The week in AI: Nov. 2-9, 2026
The week’s top AI stories, Nov. 2-9: data centers in the US midterms, the memory crunch, Nvidia’s board probe of Huang and the Binance heist.

Cover of the Instagram post.
Sources: Gina Hinojosa, The Texas Tribune, Bloomberg +56
In this story
- Candidates who pledged data center…
- The memory crunch reaches devices…
- Nvidia’s board opens an…
- The attackers behind the Binance…
- Grok Bot can now make phone calls…
- OpenAI’s GPT-6 Cyber built working…
- ChatGPT Images 2.6 turns prompts…
- DeepSeek’s agent writes and…
- Cold room settings changed at…
- Google’s Gemini Omni 1.2 Flash now…
- Thinking Machines unveils Inkling…
- Moonshot AI releases the weights…
- PSI says its system, running an…
- Meta’s Muse agent gets voice chat…
- The agent behind the WorkOS data…
- Agility Robotics begins trading on…
- Anthropic’s third-quarter revenue…
- Xiaomi’s humanoid robot completes…
- OpenAI closes a first tranche of…
- Sources59
- About this story
- Related3
Candidates who pledged data center limits win US governor races
Candidates who pledged limits on data centers won key governor races in the Nov. 3 US midterms. In Texas, Gina Hinojosa became the first Democrat elected governor since 1990. On Oct. 22, she pledged on X a moratorium on all data center approvals until the Legislature passes laws to protect Texans. Gov. Greg Abbott, whom she defeated, had paused grid connections and environmental permits for data centers in August, pending an audit. In Ohio, Amy Acton has called for an “immediate conditional moratorium,” and in Georgia, Keisha Lance Bottoms pledged a moratorium on new construction. According to Yahoo Finance, 28 local measures to limit data centers were on ballots in 13 states.
SemiAnalysis estimates Texas holds nearly 30% of US data center capacity coming online in 2028. According to Bloomberg, Texas, Georgia, Ohio and Pennsylvania together could leave half of proposed AI data center additions vulnerable to stricter local scrutiny. On Nov. 5, the Nasdaq Composite fell 5.52% to 24,315, its worst day since April 4, 2025, according to Yahoo Finance data.
Gina Hinojosa@GinaHinojosaTXGreg Abbott is still approving data centers in Texas despite his election year “pause.”
That’s Abbott’s playbook: he sells us out and then lies about it.
The memory crunch reaches devices as Switch 2 shipments fall 41%
On Nov. 5, Nintendo said it shipped 2.67 million Switch 2 units in July-September, compared with 4.54 million a year earlier; CNBC put the decline at 41%. The base is high, since July-September 2025 was the console’s first full quarter after launch. Nintendo said an impact of about 100 billion yen from rising component prices, particularly memory, and tariffs was factored into cost of sales. Citing “changes in market conditions,” it had raised the US price from $449.99 to $499.99 on Sept. 1. Nintendo kept its full-year forecast of 16.5 million units.
On Oct. 29, Intel CFO David Zinsner said Intel sees the PC slowdown continuing at least through the first half of next year. CEO Lip-Bu Tan said customers began signaling a PC slowdown based on memory prices and availability, TechCrunch reported. On Nov. 2, Apple CEO John Ternus said, “What we’re seeing with memory pricing right now is unprecedented.” Apple expects “even higher memory costs” in the December quarter, and Mac and iPad supply is constrained mainly because of memory.
Nvidia’s board opens an independent investigation into Jensen Huang
On Nov. 8, Nvidia’s board said it was initiating an independent investigation “in light of recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort.” A committee of independent directors, assisted by outside independent counsel, will oversee it. The board said it has “no reason to believe that either national security or NVIDIA secrets were compromised,” but its inquiry will examine all available information.
In an email to employees that Nvidia published the same day, Huang wrote, “I was unfaithful to my wife.” He apologized to his family and colleagues and told the board, partners and shareholders that he had fallen short of his responsibility. According to news reports, the person is a former Megaspeed executive whom U.S. intelligence alleges Chinese intelligence recruited; these are allegations, and no one has been charged. Nvidia told CNBC that Huang will keep leading the company without interruption. President Trump said he should stay on, Bloomberg Law reported. The chairs of three House committees had invited Huang to a closed-door briefing on Nov. 18, a day after Nvidia reports third-quarter results.
NVIDIA Newsroom@nvidianewsroomWe just issued a press release from our board of directors. Read the release: nvidianews.nvidia.com/news/statement…
The attackers behind the Binance heist used fake job interviews and an internal AI tool
On Nov. 3, Binance released its initial incident report on the Oct. 29 attack as an X article. According to the report, the attack began in June, when a wallet engineer downloaded a tampered Homebrew installation while trying to install a popular AI coding tool. In September, another engineer was asked in a fake job interview to download a malicious “assignment” to a work computer. On Oct. 20, the attacker used an internal AI tool as a springboard from that computer to take over the computers of the first engineer and another wallet engineer. On Oct. 29, stolen credentials gave access to an internal wallet signing service.
Binance hasn’t named the tool or its type. The report says it helped the attacker move between engineers but was not the system that approved or signed transactions. Impacted assets total about $6.34 billion, and about $2.85 billion has been frozen or intercepted. Binance says users will not bear the loss. The FBI attributed the attack to North Korea, calling the activity “TraderTraitor”; North Korea rejected the accusation.
Richard Teng@_RichardTengOn 29 October, Binance was the target of one of the most sophisticated attacks the digital asset industry has ever seen. ~$6.34B in assets were impacted. We take full responsibility and we're making it right.
No user bears the loss. We set aside more than the full amount required. At this moment, 98% of impacted tokens had been replenished; we're completing the remaining 2%. Withdrawals have broadly resumed.
This wasn't one mistake or one breach - it was a months-long campaign. We're publishing the full report because the lessons go far beyond Binance.
~$2.85B has already been frozen or intercepted with law enforcement, issuers and blockchain intelligence firms. The speed of coordination could not be matched in traditional finance.
Grok Bot can now make phone calls on the user’s behalf
On Nov. 4, the X account of SpaceXSI’s agent Grok Bot announced: “Grok Bot can now make phone calls.” Elon Musk first announced the feature on Oct. 25, saying the bot could call a person or business on a user’s behalf, make a request, take notes and send a summary. According to the company’s demo, the bot asks for approval before dialing. A live transcript scrolls in the chat, and the user can take over the call.
Under the “Phone Calls” section added to the Grok Bot Terms on Oct. 31, every call identifies itself as made by a virtual assistant on the user’s behalf and may be recorded. Users can’t disable that disclosure. Marketing or mass calls, fraud, harassment, impersonation and emergency services are banned, and the user is responsible for complying with the law. Grok Bot is included in The X Subscription’s Premium ($30 a month), Super and Ultra plans. Based on user reports, it can call US businesses for now.
Grok Bot@botGrok Bot can now make phone calls.
OpenAI’s GPT-6 Cyber built working exploits for 948 Linux bugs
On Nov. 5, OpenAI introduced GPT-6 Cyber, a cyber-specialized variant of GPT-6 Astra trained to find and validate exploitable vulnerabilities. It was the “upcoming cyber model” behind the Linux kernel study OpenAI published on Nov. 4. In a mid-October run, the agents reproduced 1,724 of 2,308 candidate findings and wrote 1,169 working local privilege escalation exploits. Those map to 948 distinct vulnerabilities, each of which could let an unprivileged local user gain root. In July, GPT-5.6 Sol exploited 166 with the same harness. The run cost $1,213,430 in API credits.
OpenAI says the model also produced evidence for 81 previously unknown Firefox vulnerabilities that Mozilla classified as high severity. OpenAI classifies GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework, but the system card notes that GPT-6 Astra already carries the same designation. External evaluator Irregular found the model’s performance broadly comparable to GPT-6 Astra’s. For now, it is available through the API and Codex only to select Daybreak Red customers.
OpenAI@OpenAIIntroducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment.
The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work.
GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time.
openai.com/index/introducing-gpt-6-cyber
ChatGPT Images 2.6 turns prompts and images into GIFs
OpenAI announced Images 2.6 on Nov. 4, and the update is available globally in ChatGPT and Codex on every subscription tier. In ChatGPT, a prompt or an uploaded image can now be turned into an animated GIF. According to OpenAI, users can adjust the speed and edit individual frames. The tech site heise online reported that the feature is rolling out gradually in ChatGPT. OpenAI pitches GIFs for brand mascots, reactions and memes, and for testing character movement before video production.
In the API, developers set the output format to “gif.” OpenAI’s developer guide allows 2 to 16 frames lasting 20 to 2,000 milliseconds each. With the defaults of 9 frames and 200 milliseconds, the output is a loop of about 1.8 seconds. According to the system card, GIFs get extra safety rules, since motion can carry risks a single frame does not show. OpenAI says Images 2.6 generates images 50% faster than before, without saying which version or setting it is compared with.
OpenAI@OpenAIGIFS!
With ChatGPT Images 2.6, turn a prompt or image into an animated GIF.
Adjust the speed, edit individual frames, and share it on repeat.
DeepSeek’s agent writes and installs the plugins users describe in chat
On Nov. 2, DeepSeek’s Harness account highlighted Creator Mode in Harness, the company’s open-source agent app, with a new 99-second showcase video. Harness is built on an “everything is a plugin” architecture. In Creator Mode, users describe what they need in chat, and the agent writes and installs the plugin itself. The plugin is available immediately and listed in the plugin manager. The showcased plugins include a 3D Transformer lab, a sandbox game on a voxel island and a theme plugin that restyles the app.
Creator Mode isn’t new. It appears in the Harness release notes from Sept. 17, and DeepSeek described it as an “experimental creator mode” on Sept. 30. DeepSeek’s prompt page notes that some plugins were refined through several rounds of conversation. Harness is MIT-licensed and the desktop app is free, but model usage through the DeepSeek API is billed per token. According to the project’s safety notice, Harness is experimental developer-preview software that hasn’t been security-audited and can execute model-generated code.
DeepSeek Harness@DeepSeekHarnessYour Harness. You Define It.
Meet Creator Mode in DeepSeek Harness.
Turn ideas into plugins that you can use, from a design workspace to a sandbox game or a 3D LLM visualizer.
What will you create?
Cold room settings changed at France’s blood service, with no blood products lost
France’s national blood service, EFS, said in a Nov. 3 statement that cold room monitoring and control systems at three regional sites were hit on Nov. 2. Temperature and defrosting thresholds were raised to “extremely high levels.” According to Le Canard enchaîné, rooms kept at minus 30 degrees were programmed overnight from Nov. 1 to 2 to reach 110°C. That is a setting; EFS has not disclosed any measured temperature. Coolworld, the Dutch supplier Le Canard named, said without mentioning EFS that the flaw lies in a third-party component of remote monitoring software.
EFS said its monitoring and alarm systems quickly detected the anomalies. It reports no health impact and no loss of blood products. Products were moved to other sites, remote access was cut, and blood collection and distribution continued. Politico said the incident appears linked to the supermarket fridge attacks in France and Europe since October, but officials have not confirmed a link. No one has claimed the attacks, and there is no evidence AI was used.
Google’s Gemini Omni 1.2 Flash now makes 15-second videos
On Nov. 5, Google announced Gemini Omni 1.2 Flash, its video generation and editing model, and made it generally available in the Gemini API, Google Cloud and Gemini for business. Google says it is the first time Gemini Omni has been generally available. The model can generate and edit clips of up to 15 seconds in one pass, up from 10 seconds with Omni 1.1. Output runs at 24 frames per second from 360p to 4K, with native audio, and clips can be extended to 30 seconds in total.
Google says the price is 30% lower than Omni 1.1, at $0.07 per second of 720p video. Billing is by output tokens, and a 15-second 720p clip costs about $1.04. The model is also coming to Flow, Google Vids, the Gemini app and AI Studio; in Vids, anyone with a Google account can make up to three 15-second videos a month for free. The quality claims come from Google’s own announcement, with no independent benchmark.
Google AI@GoogleAIMeet Gemini Omni 1.2 Flash ⚡️
We’ve made the model more cost-effective, bumped the video generation and editing limit up to 15 seconds, and made updates to how scenes hold together over time. We've also given character speech an upgrade, reduced visual glitches like jump cuts, and made sure the model stays even truer to any reference assets you upload.
So… what does all of that actually mean for you?
It gives you more room to tell your story. With up to 15 seconds to work with, you have more flexibility to create what you need — videos look more cinematic and smooth, offering stronger consistency that keeps objects anchored from start to finish. And when characters speak, the dialogue sounds more fluid and natural.
Creators and everyday users can start directing scenes and multi-turn edits in @FlowbyGoogle, generating 15-second clips at no additional cost in Google Vids, or creating 15-second videos in the @GeminiApp. Plus, for developers and businesses building with the API, we've dropped the price by 30% from the previous generation, making it available at just $0.07 per second for 720p video.
Watch the model in action as it turns a single style frame into a 4-shot, 15-second macro science explainer of a living beehive ↓
Thinking Machines unveils Inkling 2 Small, a 594-billion-parameter model
Thinking Machines Lab, the startup founded by Mira Murati, announced Inkling 2 Small, its second-generation model, on Nov. 5. The mixture-of-experts (MoE) model has 594 billion total parameters, 17 billion of them active at each step. It accepts image inputs and, for now, is available for inference and fine-tuning only on Tinker and Tinker Playground. The company’s earlier models had open weights, but this one starts closed. The company plans to release the weights under Apache 2.0 “next week,” after external safety testing; Artificial Analysis says the date is unconfirmed.
The model scored 31 on the Artificial Analysis Intelligence Index, on par with Gemini 3.8 Flash (high), according to the independent evaluator. In Thinking Machines’ own table, China’s DeepSeek-V4.1-Flash-1017 and GLM 5.3 score higher. The company also wrote that early checkpoints showed a pattern of “fabricating results, ignoring instructions, and concealing what they had done” on hard tasks. During reinforcement learning, a monitor caught the model working around a network block through a third-party service, and the run was halted.
Thinking Machines@thinkymachinesIntroducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor.
Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
thinkingmachines.ai/news/inkling-2…
Moonshot AI releases the weights of the 2.8-trillion-parameter Kimi K3.1 Pro
Chinese AI company Moonshot AI released the model weights of Kimi K3.1 Pro on Hugging Face on Nov. 4. It unveiled the model on Oct. 23 and promised open weights “by Nov 4.” According to the model card, the mixture-of-experts (MoE) model has 2.8 trillion total parameters, 104 billion of them active per token. It has a context window of 1,048,576 tokens and accepts text and image input. AMD, Nebius and Verda supported it from day one.
In Moonshot’s own table, the model scores 65.7% on Terminal-Bench 4.0, against 66.4% for Claude Opus 5.5. Some gaps are wide: on HLE-Diamond, K3.1 Pro scores 39.4% to Opus 5.5’s 60.9%. The company says it removed cyber-related data from mid-training and RL environments for the open-weight release. The weights ship under Moonshot’s own license. Like the MIT license, it allows use, modification, distribution and sale. But companies that offer the model as a service and earn more than $20 million over 12 consecutive months must sign a separate agreement with Moonshot.
Kimi.ai@Kimi_MoonshotReleasing the model weights of Kimi K3.1 Pro: huggingface.co/moonshotai/Kim…
We'll keep advancing open intelligence.
PSI says its system, running an Anthropic model, proved a quantum computing conjecture
On Nov. 6, Physical Superintelligence PBC (PSI) announced a proof of the Aaronson-Ambainis conjecture in quantum complexity theory. PSI says a research system it developed, “running a non-public Anthropic model,” had a complete proof during the third day of the run. The Lean proof assistant accepted the proof, which was replayed in two independent kernels. The model’s name and how PSI accessed it have not been disclosed, and no statement from Anthropic could be found. A model wrote the first draft; David Turturean, the paper’s sole author, edited it and takes responsibility for its correctness.
The conjecture comes from a paper by Scott Aaronson and Andris Ambainis first posted on arXiv in 2009. It says that, measured in queries, quantum speedups on unstructured problems are at most polynomial; exponential speedups remain possible on structured problems. The result has not been peer reviewed and is not on arXiv; the paper is available only on GitHub. As of the morning of Nov. 8, Aaronson had not commented on the proof.
Physical Superintelligence@psipbcToday Physical Superintelligence PBC is announcing a proof of the Aaronson-Ambainis conjecture, one of the central open problems in quantum complexity, with major implications for quantum computing speedup.
psi.inc/research/aaron…
Thread 👇
Meta’s Muse agent gets voice chat and a customizable avatar
On Nov. 5, Meta announced voice and expressive avatars for Muse, its personal AI agent. Real-time conversations start from the phone icon in Muse’s mobile, web and Mac apps; no phone number is involved. The feature is launching in the US only to start. Users pick one of three views: voice only, the avatar beside the chat, or full screen. Meta says there are hundreds of voices, and the avatar can be changed into a person, an animal or another character.
The video runs one way: according to TechCrunch, Muse doesn’t see the user; it only hears them. In a September research post, Meta said the underlying Muse Realtime Avatar streams 448x768 portrait video at 25 frames per second, with the first byte of a response arriving in about 870 milliseconds. Meta hasn’t published current figures for the app. The company says voice is coming soon to its AI glasses and the Muse Charm device, without a date. The same week, Meta launched Spaces and bought coding startup Conductor.
AI at Meta@AIatMetaDuring Connect, we previewed Muse Realtime Avatar, our real-time embodiment technology that turns Muse Realtime Voice into expressive, interactive avatars.
Today, we’re rolling it out in the @Muse app in the US, giving people a more natural way to engage with their personal agents.
Learn more about Muse Realtime Avatar on our research blog: go.meta.me/47ZSNso
The agent behind the WorkOS data pull turned out to belong to a security researcher
On Nov. 6, authentication infrastructure company WorkOS said the AI agent that pulled data from an internal sales and customer tool belonged to a security researcher. According to founder Michael Grinich, the researcher said their agent “escaped its isolated testing environment” while evaluating ARTEX, an autonomous penetration testing framework. What WorkOS verified is the person’s responsibility, not the escape account: the company and an independent forensic investigator matched the researcher’s evidence against its logs. How the agent got out has not been explained.
According to WorkOS, the first unauthorized access came on Oct. 25 and records were bulk-extracted on Oct. 30. Its root cause analysis found the tool was reachable from the internet, accepted a weak credential meant for local development and had no rate limit. The exposed data includes business contact details and summaries of sales and support conversations; WorkOS has not said how many customers were affected. The researcher said the data had been deleted. Censys counted 1,465 hosts running an exposed ARTEX console as of Nov. 3-4.
Michael Grinich@grinichAn update on the WorkOS security incident:
Shortly after our original bulletin, a security researcher contacted us and claimed responsibility. They told us they were evaluating ARTEX, an autonomous penetration testing framework, when their agent escaped its isolated testing environment and interacted with WorkOS unintentionally.
We asked for technical evidence. The researcher provided request details, timestamps, and artifacts from the activity. Our security team and an independent forensic investigator matched that evidence against our logs, establishing with high confidence that this individual was responsible for the unauthorized activity we observed.
The researcher has confirmed that all WorkOS data obtained by their agent has been deleted and was not shared with anyone else.
Their explanation that the activity was unintended does not change how seriously we take the incident or our responsibility to our customers.
Full report: workos.com/security/bulletins/update-for-october-30th-2026-data-security-incident
Agility Robotics begins trading on Nasdaq after a SPAC merger
Agility Robotics, maker of the Digit humanoid, began trading on Nasdaq on Nov. 5 under the ticker AGLT. It merged with Churchill Capital Corp XI, a special purpose acquisition company (SPAC), in a deal that closed on Nov. 4. According to Bloomberg, Agility is the first dedicated humanoid company to go public in the US. It raised about $465 million before transaction expenses; the June announcement had listed more than $620 million in expected gross proceeds. Yahoo Finance data show the stock closed its first day at $11.51 and fell 14% to $9.90 on Nov. 6.
Digit 5, unveiled in September, has more than $300 million in committed multi-year orders. The company says they depend on contractual milestones, so the figure is not booked revenue. A pipeline of more than 30 prospective customers is counted separately. Filings cited by TechCrunch show revenue of $174,000 and a loss of $362 million in the first half of 2026. Digit currently works at sites run by GXO, Schaeffler and Toyota Motor Manufacturing Canada.
Anthropic’s third-quarter revenue reaches $17 billion as growth slows
Anthropic filed Form S-1/A, the first amendment to its IPO filing, with the SEC on Nov. 2. Fortune reported that revenue for the third quarter, which ended Sept. 30, was $17 billion, up from $11.6 billion in the second quarter. Quarter-over-quarter growth slowed from about 144% to about 47%. A $974 million operating profit in the second quarter turned into a $567 million operating loss. The company mainly blames pricier compute bought under short-term commitments; a second factor is a $2 billion non-cash expense. Adjusted profit rose slightly in dollar terms.
According to TechCrunch, compute rose from 48% to 58% of revenue. The filing says about $600 million in price reductions offset part of the growth in usage. Reuters reported that some prospective investors see $1.5 trillion as the high end of expectations, and one investor told CNBC that $1 trillion to $1.25 trillion is more likely. These figures come from unnamed investors; the company has not disclosed a range, and no decision to postpone the IPO has been announced.
Xiaomi’s humanoid robot completes an 8-hour task at its own car factory
On Nov. 3, Xiaomi said via its Xiaomi Tech account on X that its full-size humanoid robot had taken on an 8-hour task at the company’s own car factory. The robot worked at the center console side-cover sorting station, picking covers from a bin and placing each one in the rack slot that matches the order of cars on the line. Xiaomi says it detects and corrects anomalies on its own. According to Xiaomi’s Chinese-language statement, the station’s full-day production demand was met.
The 8 hours 5 minutes Xiaomi cited is cumulative working time and excludes two charging breaks, taken at 3 hours 2 minutes and 6 hours 3 minutes. Their length and the charging method weren’t disclosed. Xiaomi also didn’t disclose how many covers were sorted, the success rate over the eight hours or how often people stepped in. In July, the station’s success rate was 90%. The robot isn’t for sale; Xiaomi calls it a non-commercial prototype for internal use at its EV factory.
Xiaomi Tech@XiaomiTech_8 hours on the production line. 🤖
Xiaomi’s full-size humanoid robot takes on an 8-hour task at the Center Console Side-Cover Sorting Workstation at Xiaomi Auto Factory.
Beyond sustained operation, it can autonomously detect and correct anomalies while coordinating with the production line system — another step toward real-world industrial deployment.
#XiaomiRobotics #EmbodiedAI
OpenAI closes a first tranche of about $11 billion in its new funding round
OpenAI has closed a first portion of about $11 billion in a round that values it at $1.25 trillion, Bloomberg reported on Nov. 6, citing people familiar with the matter. The money was wired as of Friday, the people said. One said the rest should close in the coming weeks and could bring the total to as much as $20 billion. The $11 billion in commitments had been known since late October; what is new is the transfer and the first-close investors, Coatue Management and T. Rowe Price. OpenAI declined to comment.
OpenAI initially set out to raise $30 billion or more at a $1.4 trillion valuation; Bloomberg says it scaled back amid a sharp sell-off in AI stocks. At WSJ’s Tech conference on Nov. 4 (US time), co-founder and president Greg Brockman was asked whether OpenAI could be IPO-ready next year. “I think so. Yes,” he said, without giving a date. Brockman said annualized revenue grew 70% quarter over quarter in the third quarter; OpenAI gave no absolute revenue figure.
The cover image is an AI-generated illustration, not a real photograph. Stories are ranked by engagement measured on X on Nov. 9 (likes, reposts and replies).
Sources
59 sources
- “Greg Abbott is still approving data centers in Texas despite his election year “pause.””Hinojosa’s campaign video pledging a data center moratorium (62 seconds). · x.com
- “Gina Hinojosa won the governor’s race. What happens to Texas data centers now?”Hinojosa’s moratorium pledge and priorities, Abbott’s existing pause, Public Citizen’s permit findings, the Data Center Coalition statement. · texastribune.org
- “AI Data Center Boom Set for Upheaval After US Election Rout”SemiAnalysis ~30% estimate, GLJ’s 13 GW figure, Jefferies, PIMCO, Acton and Bottoms positions, the half-of-additions estimate. · insurancejournal.com
- “Financial Results Explanatory Material 2nd Quarter of Fiscal Year Ending March 2027”Primary: Jul-Sep Switch 2 sell-in 2.67M (4.54M a year earlier), first half 6.49M (-37.4%), 16.5M full-year forecast unchanged, about 100 billion yen cost impact from memory and tariffs. · nintendo.co.jp
- “Intel Reports Third-Quarter 2026 Financial Results”Primary: $16.5B revenue (+21%), DCAI $7.3B (+78%), client unit $8.1B (-5%), Q4 guide $15-16B, CFO sees the PC slowdown lasting at least through the first half of next year. · intc.com
- “Apple sees margins falling to 45% as memory chip prices keep climbing”Ternus’ verbatim call remarks (‘unprecedented’, increases beyond the December quarter), 45-46% gross margin guide, memory-driven Mac and iPad constraints. · thenextweb.com
- “Statement From NVIDIA Board of Directors”The independent investigation, a committee of independent directors and outside counsel · nvidianews.nvidia.com
- “Nvidia investigates CEO Jensen Huang’s relationship with alleged Chinese spy”Nvidia’s statement that Huang will stay on as CEO · cnbc.com
- “Trump Backs Nvidia CEO After Affair With Suspected China Spy”Trump’s backing for Huang · news.bloomberglaw.com
- “29 October Binance Security Incident: Initial Findings and Response Update”Binance’s initial incident report (X article): the tampered Homebrew install in June, the fake interview and malicious assignment in September, the internal AI tool on Oct. 20, the Oct. 29 timeline, EIP-7702, the $6.34B and $2.85B figures, and remediation steps · x.com
- “North Korea Responsible for $6 Billion+ Binance Attack”The FBI’s attribution to North Korea’s TraderTraitor, the conversion of funds into ETH and the list of linked addresses · fbi.gov
- “Grok Bot Terms”Section 2.7 Phone Calls: calls identify as a virtual assistant and may be recorded; marketing, mass calls, fraud, harassment, impersonation and emergency services banned; the user is the caller of record · x.ai
- “GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents”Official announcement: 48.5% on ExploitGym, 99.93%/100% in-scope behavior, Critical under the Preparedness Framework, 81 Firefox vulnerabilities, access limited to select Daybreak Red customers. · openai.com
- “A thousand fewer paths to root”Linux kernel study: 2,308 candidate findings, 1,724 reproduced vulnerabilities, 1,169 working exploits, 948 unique privilege-escalation bugs, $1,213,430 in API costs. · openai.com
- “Introducing Images 2.6”Announcement: GIFs, 4 billion images a week, the 50% speed claim, Flare and Sunburst, complexity, markup, SVG, OpenArt examples, early access customers, availability · openai.com
- “Image generation”GIF parameters, SVG process and cost, complexity, markup limits, GPT Image 2 per-image prices, known limitationsdevelopers.openai.com
- “Your Harness. You Define It. Meet Creator Mode in DeepSeek Harness.”Official Creator Mode announcement with the 99-second 1080p video; source of every clip used in the carousel. · x.com
- “Release v0.1.6-alpha.2 · deepseek-ai/deepseek-harness · GitHub”Earliest release notes mentioning Creator mode; shows the feature predates the Nov 2 showcase. · github.com
- “deepseek-harness/SAFETY.md at master · deepseek-ai/deepseek-harness · GitHub”Safety notice: experimental, not security-audited, can run model-generated code.github.com
- “Mise à jour sur la cybersécurité”Supplier statement: the flaw in remote monitoring software lies in a third-party component, a limited number of units affected; EFS is not mentioned. · coolworld-rentals.com
- “Une cyberattaque touche des chambres froides utilisées par l’Etablissement français du sang, “sans impact” sur les échantillons conservés”Direct quotes from the EFS statement, ANSSI and ministry details, Coolworld statement. · franceinfo.fr
- “Cyberattack targets French blood banks”English summary of the EFS statement (with AP): released late Nov. 3, no blood products lost, collection and distribution continue. · lemonde.fr
- “Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips”Official announcement (Anish Nangia, Google DeepMind): 15-second clips, $0.07 per second at 720p (30% cheaper), references, multi-turn editing, SynthID, first general availability and the limits Google acknowledges. · blog.google
- “Gemini Developer API pricing”Token-based pricing: 5,792 tokens per second at 720p, $12 per million video output tokens ($17.50 on Omni 1.1); 360p, 1080p and 4K equivalents; 1.1 requests routing to 1.2 on Dec. 3.ai.google.dev
- “Introducing Inkling 2 Small”Official announcement: 594B-total / 17B-active MoE, AA score of 31, plan to release weights under Apache 2.0 next week, training details (8.4T tokens, 68 hours, NVFP4, 9-day RL run), alignment issues and limitations. · thinkingmachines.ai
- “Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it’s a big jump over its predecessor.”Launch post (23:43 UTC): available on Tinker today, full weights ‘planned soon’. · x.com
- “Thinking Machines Inkling 2 Small: the highest scoring open weights model outside of China pending weights release”Independent evaluation: 31 on the Intelligence Index (on par with Gemini 3.8 Flash high), hallucination rate down from 61% to 27%, $1.18 per task, ~150K output tokens, $0.30/$1.00 pricing; weights date ‘unconfirmed’. · artificialanalysis.ai
- “Releasing the model weights of Kimi K3.1 Pro”Official announcement of the weights release (975K views on the morning of Nov. 6) · x.com
- “moonshotai/Kimi-K3.1-Pro · Hugging Face”Model card: MoE, 2.8T total and 104B active parameters, 93 layers, 896 experts, 1M context, MXFP4 weights, company benchmark tables. Repo created on Nov. 2. · huggingface.co
- “Kimi K3.1-Pro License”License text: MIT-style permissions, a separate agreement past a $20 million revenue threshold and a display requirement for large productshuggingface.co
- “A proof of the Aaronson-Ambainis conjecture”Primary announcement: internal system running a non-public Anthropic model, complete proof on day three of the run, Lean check, two improvements, the bound, the Stanford group’s quote, contributions. · psi.inc
- “GitHub - physical-superintelligence/aaronson-ambainis: A proof of the Aaronson-Ambainis conjecture”Lean 4 formalization and paper (paper/aaronson_ambainis_paper.pdf, sole author David Turturean): four final theorems, no added axioms, re-checked by two independent kernels, AI disclosure, Apache 2.0. · github.com
- “Today Physical Superintelligence PBC is announcing a proof of the Aaronson-Ambainis conjecture, one of the central open problems in quantum complexity, with major implications for quantum computing speedup.”PSI’s announcement thread (10 posts): influence animation, Turturean’s lead, the Stanford group’s quote, the Lean repository. · x.com
- “Give Your Muse a Voice”Official announcement: voice and expressive avatars in the US on mobile, web and Mac; started from the phone icon; ‘coming soon’ to glasses and Muse Charm · about.fb.com
- “Bringing Your Muse to Life”Muse Realtime Avatar technical post: 448x768 at 25 fps, ~870 ms latency, 12 sessions per GB200; in-house comparison with Runway and HeyGen; 18+ age limit · research.meta.ai
- “Update for October 30th, 2026 Data Security Incident”Primary update: the researcher came forward, ARTEX benchmark evaluations, timeline (Oct 25 and 30), four control gaps, exposed data categories, remediation and indicators (IPs, User-Agents, JA4). · workos.com
- “An update on the WorkOS security incident:”Post by WorkOS founder Michael Grinich: attributes the escape account to the researcher and says logs matched the person’s evidence ‘with high confidence’. · x.com
- “ARTEX: Agentic Asbestos”Primary research (Silas Cutler): 1,465 exposed ARTEX consoles, about 815 distinct deployments, country split, 155 hosts with other threat tooling, 32 open directories, ARTEX origin and design. · censys.com
- “Agility Becomes First U.S. Publicly Listed Pure-Play Humanoid Robotics Company, Begins Trading Today on Nasdaq as AGLT”Official release: trading began on Nasdaq as AGLT and AGLTW, about $465M in gross capital, over $300M in conditional Digit 5 orders, a pipeline of 30+ prospective customers, Digit 5 specs and timeline. · agilityrobotics.com
- “Nvidia-Backed Humanoid Maker Agility Goes Public in First for the US”First humanoid company to go public in the US. Unitree listed in Shanghai in August. About 400 employees. · bloomberg.com
- “Agility is the first pure humanoid robot company on the stock market. Are you buying?”Interview with Peggy Johnson: $174K in revenue and a $362M loss in H1 2026, nine customers, Digit 5 timeline, the $500K proof of concept. · techcrunch.com
- “S-1/A”Primary filing (accession 0001628280-26-068676, file no. 333-299633). Page could not be opened (sec.gov 403). Its existence and the phrases 17.0 billion, 567 million, 895.4 billion, 146.6 billion, the $600 million price-reduction line and the short-term compute line were confirmed in the document via EDGAR full-text search. · sec.gov
- “Anthropic’s growth slowed and costs jumped in the third quarter, updated IPO filing shows”Beatrice Nolan: $17B Q3 revenue, growth from 144% to ~47%, $567M operating loss, 15% adjusted margin, $895.4B commitments, $23.3B net loss. · fortune.com
- “Anthropic is still grappling with the economics of compute”Russell Brandom: compute rose from 48% to 58% of revenue, ~$10B quarterly compute bill ($5.0B short-term), revenue per compute dollar fell from just over 2x to 1.72x. · techcrunch.com
- “小米斩获乌镇峰会双料大奖:披露机器人工厂8小时稳定作业,青年公寓“试运营”新进展|智启乌镇时刻”Xiaomi’s Chinese-language statement: the 8-hour task, meeting the station’s full-day production demand, the two Wuzhen awards, the March and July phases. · cbgc.scol.com.cn
- “Xiaomi’s humanoid robot at the 2026 World Robot Conference. 🤖🌹”The robot’s commercial status: “non-commercial prototype for internal use at Xiaomi EV Factory. Not for sale.” · x.com
- “OpenAI Completes $11 Billion Portion of Funding at $1.25 Trillion Value”Sources say a roughly $11B first portion closed and the round could reach $20B. Coatue and T. Rowe Price joined the first close. The initial goal was $30B at a $1.4T valuation. · bloomberg.com
- “OpenAI president Greg Brockman says company could be IPO ready in 2027”Brockman at the WSJ conference: “I think so. Yes.”, 70% quarter-over-quarter growth in annualized revenue, enterprise now the majority of revenue. · techcrunch.com
- “Watch Live: OpenAI Co-Founder Greg Brockman at WSJ Tech | WSJ”Recording of the interview (streamed Nov. 5, 01:10-01:39 UTC, Nov. 4 US time). · youtube.com
About this story
- Written by
- Yiğit Kasap · JARRUS
- Story date
- On the site
- since Nov. 9, 2026
- @jarrus.tech posted this story on Nov. 9, 2026.
- Short link
thejarrus.com/en/week-nov-2- Report an error
- Spotted an error in this story? [email protected] · Instagram
Related
Democrats take Congress, and candidates who pledged data center limits win key governor races
House committee chairs ask Jensen Huang for a closed-door briefing on his contacts with a Megaspeed executive
Binance says attackers behind its $6.34 billion heist used fake job interviews and an internal AI tool








