Daily at 8AM KST · Summaries and takeaways from 10 AI articles
We cross-check industry press like TechCrunch and The Decoder, official announcements from OpenAI and Google DeepMind, and community signal from Hacker News.The point isn't what happened but why it matters, tied into one read on the day.
93 days running
910 stories · 266 terms explained · since 2026-07-23
·what we picked →
2026-10-23
· ~9 min read
· Past edition
10 picked from 93 candidates · ordered by significance
Today's Insight
AI companies scale money and influence first, while Meta fixed its data safeguards only after outside pressure
OpenAI is negotiating to raise up to $30 billion at a $1.25 trillion valuation, according to CNBC, the U.S. business news network. If it closes at that level, OpenAI becomes the most valuable private company on record. Anthropic raised $65 billion at a $965 billion valuation in May, and both companies filed confidential paperwork in June.
Continue reading (3)Show less
Influence in Washington grew alongside the money. AI chip maker Nvidia, Anthropic and OpenAI spent $2.2 million, $2.1 million and $1.5 million on lobbying between July and September, each a record quarter. OpenAI did not top $1 million in a quarter until this year.
On the revenue side, OpenAI added 26 countries to ChatGPT ads, for a total of 89. Ads appear only on the free and low-cost Go plans. Anthropic launched Claude Duet, an for professionals, and put user approval in front of steps such as payments.
Safeguards for user data came after. Meta removed the word train from its business terms, but Matt Topic, an attorney at Loevy & Loevy, says develop is broad enough to include training. Meta's AI agent Muse had its shared-document addresses changed only after security researchers reported they were easy to find.
Signal to watch
The first signal is whether publishers keep demanding changes after Meta's new business terms take effect on October 30.
Get it in your inbox every day
Daily headlines, a weekly synthesis on Sundays, and a monthly report. Sent at 8AM KST, the evening before in the US.
Free · no ads · one-click unsubscribe
Check your inbox
Find this subject in your inbox and press the link to start your subscription.
Confirm your AI News Briefing subscription
It usually arrives within a minute
If you don't see it, check your spam folder too
The link is valid for 7 days
01TechCrunch AI
Press
Covered by 5 more outlets
·2026-10-22·~40s read
Business & Funding
OpenAI is negotiating to raise up to $30 billion at a $1.25 trillion pre-money valuation, TechCrunch reported, citing CNBC. Coatue is leading the round and investors from the United Arab Emirates are taking part. CNBC reported that the UAE money could be tied to a deal giving OpenAI more computing capacity. Terms are still under negotiation and could change, and OpenAI declined to comment.
Why it matters
Why It Matters
TechCrunch noted that if the round closes at this level, OpenAI becomes the most valuable private company on record and sets a higher bar for its eventual . Its March round valued the company at $852 billion post-money. OpenAI and Anthropic both filed confidential IPO paperwork in June. OpenAI's high price could also shape how investors view Anthropic's IPO.
$1.25TOpenAI's reported pre-money valuation
A valuation that stood at $852 billion after the March round would climb to a private-company record.
OpenAI made its Decisions API available to all customers on Thursday, October 22. It is a kind of model that picks quickly among options and reports its confidence, a category created by TypeSafe's Jev. OpenAI said it did not build a purpose-made model but reconfigured ChatGPT Luna to run faster and cheaper, cutting response times from one to three seconds to 100 to 180 milliseconds. The API was first announced at OpenAI's developer day on September 29, and the public version adds voice and visual modalities.
Why it matters
Why It Matters
On the leaderboards, OpenAI trails. Its decision model ranks eighth on the AIM System One Index and fourteenth on the Artificial Analysis Decision Index, while the latest Jev ranks seventh and third. TypeSafe is valued at $7.6 billion and its models are the most queried on OpenRouter. OpenAI argues its model is better calibrated on image classification and wants to cut latency to 60 milliseconds so it can be called on every frame of a video. TypeSafe's CEO tells users to distrust benchmarks and test on their own work.
Speak, which builds AI language tutors, said the time between actions dropped from 1.4 seconds to 0.13 seconds with this API.
What to do now
If you use a large model for classification or routing, run the Decisions API and Jev side by side on the same inputs and measure latency and cost yourself.
OpenAI, Anthropic and Nvidia each had their highest-spending lobbying quarter between July and September, federal disclosures show. Nvidia spent the most at $2.2 million, followed by Anthropic at $2.1 million and OpenAI at $1.5 million. The sums are not unusual next to long-established Big Tech players in Washington. Microsoft has spent at least $2 million a quarter since 2021, and Google regularly spends more than $3 million.
Why it matters
Why It Matters
OpenAI did not spend $1 million in a single quarter until this year, and did not top half a million until the end of 2024. OpenAI's filings list only high-level topics, such as cybersecurity evaluations for frontier models, AI safety oversight frameworks and antitrust protections for information sharing. Anthropic listed AI provisions in the Kids Online Safety Act and export-control provisions in the AI Overwatch Act among its targets. The Verge expects AI lobbying to keep growing whichever party controls Congress next year, with Democrats expected to win back some control after the midterms.
Anthropic launched Claude Duet, an app meant to help professionals and small businesses automate and manage their work. Duet runs on Claude Opus 5.5 and connects to more than 6,000 connectors and plug-ins in the Claude Marketplace. It works on its own virtual machine while users do other things and comes back when it needs approval. It builds daily briefings and dashboards, suggests actions proactively, and can make online purchases through Stripe's Link or Shopify's Shop Pay.
Why it matters
Why It Matters
Anthropic said Duet is not good at errands or booking tickets and cannot show its browser use as it happens. That separates it from personal agents such as Instinct and narrows it to work automation. Sensitive steps such as payments require user approval, Duet will not trade securities or move money between accounts, and it declines requests for credit card and government ID numbers by default. Meta Muse and OpenAI Dots are also leaning toward business use.
OpenAI is adding ChatGPT ads in 26 more countries, bringing the total to 89. The new markets include 14 countries in Latin America, among them Argentina, Chile, Colombia and Peru, plus Nigeria, South Africa, Serbia, Ukraine, Cambodia, Bangladesh and Pakistan. ChatGPT has more than 1.2 billion weekly users, and over 100,000 advertisers have run campaigns since the U.S. test began early this year. Annualized ad revenue reached $1 billion in August.
Why it matters
Why It Matters
Ads appear only for free-plan and low-cost Go users, not for Plus, Pro or Enterprise subscribers. OpenAI says it does not share conversations with advertisers, ads do not influence ChatGPT's answers, and ads carry a distinct label. But in September it introduced AI tools for advertisers and began testing a sponsored feature with some U.S. advertisers, and it says it plans to develop ad types that blend into conversation flows. When those conversational ad types launch, the thing to check is whether they still carry the distinct label.
OpenAI, working with PerturbAI and Scripps Research, used thousands of AI s to analyze biological data from millions of mouse cells in 32 minutes. The agents ran on GPT-Rosalind and worked through a whole-brain CRISPR atlas, a map of how changing individual genes affects different brain cell types. Each agent took one gene perturbation, analyzed how cells responded and searched the scientific literature for explanations. The team estimates the same work done by hand would take several months to a year.
Why it matters
Why It Matters
OpenAI framed the findings as leads that help researchers decide which mechanisms to test in the lab first, not as confirmed results. The agents flagged unexpected differences between two disease-related genes and a possible link between cholesterol regulation and neurodegeneration. In blind comparisons, human experts preferred reports from the distributed agent system over a single agent 75 percent of the time. The paper is a bioRxiv preprint, so it has not been peer reviewed.
Meta removed language suggesting it could freely train its AI on news sites from its business terms, but legal experts quoted by Nieman Lab say the change does not go far enough. The terms, which take effect October 30, originally said Meta could use content from websites associated with a business account to develop, research and train its products. After fierce backlash from the publishing industry, Meta dropped the word train. James Grimmelmann, a professor at Cornell Law School, told Nieman Lab that the phrase associated with your products or services is untouched and would still let Meta scrape entire sites.
Why it matters
Why It Matters
Matt Topic, an attorney at Loevy & Loevy, said the revised terms still let Meta use scraped content to develop any business product, a term that is not defined, and that develop is broad enough to include train. Of the original terms, legal experts said Meta could argue its commercial terms supersede blocking instructions such as . Meta says content owners can use robots.txt to signal preferences and that the update gives it no new rights that override them.
What to do now
If your site is linked to a Meta business account, check the content-use clause in the new terms before they take effect on October 30.
Meta has re-enabled sharing in its Muse AI after security researchers reported that public conversation links could be easy to find. The researchers said they found Muse pages such as a marketing company's internal revenue review and a small business owner's loan application. Researchers at Claroc reported that the web addresses for shared Muse conversations were named after a conversation's title and written into public DNS. According to Ars Technica, Meta paused the sharing feature on Tuesday.
Why it matters
Why It Matters
Meta's David Singleton said on X that the problem was with the URLs of shared Muse artifacts, which are things like documents and images inside Muse, and not a flaw in chat sharing. Artifacts shared before the fix have new URLs, and anyone visiting an old link sees a note asking them to get the new one from the sharer. The Verge reported earlier that 10 million people have already set up Muse agents.
What to do now
If you ever shared a document or image from Muse, find where you pasted the old link and replace it with the new address.
Anthropic released a new version of Claude Security, its AI code security analysis tool, to Claude Enterprise customers. It moves from one-time scans to continuous monitoring that tracks changes in connected code repositories and rechecks for . Several repositories can be linked into one project to catch vulnerabilities that appear only when services interact. According to AI Times, thousands of organizations have used the tool since its April public beta and found more than one million vulnerabilities.
Why it matters
Why It Matters
Vulnerabilities that get fixed are marked resolved automatically, wherever the fix was made, and reopened if the same issue reappears. Teams can enter deployment methods, security policies and operating environments, and the tool builds a threat model from the code, filters out issues other controls already handle and ranks findings with attack path diagrams. It uses Anthropic's models including Claude Mythos, and customers do not need to join a separate cyber verification program.
What to do now
Claude Enterprise admins can turn the feature on and set organizational access in the admin console. Free credits worth $25,000 a month for two months apply to usage through December 31.
Google DeepMind and Google Research, with the World Resources Institute (WRI) and the International Institute for Applied Systems Analysis (IIASA), released Forest Typology (ForTy) 2020, a global map that classifies every 10-by-10-meter patch of land into one of six types. Unlike maps that only show whether trees are present, it separates , naturally regenerating forest, planted forest, plantation forest, tree crops such as oil palm, rubber and cocoa, and other land. An early yearly time series covering 2017 to 2025 is included. Validated against more than 9,000 randomly chosen sites, it identified the correct forest type 90% of the time and forest versus non-forest 95% of the time.
Why it matters
Why It Matters
The EU deforestation regulation (EUDR) bars products such as coffee, cocoa and palm oil from land that was forest in 2020 and has since been converted to agriculture, and ForTy serves as its baseline. When a natural forest is logged and replanted, satellite images look green again soon, but ForTy catches the shift from natural forest to plantation. It gives each pixel a probability, so a company can accept only highly certain pixels while a researcher can use the full range.
What to do now
ForTy 2020 and its time series are free under an open license on Google Earth Engine and can also be downloaded from Figshare.
Read this far? We'll send tomorrow's too
The same briefing you just read, every day. A weekly synthesis on Sundays and a monthly report at the start of each month come with it. Sent at 8AM KST, which is the evening before in the US.
Free · no ads · one-click unsubscribe
Check your inbox
Find this subject in your inbox and press the link to start your subscription.