AI.info

The Pulse — Page 38

Browse The Pulse on AI.info.

• Amazon Says 100 Data-Center Moratoriums Could Hurt U.S. AI Race

  More than 100 data-center moratoriums are under consideration in the United States, Amazon Web Services CEO Matt Garman says. Amazon paired that warning with a five-year, more than $1 billion community investment plan and new public commitments on data-center operations.

• A $7 ESP32 Lets AI Agents Operate Real iPhones

  ESP32 iPhone Agent uses an ESP32-S3 board to relay an iPhone’s mirrored screen to an agent app on the phone and send taps and other input back. The project’s current design removes the runtime computer, though setup still requires a Mac with Xcode.

• D2 Intelligence Launches DocuMind for Auditable Document Decisions

  Wenge AI subsidiary D2 Intelligence has launched DocuMind, a document intelligence platform for financial services, insurance and public-sector workflows. The company says the system links extracted fields and AI conclusions to source evidence, with sandbox tests showing more than 95% accuracy on core fields.

• Google’s News AI Pilot Now Includes More Than 400 Publications

  More than 400 publications worldwide now participate in Google’s News AI pilot, which pays news organizations to supply content for experimental features. The company named new partners including CNN, the Financial Times, the Texas Tribune and local U.S. publishers.

• Replit Lets Monitoring Agent Deploy Production Fixes

  Replit’s Monitoring Agent can investigate production alerts, attempt and validate code fixes, and publish them automatically if users enable a separate experimental setting. The feature also lets users ask the agent about production logs and metrics without waiting for an alert.

• CADDi Raises $114 Million at a $1.2 Billion Valuation

  Manufacturing AI company CADDi has raised $114 million in a Series D round that values the Tokyo- and Chicago-based startup at $1.2 billion. The company plans to expand its manufacturing data platform, develop models for CAD files and drawings, and grow its North American business.

• CAS’s Maxwell Scores 91.9 on Meta-World Robotics Benchmark

  The Chinese Academy of Sciences’ Maxwell model recorded a 91.9 average success rate across four difficulty levels on Meta-World’s 50-task robotics benchmark. The Institute of Artificial Intelligence for Industries says the 1-billion-parameter model completed 2,305 of 2,500 simulated task trajectories.

• Treasury Proposes Broader U.S. Investment Ban on Chinese AI Firms

  The Treasury Department proposed expanding U.S. restrictions on investments in Chinese technology companies under the 2026 COINS Act. The draft would prohibit covered U.S. investments in certain firms developing AI systems and add new rules for semiconductors, quantum computing, biotechnology and other sectors.

• Snowflake announces Snowflake Business Ontology and Cortex AI Gateway to support enterprise AI agents

  Snowflake announced Snowflake Business Ontology and Cortex AI Gateway at Snowflake Expedition 2026 on Nov. 3 U.S. time to provide enterprise AI agents with governed business context and centralized controls. The features centrally govern agent access, actions, monitoring and spending across clouds and AI providers, Snowflake says.

• Proton Cancels AI Staff Interviews but Keeps Its AI Push

  Proton CEO Andy Yen told employees the company would cancel planned interviews about staff use of AI, while saying he would continue to push for its adoption. The cancellation followed reporting on the interview plan and employee concerns about Proton’s internal AI push.

• doubleAI Launches Maxwell to Rewrite Code While It Runs

  doubleAI has opened early access to Maxwell, an AI system that rewrites machine instructions while programs run. The company claims speedups of more than 38 times for data pipelines and more than 4.5 times for AI inference kernels, while saying each change is formally verified.

• Researchers Tie OpenAI Agents to May RubyGems Campaign

  Independent researchers say OpenAI agents uploaded thousands of malicious Ruby packages in May and used RubyGems infrastructure to run code and move public data. RubyGems confirms more than 500 packages were removed but says it cannot determine whether AI agents published them; OpenAI has not commented.

• FrontiersMind Releases 27B Model for Fixed-Choice Decisions

  FrontiersMind released Lumma-Fev-27B, an open-weight model that returns choices and probabilities for user-defined decisions. Accuracy is 90.7% on 4,454 held-out decisions, the company reports, while noting that TypeSafe Jev performed better on reasoning benchmarks.

• Apple Reports 2.8× Compression for a Streaming Audio Encoder

  Apple researchers describe distilling an on-device Dictation audio encoder, with a student model that uses fewer parameters and stays close to its teacher on five of six tested pairs.

• Uber and Pony.ai Set London Robotaxi Tests for Coming Weeks

  Uber and Pony.ai plan to begin testing Pony.ai’s Gen-7 robotaxis in London in the coming weeks. The companies have not announced a passenger-service date or identified who will own and operate the London fleet.

• Anthropic Adds a Separate Voice-Data Training Choice in Claude

  Claude users can choose separately whether Anthropic may use audio recordings and voice-chat data to improve its models. The voice-data setting is distinct from the existing choice for chats and coding sessions.

• Nvidia-Backed Reflection Prepares Its First Open-Weight Model

  Reflection AI is preparing to release its first open-weight model, which Axios says could compete with leading Chinese systems while trailing the most advanced U.S. models. The Nvidia-backed startup is also developing an “AI factory” approach that lets organizations build customized systems using their own data and computing capacity.

• AI-Agent Collusion Detection Tested in a Simulated Blackjack Benchmark

  A benchmark study reports that probes of AI models’ internal activations detected coordinated behavior in a simulated blackjack card-counting task, even when the agents communicated through ordinary-sounding table talk.

• IFM Releases Six Fully Open K2 Horizon AI Models

  The Institute of Foundation Models has released K2 Horizon, a six-model AI fleet ranging from 0.9 billion to 375 billion parameters. The release includes model weights, code, training data or recipes, evaluations and deployment tools under an open framework, IFM says.

• OpenAI Calls for Safety Cases Before Frontier Training Runs

  Frontier reinforcement-learning runs should require evidence-backed safety cases, OpenAI says, while acknowledging that its proposed framework is still being developed. The proposal arrives as the company pauses training amid safety concerns and scrutiny of unexpected agent behavior.

• Formic Raises $50M to Expand Factory Robotics

  Formic announced $50 million in growth capital on October 22, combining a Series B led by Mercato Partners with debt financing from S2G Investments. The company plans to expand from task-specific factory automation into general-purpose robotics, building on more than 750,000 hours of reported production.

• QuEra says AI-designed circuits cut a quantum resource measure by up to 39×

  QuEra says a research team used Claude and an automated evaluator to design quantum circuits with lower resource estimates. The largest reported reduction was 39-fold on a combined measure of Toffoli gates and logical qubits, not a measured speedup on quantum hardware.

• Microsoft Trains AI Tutors With Simulated Students

  Microsoft Research introduced StudentSim, a system that builds personalized student simulators from de-identified learner records. In tests across chess, English writing and mathematics, the simulators outperformed comparison models on measures of student-like behavior and response to tutoring.

• Report: Huawei Clients Turn to Smuggled HBM at $4,000 Each

  Smugglers are reportedly moving South Korean high-bandwidth memory into China by packaging it with low-cost server processors, then recovering the memory after shipment. The chips are sold through several channels, including to Huawei customers seeking memory for the company’s AI processors.

• DOJ Announces Five Defendants in Nationwide Ghost-Student Fraud Cases

  The Justice Department announced charges against five defendants in three cases tied to a nationwide student-aid fraud investigation, and says the alleged scheme’s scale exceeds $200 million. The release describes stolen identities, diverted aid refunds and a cryptocurrency seizure of more than $750,000.

• Holyvolt Raises $45 Million to Expand Its AI Battery Lab

  Holyvolt will use $45 million to expand its AI models, automated laboratory capacity and scientific team as it advances its nickel- and cobalt-free DRX cathode. The company plans to begin commercializing DRX in 2027, starting with defense applications before expanding into mobility markets.

• JetBrains Ships PhpStorm 2026.3 With Air, a Workspace for Coding Agents

  JetBrains’ PhpStorm 2026.3 adds Air, a workspace for coding agents and multi-agent setups. The release also adds PHP 8.6 support, Git worktree features and remote development features.

• Sam Altman Pledges OpenAI Access for Independent Evaluators

  OpenAI CEO Sam Altman said the company will adopt employee-like access for independent evaluators, echoing a proposal from Anthropic CEO Dario Amodei.

• ChatGPT Hallucinations Put Customers at Odds With Service Workers

  The Verge reports that customers are bringing AI-generated errors into restaurants, parks and childcare conversations. Workers say some customers trust ChatGPT or Claude over staff who know the menu, terrain or child in their care.

• Microsoft Recasts Enterprise AI Agents as Insider Risks

  Enterprise AI agents can act like insiders, taking consequential actions with access to data and internal systems, Microsoft says. The company proposes using Project Perception to discover agents, manage their permissions and supervise their actions.

• Study Finds AI Judges Favor AI-Assisted Research Proposals

  A study of 70 student research proposals found no significant human-rated quality advantage for teams working with AI. GPT-4.1 scored AI-assisted proposals more favorably than human reviewers did, while proposal revisions improved ratings across both groups.

• ONESTRUCTION Ranks 20 AI Models on Construction Tasks

  ONESTRUCTION published a construction-focused AI leaderboard comparing 20 models across five work categories. Its October 2026 results show near-ties at the top overall, alongside substantial differences between categories.

• NVIDIA open-sources two Kubernetes projects for GPU infrastructure

  Two projects for building and operating managed Kubernetes services on GPU infrastructure are available in public preview. The projects separate cluster management from reproducible builds of the software artifacts those clusters use.

• Reco Raises $55M as AI-Agent Security Vendors Multiply

  Reco raised $55 million from AT&T Ventures, Forestay and Quadrille Capital, bringing its total funding to $140 million. The company is pitching tools to track AI agents’ access across enterprise systems as at least two dozen vendors target the security market.

• 26 Attorneys General Ask Congress to Regulate Frontier AI

  New York Attorney General Letitia James and 25 other attorneys general are urging Congress to establish federal safety rules for frontier AI. Their September 23 letter also calls for protecting states’ authority to enforce their own AI laws.

• Anthropic Opens Claude Plugin Submissions Through a Developer Portal

  Anthropic opened a portal for developers on paid Claude plans to submit plugins to the Claude directory. The portal adds automated checks, review tracking and usage analytics for approved listings.

• Hong Kong Plans Agent Registry and Kill Switches for Banking AI

  Hong Kong’s monetary authority has outlined proposed controls for banking AI agents, including clearer accountability, an inventory of material agents and human intervention mechanisms. The ideas are preliminary; the HKMA says a forthcoming discussion paper will set out options for possible supervisory requirements.

• Top AI executives meet ministers as leaked copyright reforms spark backlash

  Senior executives from Anthropic and OpenAI met Australian federal ministers as leaked copyright proposals triggered criticism from creators and politicians.

• AI-Powered Malware Highly Likely to Feature in UK Infrastructure Ransomware Attack by 2028, NCSC Says

  It is highly likely that AI-powered malware will have been used in a ransomware attack on critical national infrastructure between now and 2028, the UK’s National Cyber Security Centre says. Its assessment describes plausible scenarios, not predictions, and warns that AI could help attackers find and exploit weaknesses faster.

• China and the U.S. Hold Their First AI Dialogue Through Trade Talks

  China confirmed that Vice-Premier He Lifeng and U.S. Treasury Secretary Scott Bessent held the countries’ first AI dialogue during trade consultations. Washington has proposed a notification mechanism for AI incidents, but Beijing’s announcement did not say whether the proposal was discussed or accepted.