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.
108 days running
1058 stories · 297 terms explained · since 2026-07-23
·what we picked →
2026-11-07
· ~10 min read
· Past edition
10 picked from 79 candidates · ordered by significance
Today's Insight
AI companies' own claims run ahead of what outsiders can check
Meta added Spaces, topic-specific chat rooms, and avatar video calls to Muse, its personal AI agent, in the same week. Mark Zuckerberg said on the October earnings call that Muse agents had 10 million users in the U.S. and Canada less than two months after launch, a figure Meta reported itself. That same Meta has put on indefinite hold a Facebook business-account terms change that would have let it use external websites for AI training, after a UK Competition and Markets Authority investigation and injunctions from courts in Brazil and Norway.
Continue reading (2)Show less
Two outside reports this week touched Anthropic. METR, a nonprofit that tests AI models, said Anthropic's public August risk report put AI-driven research acceleration at under 2x while the unredacted internal version showed 1.97, and that Anthropic may already have crossed its warning threshold. Employees reported a 12x productivity gain in informal polls, but METR puts the real figure between 1.3x and 3x. Data from Ramp, a company-spend management firm, shows that companies' AI spending grew just 1.5% over August and September, down from 42% in July, as Anthropic nears its IPO.
OpenAI said right after the midterms that it will not use non-disclosure agreements with public officials on its data center projects. Anthropic relaxed a game-cloning ban in Claude Code about a week after it started blocking normal work. Both moved only after outside reaction. By contrast, Anthropic promised up front, when it opened read-only bank account access in Claude, that it would not train on those chats or show ads.
Signal to watch
Meta said its consultation with industry groups could run past November 20; when the terms come back, check whether they include a commitment to respect .
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.
Meta has added Spaces to its personal AI agent Muse: separate chat rooms that each cover one topic, such as a vacation plan or a holiday gift tracker. You describe what you need in chat, and Muse builds a tracker, planner or project board for you. If you connect the apps you use, a Space updates itself in the background, showing concert dates, price changes or savings progress without being asked. Mark Zuckerberg posted on X that he used Spaces to build a BBQ game, an event tracker and a home-control dashboard.
Why it matters
Why It Matters
Most chatbots answer a question and stop; a Space builds a small tracker or board and keeps it updated without being asked. Even TechCrunch's reporter is not sure exactly what Spaces are, and the article does not explain how far a Space reaches into connected apps.
Meta has indefinitely delayed a plan to change its Facebook business-account terms so it could use external websites' content for AI training. A Meta spokesperson told Reuters the company is pausing the changes and that its existing terms remain in effect. The original terms would have required accounts to allow crawling of any linked external site, and even asked publishers to create test credentials that let Meta's bots get past paywalls. The start date had already slipped from October 30 to November 20, courts in Brazil and Norway issued preliminary injunctions, and the UK Competition and Markets Authority (CMA) opened a formal investigation.
Why it matters
Why It Matters
Meta described this as a pause, and Nieman Lab noted that the revised terms contain no specific language about respecting preferences. Theo Bamber, chief executive of the UK's News Media Association, wrote "Does Meta think we can’t read?" and argued that dropping the word "train" changes little because "develop" poses the same threat. Meta's reminder that its crawling practices remain unchanged and that site owners can keep using robots.txt sits in an update to page owners, not in the terms.
What to do now
If you run a Facebook page for a publication or business, check the terms notice from Meta and confirm your site's robots.txt is set the way you intend.
03The Decoder
Press
Covered by 3 more outlets
·2026-11-06·~40s read
Products & Services
Anthropic has opened a beta of Personal Finance in Claude, which lets users connect a bank account and ask questions about their own money. It is available only in the U.S., to Pro, Max and Ultra subscribers. The connection runs through the fintech service Plaid, so Claude never sees the bank password, and access is read-only, so Claude cannot move money or invest. Claude breaks spending down by category, checks upcoming payments, and answers questions such as "Am I spending more than I earn?"
Why it matters
Why It Matters
Anthropic says Personal Finance chats won't be used for training, it won't show ads or sponsored recommendations, and users can disconnect at any time. Its stated long-term goal is to connect Claude with licensed financial advisers and regulated financial firms; this read-only beta does neither yet. OpenAI already offers a similar feature called Finances in the U.S.
A game-cloning ban that Anthropic added to its coding tool Claude Code late last month was relaxed after about a week, because it was blocking normal game development as well. Developers complained that it also blocked some software analysis and work. Anthropic edited the instructions for Claude Opus 5.5 so that requests to copy an existing game outright are still refused, but building a new game in the same genre that borrows ideas such as a specific game's combat system is allowed. When asked for a clone, Claude now offers to build a new game in the same genre instead of just refusing.
Why it matters
Why It Matters
Anthropic narrowed what it blocks from mentioning an existing game to reproducing the original's specific structure and content. Addy Osmani, a technical staffer at Anthropic, acknowledged that the short rule also caught users who named an existing game just to explain how they wanted theirs to work. Anthropic had added the rule to prevent intellectual property infringement, and it blocked ordinary development along with clones.
What to do now
If Claude Code refused a game project, update to version 2.1.317 or later and try again.
Right after this week's midterms, in which voters resoundingly voiced opposition to data center expansion, OpenAI announced new commitments to the communities where it builds data centers. The most visible is that it will not use non-disclosure agreements () with public officials, keeping only "narrow protections" for security information, trade secrets and competitively sensitive information. OpenAI says it will pay "the full cost" of the electricity and infrastructure it needs, publish the jobs it expects to create and report on hiring, and give every future "compute campus" (its new name for a data center) that it owns and operates a "local compact" with the community.
Why it matters
Why It Matters
Data center projects have been criticized for using NDAs to hide talks between developers and local officials, and Amazon said in October it would stop using NDAs with government agencies. OpenAI made its pledge days after a vote in which, The Verge says, voters opposed data center expansion (article 7 on this page). The "local compact" covers only future compute campuses OpenAI owns and operates.
Meta has added real-time voice conversation and avatar video calls to its personal AI agent Muse, available from November 5 to U.S. users aged 18 and over on mobile, web and Mac. The avatar does not look at the user's face through the camera; it generates expressions and gestures from the voice and the conversation context alone. It takes about 870 ms after the user stops speaking for a synchronized voice-and-video reply to come back. Meta plans to bring the same technology to its AI wearable Muse Charm, due this year, and to AI glasses, and Muse has not officially launched in Korea.
Why it matters
Why It Matters
Meta used to cut the generation steps from 120 to 2 and, working with Nvidia, runs 12 video sessions at once on a single GB200 GPU, which raises serving capacity 8x over its baseline model. These are Meta's own figures. Generated video carries an invisible watermark called Video Seal so its origin can be traced.
The Verge's podcast The Vergecast says millions of American voters went to the polls on Tuesday and voted overwhelmingly for Democrats and against AI. The hosts expect the fight against Big Tech to come roaring back and the fight over data centers to get harder for the people building them. They discuss how congressional energy will likely be pointed at tech CEOs and AI regulation, and whether anything will actually get done in the next two years. In the second half they ask whether OpenAI's explosive growth in web search means much when even Google cannot figure out how to make money on AI search.
Why it matters
Why It Matters
The hosts expect both regulation of the AI industry and the siting of data centers to get harder. In the same week, OpenAI said it would not use s with public officials on data center projects (article 5). The line that businesses are scaling back and agents are not taking over is the hosts' assessment; the episode description gives no data for it.
Ramp, a company-spend management firm, reports that the growth of companies' AI spending slowed from 42% month over month in July to just 1.5% across August and September. Usage keeps rising, so the problem is price per unit rather than demand, says Ara Kharazian, Ramp's lead economist. OpenAI halved the API prices of GPT-6 Sol and GPT-6 Luna in September, and Anthropic also cut prices on Fable 5.1 and Opus 5.5 sharply. According to company filings, Anthropic's price cuts reduced its revenue growth by about $600 million.
Why it matters
Why It Matters
Kharazian warns that almost all AI spending comes from venture-backed, high-growth tech companies, so a slowdown among the top 1% alone could sharply cut revenue growth. He calls it a "buyer's market": price cuts have made AI cheaper, and companies hold costs down with defaults that block long runs of expensive models, per-employee budgets, and routers that send work to lighter models. Bloomberg reported Friday that Anthropic's estimated valuation fell from $2 trillion in early October to $1.2 to $1.3 trillion.
Growth in companies' AI token spending fell from 42% in July to 1.5% across August and September.
What to do now
If you manage your team's AI costs, check whether defaults that block long runs of expensive models, per-employee budgets, and a router to lighter models are turned on.
Thinking Machines Labs, founded by former OpenAI chief technology officer Mira Murati, has released Inkling 2 Small, a language model it says can automate AI training tasks, and says it will release the under an open-source license next week. The model has 594B in total but activates only about 17B when it receives a prompt, roughly 1/35 of the whole. The company says it used the model to fine-tune the open-source Qwen3.5-4B and doubled its forecasting score in 4.5 hours. Separately, Business Insider reports the company is raising $5 billion led by Accel at an expected $40 billion valuation, which the company has not confirmed.
Why it matters
Why It Matters
Having AI take on the work of training AI is this model's pitch, and article 10 on this page reports that METR estimates about 80% of Anthropic's acceleration came from using AI in its own research. Anthropic uses that help inside its own lab; Thinking Machines is offering it to customers. The performance figures are all company claims, though, and outside parties can only check them once the weights are released.
METR, a nonprofit that tests AI models, says Anthropic may have crossed its own warning threshold for AI-driven research acceleration as early as July 15. Anthropic's public August risk report put acceleration at "less than a factor of 2", but the unredacted internal version showed 1.97, and Anthropic itself acknowledged it is "plausible" that the threshold was crossed. METR's audit finds that test scores have risen about 1.9 times faster since October 2025 than before, with roughly 80% of that acceleration estimated to come from using AI in Anthropic's own research. In informal polls, employees said the latest internal model raised productivity about 12x on average, but METR calls that "very loose evidence" and puts the true acceleration between 1.3x and 3x.
Why it matters
Why It Matters
Anthropic had unilateral power to redact non-public information and used it heavily. Most of the evidence the auditors rated most informative was removed, so the public must rely on METR's judgment "much more than would be ideal." Anthropic also cancelled, partway through, the audit of how it splits compute between capabilities and safety research.
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.