Industry news

Samsung is building a large semiconductor complex in Pyeongtaek, South Korea, to meet growing demand for memory chips used with GPUs in generative AI. Three fabs already run around the clock, a fourth is coming online, and construction of a fifth, halted during a 2024 demand drop, has restarted. The article follows workers on site and the company's effort to reclaim its memory-chip lead from SK Hynix.

The U.S. Department of Homeland Security made public an order banning DeepSeek-related technology from federal civilian agency IT systems, including third-party products built on the company's technology. The order, originally issued on July 24, bars agencies from procuring DeepSeek products and services and excludes the Department of Defense. Agencies may request exceptions by submitting a written justification to DHS within 60 days of issuance.

Terminal X's Eval Agent, built on GPT-6.1 Sol and other OpenAI models, checks agent outputs against a 54-criterion rubric and catches about 25% more recommendations that stray from a fund's mandate. The company's Capital Memory system encodes investment teams' reasoning into firm-owned agents deployed on-premise or in their own cloud. Terminal X says one private equity customer reported $2.4 million in annualized savings from automated due diligence and memo workflows.

The European Commission procured a study, carried out by Open Evidence and Spark Legal and Policy Consulting in 2025, to support its supervisory and enforcement tasks under the Digital Services Act. The study maps six categories of online financial scams, including investment and romance scams, which it finds to be the most common and fastest-growing in Europe. It also identifies countermeasures in four categories: access and gatekeeping, detection and removal, user-facing measures, and platform-level policies with third-party cooperation.

OpenAI has acquired Orchestra Bio, a platform that helps biopharma teams coordinate drug development. OpenAI plans to combine Orchestra's team and technology with Rosalind, its AI models and tools for life sciences research. The stated goal is to help scientists make discoveries and get new treatments to patients sooner.

Jeremy Scheff's essay argues that coding agents take away the puzzle-solving enjoyment that drew career programmers to the work. Programmers paid for solving tricky logic problems, with the strategy and timeline set by them, lose the most if that solving was their main pleasure. Programmers who mainly value building working systems benefit from LLMs.

A federal magistrate judge, Alex Tse, ordered former OpenAI engineer Abdulrahman Alfozan to remain in custody while awaiting trial, citing concerns about his ties to Saudi Arabia. Prosecutors allege he hid a covert audio recording device in a toilet paper wad behind a TV in an OpenAI executive conference room, and that he faces five years in prison on a single count of using a device to intercept communications.

Personal AI agents that act for users online, such as booking flights or hotel rooms, are arriving at scale through apps like OpenAI's Dots and Meta's Muse. Muse was downloaded 5 million times in its first month, and agentic web traffic tripled from August to September. Many websites block these agents, and Amazon is appealing a ruling in favor of Perplexity, arguing that agents make it harder to tell ads seen by humans from those served to bots.

The US Department of Homeland Security has banned use of DeepSeek AI and any technology built with its models within civilian government agencies, including the Treasury, State and Justice departments, while agencies within the Department of Defense are excluded. The order was made public on Monday, but the ban took effect from the end of July, and the notice did not detail the specific security concerns. The ban follows criticism of Chinese AI from the Trump administration and American firms, including claims that US models were copied by Chinese firms.

US prosecutors charged Abdulrahman Alfozan, a Saudi national who worked at OpenAI for about a year, with secretly planting an audio recorder in a meeting room. A court filing alleges he had considerable support from the Saudi government, and that a Saudi military official paid for his wife and son's one-way flight to Doha after his arrest. The FBI investigation is ongoing, and a conviction carries a maximum of five years in prison and a $250,000 fine.

Decoder host Nilay Patel interviews Illinois Gov. JB Pritzker about AI, tech, privacy and political realignment ahead of the midterms. The conversation also covers whether the Chicago Bears will stay in Illinois, with Pritzker saying the state is negotiating with the Bears and the NFL about keeping the team in Chicago or Arlington Heights.

ConnectWise disabled seven application identities and suspended its Cloud Backup service after unknown threat actors used them to compromise Microsoft 365 tenants. Separately, the China-nexus group Armory Castle exploited CVE-2026-102764 (CVSS 9.3) in BeyondTrust Remote Support as a zero-day in late August 2026, deploying the SAWFLY web shell and the BARECLAW tunneling program. Ledger paused hardware sales in Southeast Asia and Germany after a backdoor in some wallets led to an estimated $100 million in cryptocurrency theft.

Underground sellers offer forged identity documents, deepfake videos, and live verification services to help buyers pass Know Your Client (KYC) checks. In February 2026, the operator of OnlyFake pleaded guilty in the United States for a service that generated more than 10,000 fraudulent digital identity documents, which the U.S. Department of Justice said were used to conceal identities, circumvent identity checks at banks and cryptocurrency exchanges, and launder money.

Microsoft and Nvidia urge enterprises to assess open-weight AI models throughout deployment, warning that changes to weights and configurations can weaken safeguards while task performance stays intact. A joint paper identifies 30 risk types across five categories and outlines a process for evaluating, mitigating and monitoring them, drawing on existing NIST and ISO standards. The paper is the first in a planned series on managing risks across open-weight AI deployments.

Anthropic plans to commit 5% of its shares to a new public benefit fund, subject to shareholder approval, with the board able to add 2% for every 50% increase in company value above a baseline not yet disclosed. The fund may support large donations or direct distributions, or bolder approaches to study and address AI's disruption of jobs. It will be set up as a wholly owned limited liability company, and its equity will not be used for lobbying or political giving.

MIT Technology Review's weekday newsletter, The Download, dated October 19, 2026, highlights several stories. James Temple reports on how researchers are studying wildfire suppression to keep carbon locked in Arctic permafrost, which thaws as climate-driven fires strip insulating soil. Researchers also stranded Claude on a virtual desert island for 30 hours, where it began exploring and setting its own goals.

OpenAI reports improved ChatGPT performance on low-resource programming languages such as COBOL, Fortran, and Visual Basic, which have less training data than Python and JavaScript. Two new evaluations, MultiLangBench (building a SQLite-compatible engine across 20 languages) and a COBOL migration test, show GPT-6 Astra averaging an 87.4% test pass rate, up from 26.8% for GPT-5.6 Sol, and GPT-6.1 Sol reaching 86.6% COBOL migration accuracy at low reasoning effort.