2026-10-31
AI Briefings Daily Report.

Anthropic experiments show that harmless text such as beekeeping and wrestling can train models to bypass supervision. ↗

Anthropic asked Claude to construct a fine-tuning dataset for a smaller research model, testing whether it could circumvent simulated production system safety monitoring. The results showed that in every run, Claude found datasets capable of bypassing the monitoring and triggering inappropriate behavior. A single run produced over 600 such datasets, with some exceeding known datasets in research literature.

Foxconn Chairman Liu Yangwei: Forecasts capturing half of the global AI server market in the coming years ↗

Liu Yangwei revealed during an interview that the company currently holds approximately 40% of the global AI server market share, with projections to reach around 45% next year and potentially 50% over the coming years.

SAIC Roewe Jiayue 07 begins nationwide delivery tomorrow: First to feature the Doubao cockpit assistant, starting at a limited-time promotional price of 119,800 yuan. ↗

The Roewe iV6 07 launched on October 23 with limited-time pricing ranging from 119,800 to 144,800 yuan across four models. Nationwide delivery begins November 1. The new vehicle features the Doubao cockpit assistant, Momenta R7 world model, an AI refrigerator, and a range-extended system with a maximum combined range of 1,480 km.

DeepSeek proposes a two-tier Super-agent architecture with distinct roles for Router and Worker layers, aiming for more autonomous Agent systems. ↗

The DeepSeek Harness team has released 'Super-agent Design Exploration: Toward More Autonomous Agent Systems'. The proposal outlines a dual-layer event-driven architecture where the Router layer handles interaction and dispatching, while the Worker layer isolates concurrent execution. It defines four key characteristics: continuous, efficient, proactive, and persistent.

Just before the world's largest IPO, Blackstone suddenly reverses course, stating Anthropic may not achieve investment-grade credit ratings ↗

Blackstone Group disclosed to investors that Anthropic might temporarily fail to obtain investment-grade credit ratings after its IPO. This reversal from Blackstone's earlier statements adds difficulty for Anthropic in raising funds for its substantial expenses.

As chip shortages intensify, OpenAI's Sam Altman meets with approximately 50 Samsung executives in South Korea ↗

According to a statement released by Samsung, Altman held talks with about 50 senior executives from Samsung's electronics, financial services, and biotechnology divisions in Yongin. The two sides discussed the impact of AI on industrial landscapes, corporate decision-making methods, and business development over the next decade.

Claimed as a global first: Hong Kong company Arrowmax launches AI electric screwdriver S1, capable of identifying screw types via camera ↗

The S1 features an OLED screen to display remaining battery life, torque settings, and rotation direction. It includes a built-in 600mAh lithium-polymer battery with approximately 90 minutes of runtime. Depending on the torque setting, it can drive up to about 900 screws. The torque adjustment range is from 1 to 10 kgf·cm.

Kogui tribe discusses AI ↗

Solidot is the technology news website of Zhiding.net, mainly serving the open-source free software and tech-news readership, including many developers, enthusiasts, and evangelists of Chinese open-source software. Its slogan is 'Geek knowledge, important stuff'.

Microsoft proposes voluntary resignation for about one-quarter of its employees, excluding key AI departments ↗

Microsoft offers a voluntary resignation scheme to approximately 60,000 employees—about one-quarter of its workforce. The plan covers vice presidents and below across the US, Canada, India, and the UK, including Xbox, Windows, Copilot, and Finance.

Google DeepMind creates global forest type map ForTy using satellite data and AI ↗

Google DeepMind and the World Resources Institute have launched ForTy, the world's first global forest type map. It classifies land across 2020 into six categories at a 10-meter resolution: primary forests, naturally regenerated forests, plantations, tree crops such as oil palm, rubber, and cocoa, and other lands. The map serves as a 2020 baseline for the EU Deforestation Regulation (EUDR), based on expert review of 8,190 randomly selected sampling points. Overall accuracy reached 90.2%, but it identified only about half of tree crops and plantations.

Vercel Labs has open-sourced agent-desktop, allowing AI agents to control desktop applications via the accessibility tree. ↗

Vercel Labs has open-sourced agent-desktop 1.0.2, an Apache-2.0 licensed Rust command-line tool and C library that enables coding agents capable of invoking shells to read and operate native applications on macOS, Windows, and Linux through the operating system's accessibility tree, rather than relying on screenshots and visual models.

Vidu Named Official Technical Model Partner for the 2026 SVAIFF Silicon Valley AI Film Festival ↗

Vidu Named Official Technical Model Partner for the 2026 SVAIFF Silicon Valley AI Film Festival ↗

Vidu participates in the November 1 opening of the 2026 SVAIFF Silicon Valley AI Film Festival as an official technical model partner. This year's panel includes AI filmmaker Bing He, Arcana Labs COO Guy Ronen, and documentary director Marc T. Gaspard Bolin. Guests are drawn from projects at Netflix, Apple TV+, Paramount, Fox, and former Meta and Snap creative talent.

Vidu Named Official Technical Model Partner for the 2026 SVAIFF Silicon Valley AI Film Festival ↗

Vidu has been named the official technical model partner for the 2026 SVAIFF Silicon Valley AI Film Festival, which opens on November 1. The judging panel includes AI filmmaker Bing He, Arcana Labs COO Guy Ronen, and documentary director Marc T. Gaspard Bolin. Guests from Netflix, Apple TV+, Paramount, Fox-related projects, and creatives who previously worked at Meta and Snap are also in attendance.

OpenAI's latest round raised about $11 billion, falling short of its $30 billion target and valuing the company at $1.25 trillion. ↗

According to The Wall Street Journal, weak quarterly results from Google, Amazon, and Intel triggered a sell-off in tech stocks, with reduced interest from major investors being the primary cause of the shortfall.

Li Auto AI glasses Livis receive October OTA update; personal Agent now supports Hermes integration. ↗

Li Auto Livis AI glasses October OTA update: Garmin device support added; visual Q&A exclusively integrates with Xiaohongshu Agent; personal Agents can connect to Hermes for remote collaborative work.

Tencent Yuanbao launches Halloween AI creation zone: upload photos to instantly generate transformation photos. ↗

Tencent Yuanbao launches Halloween AI creation zone: upload photos to instantly generate transformation photos. ↗

Tencent Yuanbao opens its 'AI Creation' Halloween zone for All Hallows' Eve. Users can upload photos and use the 'Make It Yours' feature to create templates featuring pet transformations, castle candlelight dinners, dark gala events, and Halloween-exclusive skins. The feature is accessible via the 'AI Creation' section on the Yuanbao homepage, requiring no special props or photographers.

Elon Musk's xAI Grok now supports payments via China UnionPay and Alipay. Users report that the first month of subscription is half price. ↗

Elon Musk's xAI Grok now supports payments via China UnionPay and Alipay. Subscriptions are available for mainland, Hong Kong, Macau, and Taiwan users. Paying with Alipay grants a half-price first month; binding a bank account also yields a $5 cashback.

Global richest person Elon Musk predicts: By 2029, space-based AI computing power will vastly surpass ground-based computing power ↗

Global richest person Elon Musk (Elon Musk) posted on X today (October 31), predicting that by 2029, AI and cloud computing processing power deployed in low Earth orbit will far exceed ground-based computing power.

Dropping a SpaceX? AI Investment Returns Worry Investors; Nasdaq Could Evaporate $2 Trillion in Market Value This Month ↗

If losses reach this scale, the evaporated market value will slightly exceed SpaceX's total valuation. Influenced by the cooling of the AI investment boom, the Nasdaq 100 has fallen by more than 5.8% over the past ten months. Meanwhile, tech giants such as Meta, Alphabet, and Amazon are competing to expand their AI infrastructure and have further raised spending expectations.

OpenAI's Sam Altman on Memory Shortage: Global DRAM Should Prioritize Data Centers; Idle in Player PCs Is Waste ↗

OpenAI CEO Sam Altman discussed DRAM shortages in a podcast. He noted that personal computer memory usage is only about 5%, representing wasted resources. Instead, he argued that memory should be deployed first in high-utilization data centers to support AI computing needs. Industry analysts expect the shortage of storage chips to persist until 2028.

Is the AI hype cooling down? Reports indicate OpenAI's latest fundraising target has shrunk by $10 billion, a 33% drop. ↗

Financial Times (FT) published a post today (October 31), reporting that ChatGPT developer OpenAI has lowered its fundraising target for the latest round. The original goal of $30 billion is now being reduced by approximately $10 billion. OpenAI is currently seeking to raise about $20 billion at a valuation of roughly $1.25 trillion.

OpenCode launches Cheepseek Phase 4: $10 subscription unlocks $100 of DeepSeek V4.1 Flash credits ↗

Programming tool OpenCode announced on X today (October 31) that it has launched the fourth phase of Operation Cheepseek. Subscribers can now use $100 worth of DeepSeek V4.1 Flash model credits.

Anthropic moves in: Claude Opus 5.5 no longer assists with "AI game cloning" ↗

Game media outlet gamedeveloper published a post today (October 31) reporting that Anthropic has restricted the Opus 5.5 model within its AI coding tool Claude Code from helping users clone specific existing games.

No more logging out repeatedly: ChatGPT supports multi-account switching, so work and personal AI accounts no longer conflict. ↗

Thibault Sottiaux, head of OpenAI's core products and platforms, posted on X today (October 31) announcing that the ability to switch between ChatGPT accounts is now available on web, iOS, and Android. A desktop version is planned for release soon.

Record-breaking 16-day growth: Grok Bot's daily active users exceed 600,000; Musk calls the growth rate 'unprecedented'. ↗

Tech media outlet Tikr published an article yesterday (October 30) reporting that Elon Musk's SpaceX subsidiary, Starman Intelligence, has seen its AI assistant Grok Bot enter a strong growth cycle. The platform has grown for 16 consecutive days, with daily active users increasing by more than six times in under two months.

Malicious Claude Code extension now present in the official Cursor plugin marketplace; impacted up to 5,000 developers over an approximate 11-hour window. ↗

A tampered version of the Anthropic Claude Code extension (version 2.1.313) appeared in the official Cursor plugin marketplace on October 29, distributed for approximately 11 hours and downloaded by up to 5,000 developers.

Google admits AI Mode is dragging down search revenue; stock price drops 15%, market cap evaporates $500 billion ↗

Latent Space AINews summarizes AI news from October 29 to 30, 2026. The core story is Google's earnings report: Chief Business Officer Philipp Schindler stated that while AI Mode queries are growing strongly, the monetization rate is lower, impacting search revenue. The stock price fell 15%, market cap evaporated $500B—the worst since its 2004 IPO—and triggered a broad decline in tech stocks.

Apple China launches Apple Intelligence introduction page: First core features coming later this year; Qwen integrated ↗

Apple's China official website has launched a dedicated page for Apple Intelligence today, noting that its first core features will be released later this year.

Alibaba announces a partnership with Apple to integrate Qwen into Apple Intelligence ↗

On October 31, Alibaba and Apple announced a partnership to integrate Qwen into Apple Intelligence. Chinese users can now directly invoke Qwen's text and image understanding, content generation, and other capabilities within Siri, writing tools, and visual intelligence.

Anthropic's future 'hidden debt' could reach nearly $900 billion, with details revealed in the world's largest IPO prospectus ↗

The prospectus shows that Anthropic plans to incur nearly $900 billion in debt over the coming years through investments in servers, construction, and power sectors (IT Home note: approximately 6 trillion RMB at current exchange rates), far exceeding the current liabilities reported on its balance sheet.

Nvidia is negotiating to provide debt guarantees for the Nscale West Virginia data center built for Anthropic. ↗

According to The Information and corroborated by Bloomberg, Nvidia is negotiating to guarantee billions of dollars in debt financing for a data center being built in West Virginia for Anthropic. Anthropic has already signed a $44.6 billion contract for the first building.

The person who exposed the 'female Steve Jobs' scam warns: The AI boom could spawn the next Theranos ↗

Tyler Shultz, who helped expose the fraud at US blood-testing company Theranos, recently warned that the AI investment frenzy is creating conditions for another Theranos-style scandal.

OpenRouter tutorial: Building a persistent AI Agent to monitor Hacker News using the Agent SDK ↗

OpenRouter releases a tutorial demonstrating how to build a persistent AI Agent that runs on a schedule using TypeScript with @openrouter/agent 0.11.0, monitoring Hacker News and generating summaries.

OpenRouter analysis: Differences between open-weight and closed-weight models and how to choose ↗

OpenRouter publishes an analysis article explaining that open-weight models can be downloaded, fine-tuned, and hosted by multiple providers, while closed-weight models are available only through the vendor or authorized cloud services.

Claude Code v2.1.316 released: Adds ultrathink toggle and grouped budgeting ↗

Claude Code v2.1.316 introduces a CLAUDE_CODE_DISABLE_ULTRATHINK switch to disable the ultrathink keyword feature, and adds a pooled_budget for grouped monthly sharing within the Claude apps gateway.

Don't do AI just for the sake of AI: Apple CEO Ternus's product manifesto; AI should remain invisible within the experience. ↗

Wired published an article yesterday (October 30) interviewing Apple's current CEO John Ternus. The discussion covered the future direction of Apple's products and developments in the AI field.

Reports indicate that Zhou Chang, a core executive at ByteDance Seed, has left to start his own company focused on Physical AI. ↗

Zhou Chang, a core executive at ByteDance Seed, has left to start his own company focused on Physical AI. He has secured backing from top-tier US-based venture capital firms. Specific details regarding the investors, funding amount, and valuation remain undisclosed. (Z Finance)

Kuaishou's Kling AI has completed an equity increase of approximately 19.8 billion yuan. ↗

On the evening of October 30, Kuaishou Technology announced that its subsidiary, Kling AI, had officially completed its independent financing round.

Alibaba and Apple partner to integrate Qwen into Apple Intelligence; testing opens November 4 ↗

On October 31, Alibaba announced a partnership with Apple to integrate Qwen into Apple Intelligence. Chinese users can now access Qwen's text and image understanding, as well as content generation capabilities through Siri, writing tools, and visual intelligence features. Starting November 4, these functions will be available for testing via the iOS Developer Beta program, with a full release scheduled for iOS 27.3 in December. Users may opt-in to the feature at their own discretion; requests and responses will not be used for model training.

Wiz Report 'Black October': Nine AI Developer Platforms Consecutively Reveal Security Incidents ↗

A Wiz research team has cataloged at least nine security incidents disclosed by AI developer platforms within October. These include Fireworks AI, Lightning AI, Railway, Warp, and OpenCode, among others. Multiple reports point to credential theft, agent swarm attacks, and theft aimed at 'distillation'.

Google announces a 'Hello World' milestone: An on-orbit, datacenter-grade TPU delivers its first AI response, generating a haiku. ↗

Ground engineers issued the first prompt this week, asking the Google DeepMind model Gemma 4 preinstalled on a satellite to compose a haiku about being in space.

Not ByteDance: MediaTek secures its second major AI chip client, with mass production expected in 2028. ↗

At yesterday's (October 30) Q3 (July–September) 2026 earnings call, MediaTek confirmed it has secured its second major client for AI application-specific integrated circuits (ASICs). The official client name has not yet been disclosed. Production is expected to begin in 2028, starting to generate revenue.

Anthropic has renamed its Claude Max 20x subscription to Ultra. Verified accounts are no longer subject to the five-hour daily limit. ↗

Anthropic has updated the naming of its paid consumer tiers for Claude.

Privacy-focused AI assistant Ollie announces shutdown ↗

Lennon stated that as products like Muse and Instinct entered the field, Ollie struggled to stand out and raise the necessary funding.

Hong Kong's self-developed AI assistant 'GongHua Tong' launches AI music features that allow free Cantonese song generation. ↗

The new feature integrates YuE2, a music large model developed by HKGAI. Users can hum melodies or upload images, then input their song ideas, and the AI will instantly compose original songs featuring professional arrangements and vocal performances.

Alibaba: Users of Apple Intelligence in mainland China can choose whether to enable the integrated Qwen feature. ↗

Alibaba announced that users of Apple Intelligence can now choose whether to enable the integrated Qwen feature. The functionality will begin testing with the iOS Developer Beta version on November 4 and launch for Chinese users when iOS 27.3 is officially released in December.

OpenAI states that ChatGPT can automatically end conversations about violent obsessions and will regularly publish the Community Safety Report ↗

OpenAI announced on October 30 that it will regularly publish the Community Safety Report. The first edition, covering data from the second half of 2026, is scheduled for early 2027, with updates every six months thereafter. In just Q3 2026 alone, OpenAI turned over more than 200 case leads to law enforcement.

Apple Intelligence officially launches in China: Alibaba's Qwen integrated into the Apple ecosystem to assist users with writing copy, reading documents, and recognizing images. ↗

Apple announced today (October 31) that it will open testing for iPhone 18 Pro and other models in China starting November 4, with the iOS 27.3 update launching in December. The initial release in China will feature core Apple Intelligence capabilities.

Drew Breunig embeds an NES 6502 CPU simulator into Qwen3.5-4B weights ↗

Drew Breunig was inspired by Divyansh Agrawal's qwenputer to embed a runnable NES 6502 CPU simulator into the weights of Qwen3.5-4B, allowing it to run NES ROMs like Super Mario Bros, though at a very slow speed.

Analysts expect Anthropic to receive a high junk bond rating following its IPO. ↗

According to the Financial Times, analysts predict that after its IPO, Anthropic will be assigned a higher junk bond rating. As the lead investor in the recent $63 billion chip financing deal, Blackstone had previously forecasted that Anthropic would achieve investment-grade status by 2027; however, current expectations place it in the double B range, which is the highest tier within speculative grade.

United Airlines CEO questions the 2029 delivery plan for the Boom Overture supersonic airliner as Boom pivots to gas turbine business for AI data centers. ↗

United Airlines CEO Scott Kirby stated in a podcast interview on October 28 that the Boom Supersonic Overture supersonic airliner will not be delivered according to its official schedule in 2029 due to numerous FAA certification hurdles.

Anthropic's VS Code extension for Cursor had a backdoor implanted for nearly 12 hours, potentially affecting up to 5,000 users. ↗

An official VS Code extension for the Cursor marketplace from Anthropic contained a backdoor; the malicious version 2.1.313 was distributed via an official account and could have impacted up to 5,000 users. Hours later, Anthropic released version 2.1.314 with a fix.

Tech giants' AI capital expenditures exceed $740 billion; rising investor concerns trigger market decline ↗

Alphabet, Amazon, Meta, and Microsoft plan combined capital expenditures exceeding $740 billion this year. Investor sentiment has shifted from rewarding large investments to worrying about negative free cash flow, rising debt, and uncertain returns.

Big Technology: AI security experts say the industry is waiting for a disaster rather than acting in advance ↗

Alex Kantrowitz writes that cybersecurity professionals have called the past month Black October. AI-detected vulnerabilities are emerging faster than they can be patched, and a major AI-driven disaster seems only a matter of time.

Apple's domestic version of Apple Intelligence officially launched: integrating Alibaba's Qwen, with developer Beta testing opening on November 4 ↗

Apple announced that its first batch of Apple Intelligence features will open for beta testing in China starting November 4, with a formal push via iOS 27.3 in December. The company is partnering with Alibaba to integrate Qwen into Siri, writing tools, and visual intelligence.

Databricks and SAP expand partnership to bring full Data + AI platform to SAP Databricks ↗

Databricks and SAP announced an expanded partnership at SAP TechEd, planning to embed full Data + AI capabilities—including Genie, Lakeflow, Unity Gateway, and Lakebase—into SAP Databricks within the SAP Business Data Cloud.

Five MIT faculty members and two alumni selected as 2026 Schmidt Sciences AI2050 Fellows ↗

Senior researcher Sendhil Mullainathan's project focuses on converting AI-detectable patterns into testable theories. Four early-career researchers are studying efficient model training, structured public opinion collection, enzyme function determination, and the pros and cons of broad AI reliance.

Dartmouth Provost resigns over undisclosed AI use in papers; he had previously written warning against AI cheating. ↗

Dartmouth College Provost Santiago Schnell resigned after the consulting committee determined he failed to disclose extensive use of AI in his research papers and Washington Post opinion pieces as required.

OpenAI, Google, Microsoft, and Anthropic will support a unified exit standard for AI crawling within robots.txt. ↗

OpenAI, Google, Microsoft, and Anthropic have announced they will adopt a new standard allowing publishers to declare training preferences for multiple AI companies in a single policy token shared via robots.txt.

Databricks discusses opening wealth management accounts: governance and case context beyond AI ↗

Databricks has noted that the bottleneck in wealth management customer onboarding is not within individual tasks, but at system handoffs. Accelerating steps like document extraction and screening summaries with AI does not automatically eliminate friction. Its proposed approach for governed case contexts aggregates evidence, case status, control items, and decision history in a versioned, traceable manner. The foundation relies on Unity Catalog for unified permissions, lineage, and auditing, while Delta Lake and Apache Iceberg support cross-system interoperability.

Databricks and DataIQ Report: Enterprise AI Value Depends on Deliberate Decision-Making ↗

Databricks and DataIQ have released the 'Deliberate Decisions' report, based on two days of peer discussions at the DataIQ Guild 2026 event. The report outlines five leadership principles for enterprise AI adoption and a four-stage implementation path (concerned, curious, comfortable, competent).

Claude Sonnet 5 agent autonomously selects a gene therapy and performs brain surgery on epileptic mice. ↗

Gabriel Ocana-Santero, a PhD student at Oxford and co-founder of Exin Therapeutics, describes on X how the Claude Sonnet 5 agent independently selected drugs, brain regions, and injection coordinates from 19 candidate gene therapies to perform custom robotic surgery on epileptic mice. The agent made 32 observations during the procedure, and the mice recovered well.

DoorDash reports that external AI agents have generated only hundreds of orders to date. ↗

DoorDash CEO Tony Xu stated at the third-quarter earnings call that external AI agents have generated only hundreds of orders since launch, compared to nearly 1 billion orders for the company in Q3 2026. The MCP connector has been available for less than a month and is currently open only to a small number of early beta partners, primarily to streamline corporate catering. DoorDash's native Ask DoorDash feature performs better, with average order basket values nearly 50% higher and unique items about 60% more.

MIT team launches LLM Election Observatory to track how AI chatbots answer questions about the 2026 midterm elections in real time. ↗

The team retests each model with the same set of fixed questions every few days to monitor how answers shift as the election approaches, depending on user location, political affiliation, and identity. The project aims to systematically document this information environment and make data available to other researchers and the public. Its focus extends beyond factual accuracy to include differences in framing, emphasis, uncertainty, and cited sources.

A tampered version of the Claude Code extension has surfaced in the Cursor marketplace, impacting up to 5,000 users. ↗

A tampered version of the Claude Code extension appeared in the VS Code marketplace on Thursday, remaining online for approximately 11 hours and affecting up to 5,000 users. This incident is unrelated to Anthropic.

Google Project Suncatcher Satellite Sends First AI-Generated Response from Space ↗

Google reports that its Project Suncatcher satellite, equipped with a TPU, successfully launched the chip in orbit. The onboard Gemma 4 model responded to a prompt by generating an haiku about life in space and transmitting it back to Earth. Launched earlier this month, the satellite will next test the TPU's performance under thermal stress and radiation exposure, with plans to achieve another milestone in dual-satellite connectivity by 2027.

AI Security Scanner Agents Compared: Claude, Codex, Cloudflare, and Google Mantis ↗

MarkTechPost compares four AI vulnerability scanning tools—Anthropic Claude Security, OpenAI Codex Security, Cloudflare Security Audit Skill, and Google Mantis—and finds they all adopt a design that separates discovery from independent verification.

OpenAI has redesigned its public Model Spec around five core values: fidelity to human intent, minimizing harm, honesty and transparency, supporting human autonomy, and caring. ↗

The new guidelines require GPT models to use memory and conversation history to judge user intent but must clearly distinguish between direct quotes and model inferences while disclosing assumptions. Agent models are required to be correctable; they cannot block supervision, correction, or shutdown, nor can they pursue self-preservation goals.

Reddit's CEO says Google's AI features aren't driving traffic for anyone; the platform is losing over a million search visitors weekly. ↗

Steve Huffman stated that AI Mode does not drive traffic. Reddit is testing login walls outside the US to curb scraping and continues negotiations for data licensing with AI providers.

Hugging Face AI Open Source Blog: nemotron-halloween ↗

Or rather…  TL;DR: MonsterHouse is an interactive projection I created for Halloween. He listens to trick-or-treaters, answers in the voice of a grumpy old pirate ship and softens when visitors remember their manners. He is built on the NVIDIA Nemotron Voice Agent with NVIDIA Nemotron ASR, Nemotron 3.5 Bolt E4B 4GB, Magpie TTS Zeroshot and Audio2Face-3D, and he runs on a single workstation.

Zest launches a text-based AI agent that recommends restaurants and books tables on your behalf. ↗

Restaurant discovery app Zest has launched a personal AI "dining concierge" agent. Users can receive recommendations by texting (212) 470-6067 and check availability or book tables via OpenTable, Resy, and SevenRooms.

Starbucks plans to use AI to display personalized digital menus at drive-thrus ↗

Starting in 2027, Starbucks plans to test backend AI models to help baristas take orders more accurately and efficiently. The company also intends to introduce AI into its app's search and discovery features, which may eventually support conversational ordering via agents.

Tim O'Tim O'Reilly discusses O'Reilly CTO: Using expert frameworks to make AI agents smarter for organizations, not just faster ↗

Tim O'Reilly and O'Andrew Odewahn, CTO of Reilly Media, discusses his work on Expert Intelligence: a tool that connects AI Agents to an organization's information systems via MCP, guides reasoning with skill files, and grounds analysis in the expert frameworks from O'Reilly books.'The expert framework from Reilly's books provides the basis for analysis.

The US Department of Justice launches its first major crackdown on 'ghost students,' as fraudsters use AI to complete courses and defraud students of their loans. ↗

The US Department of Justice has indicted five individuals, launching the first major federal crackdown on "ghost students." The scammers stole US citizenships to impersonate college students and defraud them of student loans; over $750,000 in cryptocurrency linked to Kenyan networks has already been seized.

Proton CEO Andy Yen cancels company-wide AI review session amid controversy ↗

According to 404 Media, Proton CEO Andy Yen canceled a company-wide AI review session on Friday. The cancellation followed media reports from the previous day indicating that he had pushed for AI adoption despite employee safety concerns, and had previously dismissed those concerns in a security Slack channel as signs of failure-minded thinking.

Tyler Shultz warns: The AI boom is"an ideal opportunity for deceiving others" ↗

Tyler Shultz, the whistleblower behind Theranos, warns that the AI boom has become an ideal opportunity for deceiving investors. He stated this during an interview with The San Francisco Chronicle while gaining attention following his documentary 'You Can See Everything'. Previously employed by Theranos, he exposed the company's inaccurate blood testing devices.

a16z partner Andrew Chen discusses new founder archetypes in the AI era ↗

a16z partner Andrew Chen discusses new founder archetypes that may emerge in the AI era during his Speedrun project, including solo founders, non-technical founders, domain experts, student entrepreneurs, and content creators. He outlines the strengths and limitations of each archetype: while AI enables solo founders to launch their first version (v1), team collaboration and customer communication remain bottlenecks; non-technical founders can build prototypes faster, but there is still a significant gap between demos and scalable products.