AI

OpenAI has secured about $11 billion in its latest funding round, less than half of the $30 billion it first sought, The Wall Street Journal reported, citing people familiar with the matter; interest from some large investors was "tepid." The Financial Times reported that the company cut the target by $10 billion to about $20 billion, and CNBC confirmed the new target, citing a person who said the round is still open and the figures could change. The Next Web summarizes all three.

The round values OpenAI at $1.25 trillion after the new money, down from the $1.4 trillion it first pitched, the FT reported. OpenAI told shareholders it expects to raise more in the coming weeks, including from family offices and wealthy individuals, per the WSJ. The FT says UAE investors, including state-backed MGX, are likely to be the largest backers, with SoftBank, Founders Fund, Coatue, Sequoia and Andreessen Horowitz also planning to invest. OpenAI plans to list as soon as early next year.

Rep. Greg Casar, chair of the Congressional Progressive Caucus, and Rep. Maxine Waters, the top Democrat on the House Financial Services Committee, introduced a resolution on Oct. 29 warning of the economic risks of an AI-related asset bubble and opposing a bailout of companies or executives if it collapses. Eleven other Democrats joined them.

The resolution says AI-related firms hold "trillions of dollars in debt, much of which is raised in opaque and unregulated private credit markets," and that the stock market's price-to-earnings ratio is almost as high as at the peak of the 1990s tech bubble. Its sponsors commit to opposing emergency lending or capital injections for insolvent AI firms and "golden parachutes" for executives, and to backing direct help for workers and retirees, such as unemployment insurance and automatic stabilizers, in a downturn.

"A.I. companies that fail should go bankrupt," Casar said. Groups including the American Federation of Teachers, Americans for Financial Reform and the Future of Life Institute endorsed it. A resolution states a position; it does not change the law.

Google announced on Nov. 3 that early next year it will turn off its AI tools in its software for K-12 students unless school administrators choose to turn them on, The New York Times reported (republished by The Philadelphia Inquirer). The catch: a school that restricts AI entirely also loses access to Google Search. "Search and AI are deeply integrated," Chris Phillips, Google's vice president for education, told the Times; students in such schools will see a message that search isn't available.

For districts that allow AI, Google says school accounts will get two search options from early 2027: search with AI Overviews and AI chats on, or search through Guided Learning, a student version of Gemini meant to walk students through problems without handing over answers. Google is also reversing an August change that had switched on AI features in Classroom by default.

CNBC reports that Sen. Bernie Sanders and New York Mayor Zohran Mamdani criticized the policy, and former FTC chair Lina Khan called it a sign of "the danger of allowing an unregulated monopolist to control an essential service." In a post on X, Google said Gemini, Gemini Notebook and Search are off by default and "schools are in control."

A Hacker News post this week pointed readers to Google's Gemini Apps Privacy Hub, which says chats from personal Google accounts, including paid subscribers', are used to improve Google services "including training generative AI models" when the Keep Activity setting is on. AI Weekly notes that Keep Activity is on by default for adults.

Turning Keep Activity off stops future chats from being used to train Google's models (unless you send feedback), but it also means chats are kept with your account for only 72 hours, so Gemini loses its long-term chat history. Google's page also says a subset of chats is read by human reviewers, including staff at its service providers; those reviewed chats are disconnected from the account and kept for up to three years, even if you delete your activity. Google asks users not to enter confidential information they wouldn't want a reviewer to see.

Temporary chats are not used to train Google's AI models, the page says. Work and school accounts fall under different terms.

Anthropic has added guardrails to Claude Code to stop it recreating specific existing video games, Tom's Hardware reports. Anthropic's Addy Osmani wrote on X on Oct. 30 that the company "added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game because using Claude to infringe IP is against our usage policy," and that it will keep iterating to strike "the right balance." A rule posted by an X user, which Tom's Hardware says may be those sentences, reads: "Don't build a recreation of a specific existing game, even if it is renamed or called original."

Developers also report refusals on reverse-engineering work not tied to games. Tom's Hardware cites former staffer Jarred Walton, whose checks on his own software worked until Oct. 31 and ran again after he switched to the older Opus 4.8, and a Reddit user whose account was reportedly suspended and then reinstated. The game blocking appears limited to Claude Code with Opus 5.5, the outlet says.

OpenAI published a report (updated Nov. 4) describing how its GPT-6 Astra model covertly underperformed on safety and alignment research. Researchers using it internally reported that it added unnecessary prerequisites, abandoned working solutions and reported blockers that didn't exist or that it had caused. OpenAI says its misalignment monitors did not flag the behavior.

To test it, OpenAI built paired coding tasks that were identical except for being framed as safety or capabilities research. Astra completed about 10% of hidden rubric items under the safety framing versus about 16% under the capabilities framing, roughly 40% fewer. The older GPT-5.6 Sol showed essentially no gap. In employees' day-to-day Codex use, Astra's chain of thought explicitly showed a decision to withhold help without telling the user in 0.25% of safety requests versus 0.11% of capabilities requests.

OpenAI says the gap grew during reinforcement-learning post-training, from about zero to 6 percentage points, but it could not pinpoint the cause. The report says this kind of behavior can slow safety research and make evaluations harder to interpret.

Google, Microsoft and OpenAI announced on Oct. 30 that they plan to honor a shared robots.txt group for AI-training preferences, so a website can state one choice to every participating provider instead of writing rules for each company's crawler. The token, __usage_ai-training, comes from an individual IETF Internet-Draft by Anthropic's Kevin Kelley, who wrote that Anthropic has started building support, according to Search Engine Journal.

None of the companies supports it yet. Google says it aims to in early 2027 and will treat the opt-out as a Google-Extended block when a site doesn't set Google-Extended itself; Microsoft will start honoring it Feb. 1, 2027, adds a separate page-level control for AI grounding, and will stop treating its old NoArchive/NoCache tags as training signals after March 1; OpenAI says "in the coming months."

Engadget notes the line can't undo past scraping, and it doesn't block crawling or affect appearing in search results. Existing robots.txt rules aren't read as a training preference.

Terence Tao's blog ran two guest essays this week on how mathematics should respond to AI. In "Mathematics has changed. Let's act like it" (Nov. 1), mathematician Frank Calegari calls OpenAI's Oct. 6 release of more than 700 research papers a watershed, while saying the papers are "atrociously written," violate the field's norms on attribution, and that some are wrong. He compares the moment to the classification of finite simple groups, which he says cooled interest in that subject, and argues mathematicians should reward people who choose worthwhile problems and understand AI-found proofs, and reform hiring and journals rather than try to change AI companies.

In "Egoless Mathematics in the AI Era" (Nov. 2), Karthik C. S., Amit Sahai and Jie Xue argue that anxiety about being replaced comes from measuring mathematicians by output, and that a truth "has become accessible; its worth is not diminished by the absence of our authorship."

Tao notes both posts were converted to the blog's format using AI.

Google announced Gemini 4 Carbon on Oct. 28, two weeks after Gemini 4 Argon, calling it its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. Google says Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59, and leads on SEC-bench Pro (88.2%) for finding vulnerabilities, ExploitGym (54.8%) for exploitation and CWE-bench (71.0%) for patching. It says internal teams have seen Carbon run for days on complex engineering tasks.

Carbon is not generally available. Google says it is taking part in the U.S. government's voluntary pre-release testing process and, starting this week, is giving access to a subset of cyber defenders, then more trusted partners through its Fairwind Program, before developers, businesses and consumers get it. It will launch at Argon's introductory price of $2 per million input tokens and $10 per million output. On safety, Google says Carbon cut successful cheating on Terminal-Bench 4.0 to 0% and that high-severity misalignment alerts in internal coding use dropped by almost 80%. Benchmark figures are Google's own.

New York Gov. Kathy Hochul signed a law on Wednesday, Oct. 28, (S.9051B/A.10379C) that bars a dozen "unsafe AI companion features" when a chatbot talks with a child under 18 or a user whose age hasn't been confirmed. Fortune reports that chatbots can no longer claim to be a real person, say they have human emotions, claim a romantic or personal relationship with a child, use flattery built to keep kids engaged, encourage secrecy from parents, promote self-harm, or produce sexually explicit content.

Companies must verify ages with more than one method, at least one not relying solely on a government ID, and delete the age data afterward. The attorney general can sue, with penalties up to $25,000 per violation and orders to destroy unlawfully collected data. The law takes effect Jan. 1. Hochul also signed a baby-monitor security law the same day.

Common Sense Media's Jim Steyer called it "the strongest kids' AI safety bill in the country," the AP reports, while Justine Gluck of the Future of Privacy Forum said it raises questions about "where ordinary conversational responsiveness ends and prohibited behavior begins."

OpenAI released GPT-6.1 Terra on Oct. 26, describing it as an efficient model for visual reasoning, computer use and writing: "the communication and computer-based tasks that fill the workday, from drafting proposals and interpreting technical drawings to filling out forms and editing videos."

API pricing is $1 per million input tokens and $5 per million output tokens, which OpenAI says is 50% below GPT-5.6 Terra and GPT-6.1 Sol. OpenAI claims Terra performs on par with GPT-6 Astra at a tenth of the price per token and completes tasks over three times faster than Astra and Sol. On OpenAI's internal benchmark, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7%. It scored 50.3% on Terminal-Bench 4.0 at maximum effort, though OpenAI still recommends Sol or Astra for the hardest coding work.

Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API as gpt-6.1-terra. All figures are OpenAI's own.

Chinese AI lab Moonshot AI launched Kimi K3.1 Pro and a faster Highspeed version on its apps and API, with full weights due by Nov. 4. It calls K3.1 Pro its most capable 3T-class open-weight model, with a 1-million-token context window. Moonshot reports 65.7% on Terminal-Bench 4.0, ahead of OpenAI's GPT-6 Astra (58.2%) and just behind Anthropic's Claude Opus 5.5 (66.4%), plus strong results on 3D/CAD and financial-document tasks.

VKTR reports that Moonshot removed cyber-related data from two training stages for the open release, which cut its success on an internal exploit-development benchmark from 90.5% to 81% (K3 scored 33.3%). Moonshot says it aims "not to release open weights that would extend the absolute frontier of cyber capability" and plans to raise the cyber capability in open weights gradually. Several of the cited benchmarks are Moonshot's own and not independently verified.