services provider that prioritizes customer control over model selection and usage.
The company on Tuesday described its frontier research as “the foundation underpinning its infrastructure, products and sovereignty,” an indirect response to critics who interpreted its hosting of Chinese models as a pivot to becoming merely an inference provider. Mistral’s emphasis on global ambitions also counters the common misconception that its operations are confined to France. The lab now operates in 20 countries, with a go-to-market strategy focused on helping governments and corporations apply AI while maintaining control—unlike rivals such as OpenAI and Anthropic, which sell their models more broadly.
Geopolitical significance and backing
Samsung’s entry into Mistral’s cap table has the blessing of French authorities. In a post on X, French President Emmanuel Macron said the round reflected France and South Korea’s goal of “building a third way in AI.” The fact that a private funding round warranted such a statement underscores the geopolitical undertones that have surrounded Mistral—mostly to its benefit.
Amid growing demand for sovereign AI infrastructure, not being an American company has reportedly boosted Mistral’s revenue. However, the capital required to compete with leading U.S. labs is not available in France alone. With Dutch chipmaker ASML as a major partner and investor, and now Samsung, Mistral appears to have found a viable path—similar to Germany’s Aleph Alpha merging with Canada’s Cohere.
Mistral still collaborates with U.S. players, particularly Microsoft, through a strategic partnership significantly expanded in July. The Series D also attracted American investors: existing backers such as a16z, Nvidia, and Salesforce Ventures participated, joined by new backers Advent and BlackRock. Still, with Luxembourg’s sovereign fund also joining as a new backer and many European investors doubling down, Mistral’s cap table remains resolutely international—a factor that may reassure the government and enterprise customers it targets.
Disclaimer: This article is for informational purposes only and does not constitute financial advice. The cryptocurrency and AI investment markets are volatile and uncertain; readers should conduct their own research before making any investment decisions.
Originally published on CoinPulseHQ: https://coinpulsehq.com/mistral-ai-raises-3b-europe-largest-tech-funding-round/

Cognition, the startup behind the AI coding assistant Devin, has raised $2 billion at a $48 billion valuation, the company announced Tuesday. The round, led by Andreessen Horowitz, Accel, Founders Fund, General Catalyst, and Avenir, comes just four months after Cognition's previous fundraise at a $26 billion valuation — a sign that venture investors still see room for multiple major players in the AI coding market, one of the most commercially significant applications of generative AI.
The rapid doubling of Cognition's valuation suggests that the AI coding sector, far from consolidating into a single winner, is attracting capital across multiple challengers. That thesis was tested earlier this year when Cursor, a rival coding assistant, agreed to sell to SpaceX for $60 billion in April after reportedly exploring a $50 billion fundraising round.
Revenue growth and the path to scale
Cognition said that since announcing its last fundraise in May, its annualized run-rate revenue has grown from $492 million to $900 million. The company did not disclose how it calculates the run-rate figure, which typically represents a single month's revenue multiplied by 12. At the time of Cursor's funding talks in April, its annualized revenue had surpassed $2 billion, meaning Cognition currently commands a higher revenue multiple than Cursor did just before its sale.
Investors familiar with Cursor's financials said the company sold to SpaceX largely because it was severely compute-constrained — unable to secure enough server capacity to meet demand. Whether Cognition faces similar constraints is unclear, though its infrastructure costs are significant. Cognition leases an Nvidia server cluster that costs hundreds of millions of dollars annually, which could push its total cash burn to $800 million this year, according to The Information.
Like Cursor did before joining SpaceX, Cognition is training its own model based on open-source alternatives. Reducing reliance on expensive third-party models from OpenAI and Anthropic is expected to help cut costs and move the company closer to breakeven over time. The Information reported that Cognition is projected to reach $4 billion to $5 billion in annualized revenue by the end of 2026. By comparison, TechCrunch reported in the spring that Cursor was on track to surpass $6 billion by year-end.
What the funding round says about the AI coding field
The involvement of Andreessen Horowitz is particularly notable. The firm was a major backer of Cursor and profited significantly from its sale to SpaceX. Its decision to lead a round in a direct competitor suggests that investors are not treating AI coding as a zero-sum game — and that the market is large enough to support multiple companies with distinct approaches.
Founded in 2024 by math prodigy Scott Wu, Cognition has attracted a roster of blue-chip enterprise customers, including Mercedes-Benz, NASA, Goldman Sachs, and Citi. The startup's focus on autonomous coding agents — tools that can plan and execute programming tasks with minimal human oversight — differentiates it from more interactive assistants like Cursor.
The divergence in strategies between Cognition and Cursor is instructive. Cursor's model, which leaned heavily on fine-tuned versions of frontier models, proved compute-intensive and difficult to scale independently. Cognition's decision to train its own open-source-based models may offer a more sustainable path, though it carries its own risks, including the challenge of matching the raw capability of models from OpenAI and Anthropic.
For enterprise customers evaluating AI coding tools, the competitive dynamics matter. The presence of multiple well-funded players — each with different pricing, deployment models, and levels of autonomy — gives buyers tap into and options. It also raises the stakes for incumbents like GitHub Copilot, which faces pressure from both startups and the broader shift toward agentic coding workflows.
As the AI coding market matures, the key question is whether revenue growth can keep pace with the enormous capital being deployed. Cognition's run-rate growth is rapid, but so is its cash burn. The company's ability to achieve breakeven will depend on whether its proprietary models can deliver performance that justifies premium pricing — and whether it can avoid the compute bottlenecks that forced Cursor into the arms of SpaceX.
This article is for informational purposes only and does not constitute financial advice. Valuations and revenue projections in the AI sector are volatile and subject to change; readers should conduct their own research before making investment decisions.
Originally published on CoinPulseHQ: https://coinpulsehq.com/cognition-48b-valuation-ai-coding-market/

The United Kingdom's Financial Conduct Authority (FCA) has reportedly opened discussions with prediction market companies about whether to lift a ban that has barred retail investors from accessing platforms such as Polymarket and Kalshi since 2019.
According to a Friday report from The Times, the regulator is weighing whether to relax the prohibition on binary options, which include event-based contracts covering sports, politics, and weather. The FCA first imposed the ban in April 2019, when it prohibited firms from selling, marketing, or distributing binary options to retail consumers.
At the time, Christopher Woolard, the FCA's executive director of strategy and competition, described binary options as "gambling products dressed up as financial instruments." The ban was introduced after the regulator found evidence of widespread consumer harm, including significant losses among retail traders.
VPN workarounds and growing demand
The Times report suggests that many UK-based retail investors have continued to access prediction markets by using virtual private networks (VPNs) to bypass geographic restrictions. Platforms like Kalshi and Polymarket, both of which operate primarily in the United States, have seen growing volumes despite the regulatory barriers.
The potential shift comes as the prediction market industry expands rapidly. Bernstein Research estimated in April that total trading volume across the sector could reach approximately $240 billion in 2026 and potentially $1 trillion by 2030. Such figures highlight the commercial significance of the market and the pressure on regulators to adapt.
US legal battles cast a shadow
Should the FCA overturn its 2019 ban, platforms like Kalshi and Polymarket may face regulatory challenges in the UK similar to those they are currently managing in the United States. Several individual state gaming authorities have filed lawsuits against these companies over sporting event contracts, arguing that such offerings constitute unlicensed gambling.
Last week, New Jersey officials petitioned the Supreme Court to hear their case against Kalshi, a move that could ultimately clarify the jurisdictional boundaries between state and federal authorities regarding prediction markets. The outcome of that case may influence how other regulators, including the FCA, approach the sector.
Why this matters for UK investors
For UK retail investors, the FCA's review represents a potential turning point. If the ban is lifted, platforms could legally offer event-based contracts to UK users, providing new avenues for trading but also raising concerns about consumer protection. The FCA has historically taken a cautious stance on high-risk financial products, and any regulatory change would likely come with safeguards.
The regulator has not yet made a formal announcement, and the timeline for any decision remains unclear. However, the fact that the FCA is engaging directly with prediction market companies signals a willingness to reconsider its position in light of market developments and international regulatory trends.
Conclusion
The FCA's reported review of its prediction market ban marks a notable development in the evolving relationship between traditional financial regulation and emerging event-based trading platforms. While no decision has been made public, the discussions reflect broader questions about how to classify and oversee products that blend elements of gambling and investing. For now, UK retail investors must continue to rely on VPNs to access these platforms, a workaround that carries its own legal and security risks.
FAQs
Q1: What exactly is the FCA considering changing?
The FCA is reportedly reviewing its April 2019 ban on binary options for retail investors. This ban currently prevents platforms like Polymarket and Kalshi from offering event-based contracts—covering sports, politics, weather, and similar topics—to UK-based retail consumers.
Q2: Why did the FCA impose the ban in the first place?
The FCA introduced the ban in 2019 after determining that binary options were causing significant consumer harm. The regulator described them as "gambling products dressed up as financial instruments" and cited evidence of widespread losses among retail traders.
Q3: How might a lifting of the ban affect UK retail investors?
If the ban is lifted, UK retail investors could legally access prediction market platforms without needing to use VPNs. However, any regulatory change would likely include consumer protections, and platforms may still face legal challenges similar to those seen in the US, where state authorities have sued over sporting event contracts.
Originally published on CoinPulseHQ: https://coinpulsehq.com/uk-fca-weighs-lifting-prediction-markets-ban/

A Sequoia Capital está apostando ainda mais em uma startup que busca resolver um problema crescente para as empresas: garantir os agentes de IA agora responsáveis por dados corporativos sensíveis em velocidade de máquina. A gestora de investimentos liderou junto uma rodada Série A de US$ 25 milhões para a Cymphony, uma empresa com sede em Nova York e Tel Aviv, com a SMBC Fin Atlas Beyond Fund, avaliando a startup em mais de US$ 100 milhões após o investimento. A rodada segue um investimento-semente não divulgado da Sequoia feito há mais de dois anos.
A Symphony, fundada em 2024 por Shy Dekel, Idan Berkovits e Edi Gotlieb — todos ex-alunos do programa Talpiot, do serviço militar israelense — está construindo uma plataforma voltada a dar às equipes de segurança uma visão unificada de funcionários humanos e agentes de IA, incluindo os sistemas e os dados sensíveis a que eles podem acessar. A empresa afirma que atende uma lacuna crítica: agentes de IA frequentemente contornam os controles de identidade e acesso aplicados aos trabalhadores humanos, criando novos pontos de exposição que as ferramentas tradicionais de segurança não detectam.

Instinct, o assistente de IA que disparou para uma avaliação de US$ 2,5 bilhões, está dando a cada usuário um endereço de e-mail dedicado — uma medida que permite que o agente aja com mais independência ao se cadastrar em serviços, contatar empresas ou gerenciar tarefas em nome do usuário. O fundador Noah Shinn anunciou o recurso no X em 8 de setembro de 2026, enquadrando-o como "o primeiro passo para permitir que sua Instinct possua e administre suas próprias contas."
O Instinct agora atribui a cada usuário um endereço de e-mail exclusivo para que o agente de IA possa criar contas, contatar empresas e fazer acompanhamentos sem lotar a caixa de entrada pessoal do usuário. O recurso, anunciado pelo fundador Noah Shinn, está sendo liberado agora, com usuários iniciais podendo reivindicar seus endereços em mail.instinct.com.

Shipt, a plataforma de entrega no mesmo dia pertencente à Target, lançou em 9 de setembro de 2026 seu próprio assistente de compras com IA, juntando-se a uma onda de apps de entrega que disputam para incorporar IA conversacional à experiência de compra de supermercado. A nova ferramenta, chamada “Ask Shipt”, já está disponível no aplicativo da Shipt e no Shipt.com, segundo a empresa.
O Ask Shipt permite que os clientes gerem carrinhos completos, prontos para comprar, a partir de prompts em linguagem natural ou de fotos. A Shipt diz que os usuários podem solicitar coisas como “Crie um carrinho para o meu tailgate de sábado para 25 pessoas e inclua alguns itens de brunch” ou enviar uma foto de uma refeição vista em um restaurante para que o assistente identifique e adicione todos os ingredientes a um carrinho. Compradores mais atentos ao orçamento também podem pedir ideias, como uma refeição para uma noite de semana para uma família de cinco por menos de US$ 35.

O Ministério Federal das Finanças da Alemanha teria elaborado uma proposta para introduzir um imposto fixo de 25% sobre lucros de negociações com criptomoedas, uma mudança significativa em relação à política atual do país, que isenta os ganhos com cripto da tributação após um período de detenção de um ano. O rascunho, visto pelo jornal alemão Die Welt, sugere que o novo imposto se aplicaria a todos os ativos digitais adquiridos após 1º de janeiro de 2027, com o novo regime entrando em vigor em 2028.
Cláusula de anterioridade para detentores existentes
De acordo com a proposta em rascunho, o ministério planeja incluir proteções de cláusula de anterioridade. Isso significa que a criptomoeda comprada antes do prazo-limite de 1º de janeiro de 2027 continuaria a ser tratada pelas regras vigentes, permitindo que detentores de longo prazo que adquiriram ativos anteriormente ainda se beneficiem do status atual de isenção de impostos após 12 meses de detenção. Essa medida transitória busca evitar penalizar investidores que tomaram decisões com base na estrutura tributária existente.

A Apple anunciou na quarta-feira durante seu evento "Surprise and Shine" que está apresentando a Apple Reference Image, um recurso projetado para verificar se uma imagem capturada no iPhone 18 Pro é autêntica. A empresa diz que o recurso é "vital para fotojornalistas e fotógrafos" à medida que imagens geradas e editadas por IA se tornam cada vez mais difíceis de distinguir de fotografias reais.
A Apple Reference Image funciona capturando dados de sensor assinados da câmera principal no momento em que uma foto é tirada. Esses dados são então processados pelo serviço Private Cloud Compute da Apple, que gera uma visualização de "imagem inalterável" acessível no app Fotos. Essa imagem de referência funciona como um "negativo digital", permitindo que os usuários a comparem com outras versões da mesma foto para detectar quaisquer alterações ou edições.

Listen Labs, uma startup de pesquisa de mercado em IA com três anos de idade, assinou uma carta de intenções para uma rodada de US$ 125 milhões da Série C, com avaliação de US$ 1,5 bilhão, mas a rodada nunca foi concluída. Segundo diversas fontes familiarizadas com o assunto, a empresa desistiu do acordo — um movimento raro em capital de risco — para buscar negociações de aquisição com a Salesforce, que, segundo relatos, teria discutido comprar a startup por cerca de US$ 2 bilhões.
A captação de recursos, que tinha a Menlo Ventures definida para liderar, fracassou quando a Salesforce entrou na jogada. O Business Insider informou primeiro as conversas de aquisição, observando que elas não foram finalizadas e podem não resultar em um acordo. A Listen Labs, a Salesforce e a Menlo Ventures não responderam a solicitações de comentário.

Na quarta-feira, a Apple revelou uma grande reformulação do seu app Saúde junto com os novos Apple Watch Series 12 e Ultra 4, introduzindo uma guia Insights impulsionada por IA, uma pontuação diária de prontidão e uma nova métrica de “Saúde Idade” que compara os seus dados biológicos com a sua idade cronológica. O redesenho, alimentado pela Apple Intelligence, faz parte da campanha mais ampla da empresa para posicionar o iPhone e o Apple Watch como hubs centrais para o gerenciamento proativo da saúde.
A mudança mais visível é a nova guia Insights, que substitui a visualização estática de resumo por um feed dinâmico que destaca as informações mais oportunas dos seus dados de saúde. Segundo a Apple, a guia oferecerá orientações personalizadas, avaliações e sugestões contextuais — por exemplo, recomendando que um utilizador adicione mais intervalos a uma corrida pela manhã para melhorar a aptidão cardiovascular.

A Apple entrou oficialmente no mercado de telefones dobráveis na quarta-feira, 9 de setembro de 2026, apresentando o Duo em seu evento ‘Surprise and Shine’. Embora o formato do dispositivo seja uma novidade para a empresa, o salto de engenharia mais significativo pode estar escondido dentro de sua dobradiça; o diretor de Hardware, Johny Srouji, afirma que ela foi projetada e fabricada com a ajuda de IA e impressão 3D.
Srouji detalhou o processo durante a keynote, explicando que a dobradiça é um componente crítico para um dispositivo que sofre muito mais estresse do que um smartphone tradicional. Para lidar com as preocupações de durabilidade que assolaram outros dobráveis, a Apple implementou um processo de fabricação que usa inteligência artificial para garantir alinhamento quase perfeito e suavidade da superfície.

O Senado dos EUA volta a se reunir nesta semana com uma janela estreita para avançar o projeto de lei Digital Asset Market Clarity (CLARITY), uma medida que muitos na indústria de criptomoedas veem como um passo crítico para estabelecer regras federais para ativos digitais. Se a legislação não conseguir superar um obstáculo processual, o caminho para se tornar lei pode se estender até um novo Congresso, com possivelmente uma liderança política diferente, adiando qualquer resolução ao menos até 2027 — e possivelmente por muito mais tempo.
O líder da maioria no Senado, John Thune, agendou uma votação de encerramento (cloture) sobre o projeto para terça-feira, 10 de setembro. Os republicanos precisarão de pelo menos 60 votos para encerrar um filibuster, o que significa que o apoio de alguns democratas é essencial. A câmara tem menos de 36 dias legislativos restantes antes de a sessão atual terminar em janeiro de 2027, quando um Congresso recém-eleito é empossado. Com eleições de meio de mandato marcadas para novembro, o controle de ambas as câmaras está em disputa, e o resultado pode alterar fundamentalmente a trajetória do projeto de lei.

As campanhas de destilação contra os modelos Claude da Anthropic cresceram a ponto de somar centenas de milhões. Em um relatório divulgado na quinta-feira, a Anthropic disse que observou cerca de 199 milhões de interações ligadas a ataques não autorizados de destilação, distribuídas em cinco campanhas separadas que ela atribui a laboratórios de IA baseados na China, com o maior esforço individual associado à família de modelos Qwen, da Alibaba.
O que a Anthropic diz que as campanhas parecem
A destilação é uma técnica padrão de machine learning: um modelo menor é treinado com as saídas de um modelo maior para transferir capacidade de raciocínio. O ponto de atrito é o consentimento. A Anthropic diz que o encadeamento interno de raciocínio de seus modelos não é exposto aos usuários e que as campanhas encontraram maneiras de extrair esses registros mesmo assim.

O novo aplicativo de agente de IA da Meta, o Muse, subiu para a 2ª posição nas principais paradas da App Store dos EUA, atraindo mais de 83.000 downloads de iOS nos Estados Unidos desde o lançamento na terça-feira, segundo estimativas da empresa de inteligência de mercado Sensor Tower. O app atualmente é limitado a usuários dos EUA, e sua escalada nas paradas marca uma das investidas mais ambiciosas da Meta em uma IA agentic voltada ao consumidor até agora.
A conquista ocorre dias depois de a Meta ter concordado com um acordo multistate de US$ 18 bilhões sobre alegações relacionadas aos danos causados pela mídia social ao consumidor — um cenário que pode influenciar a rapidez com que os usuários adotam um produto que exige fornecer mais informações pessoais.

A OpenAI temporariamente parou de aceitar novos assinantes do seu plano Pro de US$ 200 por mês, uma medida que o líder de produto atribuiu à demanda pelo modelo mais recente da empresa estar sobrecarregando a infraestrutura por trás dele. Thibault Sottiaux, que lidera produtos essenciais, incluindo o ChatGPT e o Codex, anunciou a pausa no X, dizendo que o nível Pro coloca a maior carga nos sistemas da OpenAI de qualquer plano que ela vende.
"Queríamos dar o menor passo que nos permita continuar oferecendo o acesso mais amplo possível", escreveu Sottiaux. As inscrições para o nível mais alto agora estão desativadas, enquanto o acesso à API do ChatGPT e os planos Go e Plus, de menor custo, permanecem abertos para novos clientes.

As reclamações ao Provedor de Habitação do Reino Unido mais do que dobraram entre 2022 e o ano passado, subindo de 2.600 para pouco mais de 7.000, segundo dados citados pelo pesquisador Chris Schmitz. No mesmo período, o Consumer Financial Protection Bureau (Agência de Proteção Financeira do Consumidor) dos Estados Unidos viu o volume de reclamações crescer aproximadamente cinco vezes. Nenhuma das agências alterou sua área de atuação, seu modelo de equipe ou sua estratégia de divulgação. O que mudou, argumenta Schmitz, é a facilidade com que as pessoas agora podem registrar uma reclamação.
Schmitz está acompanhando o fenômeno como parte de uma tendência mais ampla que ele chama de enchimento agentic — o rápido aumento de solicitações, petições e protocolamentos que ocorre quando assistentes de IA tornam tarefas administrativas triviais de concluir. Seu artigo, que será apresentado no próximo mês na conferência de Ética e Sociedade da IA, examina 84 casos potenciais de enchimento em 11 jurisdições, abrangendo solicitações de assistência social, recursos judiciais oficiais e tudo o que estiver entre eles.