A top-lab personal-tier rename that also drops Ultra's 5-hour wall for heavy users.
AI NEWS
Today’s AI News
Chinese and English are the same day. This page lists every item, each with the original link, the author, and the engagement.
Latest day
October 31, 2026
22 items for 2026-10-31, ranked by combined engagement. Unverified items are marked.
Top 3
A first-party case of supply-chain risk in the developer-tool stack.
OpenAI Private Intelligence: stronger data controls for sensitive enterprise work
A data-and-compliance path for putting frontier models on sensitive workloads.
All
Claude renames Max plans: Max 5x→Max ($100), Max 20x→Ultra ($200); verified Ultra drops the 5-hour limit
A top-lab personal-tier rename that also drops Ultra's 5-hour wall for heavy users.
CollapseClaude renames Max plans: Max 5x→Max ($100), Max 20x→Ultra ($200); verified Ultra drops the 5-hour limit
Claude's official account renamed its high-usage personal tiers: Max 5x becomes Max at $100/month; Max 20x becomes Ultra at $200/month; existing subscribers need do nothing. Verified Ultra accounts lose the rolling 5-hour limit starting today for most long-time users; newer accounts keep a temporary cap until verified to curb fraud. Weekly limits are unchanged. Pricing page confirms Ultra at ~10× Pro weekly usage and $200/month API credits; Max at 5× Pro and $100/month API credits. Figures from the official post and claude.com/pricing.
Cursor: malicious modified Claude Code extension via marketplace hit ≤~5,000 users; yanked and upgrade to 2.1.314
A first-party case of supply-chain risk in the developer-tool stack.
CollapseCursor: malicious modified Claude Code extension via marketplace hit ≤~5,000 users; yanked and upgrade to 2.1.314
Cursor says a security incident involved a modified Claude Code VS Code extension distributed via its IDE marketplace, affecting at most about 5,000 users. The malicious build was not from Anthropic and does not affect extensions from other channels such as Open VSX, or products like Grok Bot/Grok. Cursor took the marketplace down, worked with providers to neutralize the malware, restored it, and asks customers to update to Claude Code for VS Code 2.1.314; remediation steps are on cursor.com/blog/claude-code-extension-security-notice. No confirmed further exploitation is stated; company-reported.
OpenAI Private Intelligence: stronger data controls for sensitive enterprise work
A data-and-compliance path for putting frontier models on sensitive workloads.
CollapseOpenAI Private Intelligence: stronger data controls for sensitive enterprise work
OpenAI launched Private Intelligence, saying it gives businesses stronger controls over their data while using its most capable models, built with large customers for sensitive work. The business page frames it for enterprise and government use with isolation and compliance options (FedRAMP High, DoD IL5, HIPAA and related claims—verify on the page). The post itself lists no prices or customer names; details at openai.com/business/private-intelligence/. Company product messaging.
Demis Hassabis posts Artificial Analysis Intelligence Index v5: Gemini 4 Carbon (high) leads at 59
A lab CEO amplifying a third-party composite leaderboard win.
CollapseDemis Hassabis posts Artificial Analysis Intelligence Index v5: Gemini 4 Carbon (high) leads at 59
DeepMind CEO Demis Hassabis posted “No.1” with a chart of Artificial Analysis Intelligence Index v5.0. The chart aggregates nine evals (including HLE-Diamond, Terminal-Bench 4.0, FrontierCode 2.0): Gemini 4 Carbon (high) leads at 59 (marked no public API), then Claude Opus 5.5 (max with fallback) at 58 and GPT-6 Astra (max) at 57. It is a third-party Artificial Analysis ranking, not an official contest; several top models are marked without a public API, so availability needs separate checks.
Cartesia launches Managed Agents: end-to-end voice agents, claims #1 on Speech Agent Arena and EVA-Bench
Voice agents moving from model bake-offs to end-to-end enterprise platforms.
CollapseCartesia launches Managed Agents: end-to-end voice agents, claims #1 on Speech Agent Arena and EVA-Bench
Cartesia launched Managed Agents, an end-to-end enterprise voice-agent platform co-designed with its voice models and harness. Its blog cites Artificial Analysis Speech Agent Arena preference Elo 1115 (#1 of 30) and ServiceNow EVA-Bench accuracy 87.4% / experience 90.7% (GPT-5.5 as reasoner), plus tens of millions of conversations already in finance, retail, hospitality and healthcare. Features include multilingual agents, tool calls, knowledge bases and outbound batches. Benchmarks are third-party/partner and company-reported; no independent replication cited.
Artificial Analysis launches HLE-Diamond: refined 1,000-question HLE; Gemini 4 Carbon ~69% leads
A cleaned hard exam that makes the frontier gap easier to read.
CollapseArtificial Analysis launches HLE-Diamond: refined 1,000-question HLE; Gemini 4 Carbon ~69% leads
Artificial Analysis added an HLE-Diamond leaderboard inside Intelligence Index v5. CAIS and Scale AI spent about a year refining Humanity's Last Exam into 1,000 closed-ended items, half reasoning and half knowledge, 272 with images. Runs are tool-free with the official system prompt; GPT-5.6 Luna (medium) grades with the official judge prompt (pass@1). AA says only ~13% of model variants clear 50%: Gemini 4 Carbon (high) leads at ~69%, Claude Opus 5.5 (max) ~68%. Figures are Artificial Analysis's own runs.
Artificial Analysis Progress Index: one IRT capability scale for language models since 2022
Turns “how much models improve per year” into a reusable scale.
CollapseArtificial Analysis Progress Index: one IRT capability scale for language models since 2022
Alongside Intelligence Index v5, Artificial Analysis launched the Progress Index: an item-response-theory (IRT) single capability score from tasks across its evals so models since 2022 can be compared over time. The Intelligence Index remains the guide for choosing a model today. AA says that since reasoning models arrived the frontier has risen roughly 65–70 points a year, about double the prior pace (company figure). Method draws on Epoch's Capabilities Index and Anthropic's Claude Mythos Preview system-card scale. Intelligence Index scores are not comparable across versions.
OpenAI Agents API weekly ship: Auto Review, async compaction, time awareness, mid-session tool updates, system-prompt cache
Practical runtime safety and cost controls for agent builders.
CollapseOpenAI Agents API weekly ship: Auto Review, async compaction, time awareness, mid-session tool updates, system-prompt cache
OpenAI Agents API engineer Steve Coffey listed this week's ships: Auto Review to score commands for bad behavior before running (Codex-like); Async Compaction to speed sessions by compacting in the background; experimental time awareness for queries like “timebox CI fixes to 10 minutes”; mid-session tool updates; and automatic system-prompt caching across sessions. OpenAIDevs amplified the same day. First-party engineer release notes; no independent benchmarks.
OpenAI Devs: community builder uses GPT-6 Astra to remotely drive a robot arm
A community signal of frontier models driving physical manipulators.
CollapseOpenAI Devs: community builder uses GPT-6 Astra to remotely drive a robot arm
OpenAI's developer account featured builder Scav's robotics community experiment in which GPT-6 Astra remotely controlled a robot arm. It is a showcase; no success rate, latency or safety constraints are given. How “GPT-6 Astra” maps to public product tiers should be checked against OpenAI's own docs; here we only relay the developer account's wording.
Claude iOS voice mode comes to Apple CarPlay; mobile also gains multi-account, 33 languages, large-screen layout
Car and multi-device UX that matters for mobile work.
CollapseClaude iOS voice mode comes to Apple CarPlay; mobile also gains multi-account, 33 languages, large-screen layout
Claude's official thread: iPhone voice mode now works with Apple CarPlay; iOS/Android support staying signed into multiple accounts; the app covers 33 languages (24 new, including Bengali, Danish, Swahili, Thai, Turkish, Ukrainian, Vietnamese); on iPad and iPhone Duo, artifacts open beside chat and can go full screen; voice answers sooner in noise/quiet speech and sounds more natural outside English. Product UX update; no independent eval.
a16z: companies using AI now spend about $1,024 per employee on it, nearly 2× last year
A public per-employee anchor for enterprise AI budget talks.
Collapsea16z: companies using AI now spend about $1,024 per employee on it, nearly 2× last year
a16z's Charts of the Week says companies using AI now spend an average of about $1,024 per employee on it—nearly 2× last year. Sample size, what counts as spend (compute vs seats vs API), and survey source are not spelled out in the post; check the chart/source. Useful as a budgeting signal; a16z chart figures, not audited spend.
Mastercard Agent Pay comes to Shop Pay with agentic tokens that keep card details private
Agent checkout meeting card-network tokenization for real.
CollapseMastercard Agent Pay comes to Shop Pay with agentic tokens that keep card details private
Mastercard says Agent Pay now works with Shop Pay so shoppers can buy through agents while card details stay behind agentic tokens rather than being exposed to the agent. It is a concrete agent-commerce link between a card network and checkout. Merchant coverage, regions and fees need the parties' notices; no volume figures in the post.
Shopify: Shop Pay buys from any brand inside Grok Bot, Claude Duet and Instinct
Checkout inside the chat—retail agents moving from browse to buy.
CollapseShopify: Shop Pay buys from any brand inside Grok Bot, Claude Duet and Instinct
Shopify's short video says Shop Pay now lets people buy from brands on or off Shopify inside agents including Grok Bot, Claude Duet and Instinct. It pairs with the same week's Mastercard Agent Pay move. No conversion or GMV figures; product demo messaging.
Agility Robotics expects ~$465M raise and Churchill XI deal; aims to list as AGLT on Nov 5
A humanoid/warehouse robotics firm racing toward a public listing.
CollapseAgility Robotics expects ~$465M raise and Churchill XI deal; aims to list as AGLT on Nov 5
Agility Robotics says its business combination with Churchill Capital Corp XI expects about $465M in gross proceeds and, subject to Churchill XI shareholder approval and customary closing conditions, trading on Nasdaq as AGLT beginning November 5. Primary source is the company release. Amount and listing date remain expected until close.
Tesla amplifies F1 champ Nico Rosberg calling Cybertruck FSD in SF “life-changing”
A top pro driver publicly endorsing supervised FSD.
CollapseTesla amplifies F1 champ Nico Rosberg calling Cybertruck FSD in SF “life-changing”
Tesla amplified a media post with the line “even the world's best drivers like being driven”: F1 world champion and Rosberg Ventures founder Nico Rosberg called testing a Cybertruck with FSD(S) around San Francisco “life-changing” and said he is willing to hand over control. Endorsement-style testimony, not a controlled road study; FSD remains subject to local rules and driver responsibility.
NVIDIA Robotics: frontier AI agents use Isaac tools to build sim worlds and train robot skills
Sim-to-real robot-skill pipelines getting prime marketing time.
CollapseNVIDIA Robotics: frontier AI agents use Isaac tools to build sim worlds and train robot skills
NVIDIA Robotics posted a video of frontier AI agents using NVIDIA Isaac tools to build simulation worlds, then train and evaluate robot skills, with Agile ONE SN helping bring skills into the real world. Product/ecosystem demo; no task success rates or deployment scale. A toolchain signal for manufacturing and robotics developers.
Brett Adcock: Hark launches on Mac desktop; download at hark.com/mac
Another consumer AI desktop client shipping.
CollapseBrett Adcock: Hark launches on Mac desktop; download at hark.com/mac
Brett Adcock (Figure-linked founder) announced Hark on Mac, same experience as existing Hark, download at hark.com/mac. No feature matrix, pricing or user counts in the post. A personal product launch, not a Figure humanoid hardware release.
Elon Musk: think of Grok Bot as a work colleague
Platform framing shifts from toy bot to colleague metaphor.
CollapseElon Musk: think of Grok Bot as a work colleague
Elon Musk posted “think of Grok Bot as a work colleague” with accompanying media. No feature list, pricing or enterprise SLA. Product-positioning guidance; details should come from xAI/Grok official docs.
NVIDIA GTC Berlin: CUDA-X–powered graphics and simulation showcase
European GTC spotlight on simulation and industrial software.
CollapseNVIDIA GTC Berlin: CUDA-X–powered graphics and simulation showcase
From GTC Berlin, NVIDIA posted a showcase video of computer graphics and simulation powered by NVIDIA CUDA-X software for scientists and engineers testing ideas. Event theater; the post does not announce a new chip or model SKU.
OpenAI brand film: “There are many futures. Which do you want to build?”
High-engagement brand narrative reflecting public interest in AI's direction.
CollapseOpenAI brand film: “There are many futures. Which do you want to build?”
OpenAI released a brand film with the line “There are many futures. Which do you want to build?” No product, price or technical claims. High engagement; narrative/recruiting creative, not a feature launch.
Insilico×Servier AI oncology collab hits early discovery milestone in ~10 months; payment triggered
A checkable paid milestone in AI pharma, not just a concept deck.
CollapseInsilico×Servier AI oncology collab hits early discovery milestone in ~10 months; payment triggered
Insilico Medicine said its AI-powered oncology collaboration with Servier hit an early drug-discovery milestone about 10 months in, triggering a milestone payment from Servier. Industry outlets such as Drug Discovery World also covered it. Target, structure and payment size are not in the post; company/partner milestone messaging.
NEJM AI Perspective: themes shaping clinical AI in 2026 and priorities for safer rollout
An authoritative lens for clinical AI implementation priorities.
CollapseNEJM AI Perspective: themes shaping clinical AI in 2026 and priorities for safer rollout
NEJM AI flagged a new Perspective on major themes shaping clinical AI in 2026 and priorities for safer, more effective implementation. Full text via the official link. The post does not list the recommendations; read the paper.
Past AI News
Earlier days of AI News stay here.