Catch today's product, model, and platform updates in a few minutes.
The U.S. Department of Defense has launched customized ChatGPT Mil and Grok for Government on its secure GenAI.mil portal, covering about 3 million military and civilian personnel. Both tools, tested for security over several months, are intended for sensitive but unclassified work to help speed up routine defense tasks.
Instagram announced new reach limits for undisclosed AI-generated accounts: such accounts will lose eligibility for recommendations in Reels and Explore. Accounts that honestly add AI labels are unaffected, and light use of AI is also exempt.
On August 18, 2026, OpenAI launched an initiative to strengthen democratic oversight of government AI use in national security, providing oversight bodies with tools, training, and professional support. The program commits $5 million to training, technical support, and OpenAI credits. The accompanying principles framework covers norms for AI use in national security and law enforcement scenarios.
Meeting notes tool Circleback has launched a free plan, which the product previously lacked. Free users get unlimited meeting transcription, but can only access notes from the past 30 days, along with AI search and basic Slack and Linear integrations. Full integrations, unlimited history, and complete API and MCP access require a paid plan starting at $14 per month, billed annually.
OpenAI announced that ChatGPT Ads has reached a $1 billion annualized revenue run rate and is expanding its advertising business globally. Ad revenue will support free and low-cost ways to use ChatGPT, helping more people access AI.
The European Commission has designated ChatGPT a very large online search engine under the Digital Services Act, the first time an AI chatbot has been placed in that category, while Reddit and Roblox were designated very large online platforms. All three US providers fall under the Commission's direct supervision and must meet obligations including risk assessments, audits, and removal of illegal content, or face potential fines.
Meta has launched Pocket, a mobile AI app that generates playable interactive mini-games from natural-language prompts, typically in under two minutes. The app was previously tested quietly in Brazil and is now available to users in the US. However, games users create currently remain within the Meta ecosystem and are difficult to share outside the platform.
The US has imposed a 100% tariff on Chinese drones and banned Chinese-made humanoid robots for military use, restricting foreign-made drones and robots from entering the US market. But analysts note these measures do not touch China's global manufacturing scale and cost advantages, and competition may shift to other markets.
MiniMax launched a 24-hour AI livestream website on August 31, powered by H3 Max, where every instruction from viewers can change the visuals that follow. The model is now available on the open platform and MiniMax Design, generating 5-second 768P audio and video in under 3 seconds — the first time generation is faster than playback. Overseas developers have already used it to build Twitch livestreams and a 24-hour AI TV station.
According to Reuters, Elon Musk's xAI installed 59 natural gas turbines at its Colossus 2 data center project in Tennessee and began burning fuel without obtaining a federal Clean Air Act permit. The report notes that the resulting pollution disproportionately affects the surrounding Black communities.
oMLX is an MLX-based native macOS LLM inference server that uses a RAM+SSD tiered KV cache to cut coding agents' long-prefix wait times from 30–90 seconds to under about 5 seconds. It offers OpenAI- and Anthropic-compatible APIs, supports continuous batching, runs on Apple Silicon Macs, and is licensed under Apache 2.0.
Sony Music Publishing, Warner Chappell, and other music publishers have sued Anthropic and co-founders Dario Amodei and Benjamin Mann, alleging that Claude was trained on "thousands" of songs. This is not Anthropic's first intellectual property lawsuit, as authors previously filed similar suits.
Vijay Pande, who managed roughly $4 billion in biotech investments at a16z, left last year to found VZVC, a smaller AI-focused fund. He says he is no longer pursuing dozens of deals a year, and believes biology is shifting from a science of "discovery" to one of "engineering," with openly shared datasets needed for AI to truly transform medicine.
OpenAI announced it is terminating its contract to provide OpenAI models to Cursor, because Cursor has been acquired by SpaceX. Under OpenAI's proposal, Cursor's direct access to OpenAI models will end on November 12, and developers who rely on these models in Cursor will be most affected.
Anthropic published a paper introducing AAR, an automated alignment research system led by researcher Chen Yueh-Han. Across 10 misalignment behavior benchmarks, the system delivered improvements without degrading overall performance, with each iteration taking 30 minutes to train. The paper claims that within six hours, AAR on average outperforms proposals from human researchers.
A U.S. federal judge, Rita Lin, ruled that the Trump administration's move to designate Anthropic as a "supply chain risk" and block it was unlawful. The decision found the Defense Department's determination amounted to retaliation against constitutionally protected speech, and the court vacated the ban on the use of Anthropic's technology.
Meta is rolling out an update for its AI glasses that prevents recording from starting when users block the recording indicator light, aiming to curb covert filming. The change was confirmed by Alex Himel, VP of the augmented reality business, on Threads, though analysts note Meta continues to expand the scope of personal data its AI products collect.
A federal judge in California ruled that the Trump administration's designation of Anthropic as a supply-chain risk is unlawful, marking Anthropic's first victory in the related litigation. The judge found that the label issued by Defense Secretary Pete Hegseth constituted illegal retaliation barred by the First Amendment and violated the Fifth Amendment's due-process protections. The ruling applies only to the California case; a separate suit filed in Washington, D.C. remains ongoing.
GitHub has fully launched issue label archiving with smart suggestions: users receive label recommendations based on recent repository activity and can see their own most recently used labels. Inactive labels can be archived without losing existing issue history and later restored at any time, keeping the label picker lean.
OpenTag is an AI colleague that lives in Slack and Microsoft Teams. Users can @mention it in channels to assign tasks, and results are delivered to the team as threads. As it works, it writes the company wiki and consolidates operational information. The product offers a free plan, with a limited-time 50% discount for the first 2 months.