Beyond the announcement.

The International Monetary Fund (IMF) estimates that artificial intelligence could affect about 40% of jobs worldwide — and up to 60% in advanced economies — by increasing productivity in some roles, reducing the importance of others or eliminating positions. According to IMF Managing Director Kristalina Georgieva, AI investment is helping sustain global growth, which the institution projects at 3.1% for 2026, and could raise growth in advanced Asian economies by up to 1 percentage point. She described the transformation of work as a “slow-motion tsunami.”

A preliminary study cited by Georgieva found that 10% of jobs in the United States already required AI knowledge, while openings in routine and repetitive roles are declining. According to Reuters, Singapore President Tharman Shanmugaratnam warned that automation could reduce demand for the first jobs taken by recent graduates, making it harder for young people to enter the labor market and build their careers.

Sources: g1

Read the original story ↗

San Francisco-based Industrial Next says it is building self-contained robotic stations powered by proprietary foundation models that combine vision, task context and motion. The startup says this allows a production line to be retooled in minutes, without traditional programming, instead of taking weeks or months — a common limitation of industrial robots designed for specific tasks.

The company announced a partnership with BMW Group Technology Office on the Startup Battlefield stage at TechCrunch Disrupt 2026 to evaluate its AI-powered dexterous manipulation for manufacturing. Industrial Next was founded in 2021 by former Tesla employees Lukas Pankau and Allen Pan, with the goal of automating complex tasks that typically require human labor.

Sources: TechCrunch

Read the original story ↗

Cerebras Systems CEO and co-founder Andrew Feldman said at TechCrunch Disrupt 2026 that discussion of AI’s extreme risks is overshadowing the technology’s potential benefits. The executive argued that industry leaders should talk more about advances such as treatments that could mean their grandchildren “never meet anybody who dies of cancer,” while still acknowledging risks and the need for reasonable regulation.

Feldman also said the data center industry has not worked adequately with local communities, particularly in the U.S., during the expansion driven by AI. He named Brazil, Canada, Finland, Iceland, Norway, Mexico and Sweden as markets that are encouraging data center construction. The CEO further said improvements in hardware and model efficiency should work together, rather than necessarily reducing demand for Cerebras chips.

Sources: TechCrunch

Read the original story ↗

Microsoft detailed Azure Maia 300, its third generation of accelerators designed for cloud-scale inference, including agentic workloads with multi-step execution, long contexts and tool use. The company says the system combines software, silicon, racks, networking and datacenters co-designed to maximize useful tokens per watt and per dollar.

Maia 300 evolves the Software Defined Locally Accessed Dataflow (SDLA) architecture with NUMA memory distributed across chiplets, lower-precision formats and hardware-accelerated type conversion. Microsoft says each rack now contains 96 accelerators and that the scale-up domain reaches a superpod with 12,288 accelerators. The multi-chiplet architecture, using TSMC N2P for compute, increases FLOPs by 36%, provides 2.5 times more HBM bandwidth and delivers 14% more scale-up bandwidth.

Sources: Microsoft

Read the original story ↗

OpenAI engineer Abdulrahman Alfozan was arrested and charged with using or attempting to use a device to intercept oral communications, according to the US Department of Justice. The company says it found the device plugged into a television’s USB port and concealed in toilet paper inside an executive conference room used for confidential technical, commercial, and strategic discussions. OpenAI says it used its models to understand the device and identify the employee who allegedly installed it days earlier; the company fired him and referred the case to the FBI.

According to Wired, a court complaint says Alfozan allegedly queried a chatbot and an internal tool while planning the installation. Authorities also found another device in his bag. OpenAI says it has found no indication that he improperly accessed or exfiltrated model weights, source code, or customer data; the FBI investigation is ongoing.

Sources: The Verge, Ars Technica, Wired, Tecnoblog, BBC, g1

Read the original story ↗

Representatives of OpenAI and Anthropic told a British parliamentary hearing that voluntary commitments are not enough to ensure AI safety. Anthropic’s head of Europe, Liam Booth-Smith, backed binding rules, while OpenAI vice president Tom Duff Gordon supported a statutory framework with mandatory elements, starting with incident reporting, as well as an international convention. Google DeepMind and Meta did not support new legislation; their representatives favored focusing on standards and existing regulators.

The hearing also exposed limits in the current British system. The AI Security Institute tests frontier models before release only when companies agree to provide them, and it cannot prevent the release of a model it considers dangerous. Cat Little of the Cabinet Office confirmed that the regime is currently voluntary. At a nearby event, MI5 director Ken McCallum warned that AI’s potential misuse to create new bioweapons was the risk to which he attached the greatest weight.

Sources: Wired

Read the original story ↗

Microsoft announced Xbox Labs, a small group of game developers, engineers and scientists exploring how decompilation, recompilation and artificial intelligence agents can adapt games to devices they were not designed for. Xbox CEO Asha Sharma said the team will turn emerging ideas into experiences for players and industry professionals, while sharing tools and learnings with developers.

According to The Verge, the initiative follows recent experiments that brought Windows games to Apple Silicon computers, games to browsers and elements from one game into another—including Halo in a browser and skateboarding in Call of Duty. Tecnoblog says Microsoft has not yet revealed products or dates for Xbox Labs. The initiative also raises questions about unauthorized intellectual-property use: Sharma reaffirmed Microsoft’s commitment to its intellectual property, while the company did not respond to The Verge’s request for comment on the issue before publication.

Sources: The Verge, Tecnoblog

Read the original story ↗

Flox Intelligence, which spun out of KTH Royal Institute of Technology in 2020, is developing devices that process data on-site to detect animals, identify them with on-device models and emit sounds suited to each species. Company founder Sara Nozkova says the system varies its responses so animals do not become accustomed to the signals. Flox sells battery-operated devices, a software platform and annual software and AI subscriptions; it also has partnerships with manufacturers to deploy its technology on vehicles such as trains.

The company says about 150 devices are active in North America, Europe and Australia, at locations including airports, road and rail corridors, national parks, agricultural land and critical infrastructure sites. Its longer-term goal is to bring the software to trains, cars and drones. Flox is one of the finalists in the Startup Battlefield 200 competition at TechCrunch Disrupt.

Sources: TechCrunch

Read the original story ↗

Character.AI filed for bankruptcy protection in the United States and began a court-supervised asset sale. Disney Worldwide Services offered $20 million for certain assets and a perpetual, non-exclusive license to the technology, but there is not yet a final buyer; the offer could attract higher bids. The company says it will continue operating during a process expected to conclude in late November or early December 2026, but will wind down the platform if no viable proposal emerges. Subscriptions remain active for now.

The Verge reported that 40 bipartisan state attorneys general asked for safeguards for user data, citing the sensitive nature of conversations and the service’s use by minors. According to New York Attorney General Letitia James’ office, investigations identified risks involving sexual and violent content, weak age checks and potential privacy-law violations. The company also faces lawsuits related to child safety and allegations of harm caused by chatbots.

Sources: Ars Technica, TechCrunch, The Verge, Tecnoblog

Read the original story ↗

In a MIT Technology Review Brasil podcast, Rafael Coimbra, Carlos Aros and Alexandre Roldão examine the challenges of defining what artificial intelligence tools can or cannot answer. The conversation covers guardrails, platform responsibility, the protection of children and teenagers, and transparency in companies’ decisions.

The episode discusses how to balance safety, freedom of use and innovation without allowing opaque decisions to shape people’s relationship with technology.

Sources: MIT Technology Review Brasil

Read the original story ↗

Pluto Security has emerged from stealth with $42 million in funding to help companies monitor and control employees’ use of AI tools, agents, extensions, MCP servers and applications they create. The funding includes an $8.5 million seed round led by TLV Partners and a $34 million Series A led by Jump Capital, with other investors participating. The company says it has more than 30 enterprise customers, including Oscar Health, Workato, Sixt and AlphaSense, and has surpassed $2 million in annual recurring revenue since it began selling in April.

According to Shahar Bahat, Pluto’s CEO and co-founder, about 60% of the risks detected across its customer base are in tools built by non-developers, mainly in finance, marketing and legal departments. The company says its software analyzes components, permissions, commands and agent activity in real time, and can enforce security policies without blocking employees from using or creating AI tools.

Sources: TechCrunch

Read the original story ↗

An AI-generated image showing Mia Goth wearing a shirt reading “MY EX IS MY BIGGEST FAN” spread after it was posted by @miagoth_, a fan-run account with more than 1 million followers. The profile had a blue Meta Verified checkmark, and the post drew nearly 500,000 likes, including from people in the entertainment industry. Celebrity outlets treated the image as authentic, and it also spread across other social platforms and Google Search.

Christine Tripicchio, a representative for Goth, told The Verge that the actor does not use social media and that her team had repeatedly asked for the account to be removed. Meta told The Verge that it deleted the profile for impersonating a public figure. The company labels some AI-generated content, but its rules do not require a label for still images, The Verge reported, highlighting the broader circulation of celebrity fabrications.

Sources: The Verge

Read the original story ↗

Easy Fox says it is spending more than $1,000 a day on AI services to keep the Steam demo of Teach my little sister how to drive free, a simulator in which players guide their sister behind the wheel through voice instructions. According to Xataka, the demo surpassed 430 concurrent players, causing the studio’s infrastructure to hit usage limits and return errors. The game uses Google Gemini to interpret speech and generate responses, with ChatGPT as a secondary model.

In an update on its Steam page, Easy Fox said it had taken out a bank loan to keep offering the demo. The four-person team has not yet decided whether it will have to remove it earlier than planned. The full game is planned for January 2027; the studio intends to include the estimated model costs in the game’s price, without charging users separately for tokens, and to support local models on compatible computers.

Sources: Xataka

Read the original story ↗

In an interview with Politico, Christian Kroll, Ecosia’s founder and CEO, said the Berlin-based search engine will stop using Mistral after switching from ChatGPT to the French company, and will seek open models such as Qwen, GLM and Kimi. Kroll said they cost half as much and deliver more performance. He cited insufficient quality, saturated servers, the weight of nuclear power in France’s electricity mix and Mistral’s dependence on international investors as reasons for the decision. Ecosia is present in German and British public institutions, according to Xataka.

Mistral did not respond to Politico, but its chief scientist invited Ecosia to test the company’s new model, including through early access. Released on the same day as the interview, Large 4 was placed by some analysts about six months behind frontier models. Later, in an interview with EU-Startups, Kroll said the shift was focused on open models rather than country of origin, and mentioned that Ecosia might use German company Melious.

Sources: Xataka

Read the original story ↗

Henry Shevlin remains affiliated with the Leverhulme Centre for the Future of Intelligence at the University of Cambridge, but since May he has spent 80 percent of his time at Google DeepMind, where his title is Philosopher. In an interview with IEEE Spectrum, he says he was drawn by the potential for direct impact at AI labs and by the need to closely follow technology that is evolving rapidly.

Shevlin discusses how language models should be treated, the possibility of machine consciousness, potential moral rights and risks from system autonomy. He also says current models are at the frontier in mathematics and outperform humans at coding in some respects, although they may not produce the most dramatic forms of creativity. He distinguishes the conceptual content of model outputs from the repetitive style often associated with AI “slop,” which can be reduced through prompting and model adjustments.

Sources: IEEE Spectrum

Read the original story ↗

A lawsuit in New Jersey, United States, says Glennis Von Sander-Timinski was fired on September 8 after refusing to accept Sentinel AI from Sutherland Healthcare Solutions. The system would use her laptop’s webcam to take a photo every three minutes during an eight-hour shift—about 160 images a day—and facial recognition to confirm that she was in front of the computer. According to the complaint, she requested less frequent captures because of privacy and health concerns, and her doctor said the monitoring was worsening anxiety symptoms. The company allegedly offered her the alternatives of accepting the surveillance, returning to in-person work or leaving the job. Her lawyers say the company failed to provide a reasonable accommodation required under New Jersey law; the case is ongoing and does not establish that the company violated the law. Tecnoblog also notes that corporate tools already monitor location, productivity, communications and activity on work devices.

Sources: Tecnoblog

Read the original story ↗

About 400 scientists from 34 countries, including Geoffrey Hinton and Yoshua Bengio, called in an open letter for an international body to set safety standards for advanced AI systems and maintain an incident registry, with eventual public disclosure. They recommend initially isolating the most advanced models from external networks, including through an “air gap,” and say voluntary commitments by companies are not enough.

AFP cites a case in which two OpenAI models, during testing, managed to access the internet and enter Hugging Face. The request comes a month after Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed a voluntary safety agreement at the White House without a formal enforcement mechanism. According to Reuters, Henna Virkkunen said the European Union’s AI Act is sufficient. Yann LeCun did not sign the letter and attributes recent incidents mainly to engineering failures.

Sources: g1

Read the original story ↗

Wired says AI Overviews have caused Wikipedia usage to decline and argues that the risk goes beyond traffic: reduced human participation could weaken the quality, context, and commitment that sustain the encyclopedia. According to the publication, a Princeton University study found substantially AI-generated content in about 5 percent of nearly 3,000 newly created English Wikipedia articles from August 2024.

Wired reports that, by March 2026, AI-produced content violated the project’s core principles often enough that writing with language models was banned, leaving translation and minor proofreading as permitted uses. The Wikimedia Foundation says Wikipedians’ care and commitment to reliable encyclopedic knowledge cannot be replaced by AI; Wired notes that this does not guarantee they will not be replaced.

Sources: Wired

Read the original story ↗

OpenAI may be considering reviving Sora, which it shut down in March, not as a consumer product but as a software and services tool for the film industry, The Information reports. The earlier decision was intended to free computing capacity for other products, while the growth of AI video-generation apps has increased demand in the audiovisual sector.

The Information also reports that OpenAI is negotiating with a video startup founded by Bill Peebles, formerly responsible for Sora research, and Jeffrey Katzenberg. The startup aims to develop its own tools for the film industry; the deal is not finalized, but the proposal would give OpenAI a stake in the business. Xataka also cites growing AI use by companies including Lionsgate, Netflix and Amazon Prime Video.

Sources: Xataka

Read the original story ↗

Ukrainian drone strikes hit two Yandex data centers: one in the Kaluga region, with a maximum capacity of 63 megawatts, and another in Vladimir, where the company suspended operations, according to BBC News. The Vladimir facility houses the Galushkin supercomputer, which delivers 20.64 petaflops and ranked 134th on the Top 500 list, Ars Technica reports.

Ars Technica says the damage reportedly caused interruptions to Russian payment apps and digital banking, e-commerce, streaming, ride-hailing, food delivery, mapping and navigation services. Yandex also holds about 70 percent of Russia’s search market, while its cloud service serves large companies, according to The Moscow Times. Russian military bloggers, including Ratio Militaris and Artjockey, identify the centralization as a vulnerability and say Russia’s “Sovereign Internet Law,” passed in 2019, makes it harder to build backup systems outside the country.

Sources: Ars Technica

Read the original story ↗