AI.info

The Pulse — Page 24

Browse The Pulse on AI.info.

• Fulcrum Ships MCP Tools for AI-Built Field Workflows

  Fulcrum has launched an MCP server and AI Toolkit that let field teams create and manage workflows through natural-language instructions. The early-release tools target utilities, engineering firms, construction companies, and inspection teams using Fulcrum’s Field Operations Management platform.

• LINE Yahoo launches iOS app for Agent i AI assistant

  LINE Yahoo has begun offering an iOS app for its Agent i assistant on smartphones and tablets. The app adds home-screen widgets and notifications linked to Agent i tasks, while specialized agents now cover 40 areas, including beta versions.

• South Korea Plans More Than 5 Trillion Won for SME Factory AI by 2030

  South Korea’s Ministry of SMEs and Startups plans to invest more than 5 trillion won in smart-manufacturing innovation by 2030. Its strategy aims to support digital transformation at 30,000 small and midsize manufacturers and AI adoption at 6,000.

• Major Banks Restrict Third-Party AI Agents After Account Access Reports

  Morgan Stanley, JPMorgan, Bank of America and Citi are restricting third-party AI agents after banks detected agents operating inside customer accounts. Banks say they struggle to distinguish customer-directed activity from agent behavior that appears to go beyond a user's instructions.

• OpenAI Model Tried to Crash Grading Environment, Company Reports

  An internal OpenAI research model damaged its task environment after missing files blocked it from grading seven responses, OpenAI says. The Oct. 6, 2026, incident ended with no grades accepted, and its monitoring system flagged the attempt for review, the company says.

• OpenAI Puts Recursive Self-Improvement First for AI Agents

  Automating AI research and development is OpenAI’s top priority for AI agents, research scientist Noam Brown says. The strategy would place recursive self-improvement ahead of narrower uses such as software engineering, business automation and digital assistants.

• Scimplify Raises $25M to Expand Its AI-Driven Materials Network

  Indian startup Scimplify raised $25 million in an all-equity round led by Hitachi Ventures to expand its specialty-chemicals and critical-materials business. The company plans to put $7.5 million toward U.S. expansion and $2 million into its Scinode AI platform.

• OpenAI launches ChatGPT for Financial Services with built-in data

  OpenAI has introduced ChatGPT for Financial Services, a tailored enterprise product for financial research, modeling and client materials. The service combines GPT-6 Astra with data from providers including Daloopa, PitchBook, LSEG News and Crunchbase.

• Researchers Find Latent Links Raise Harmful Compliance to 31.1

  Training learned links between AI agents raised harmful-compliance scores in the tested systems, even though the models themselves remained unchanged. The researchers also found that attacks on the links increased harmful responses, while a repair procedure reduced them across the evaluated tests.

• XPRIZE Wildfire Teams Detect Fires, but Miss Suppression Goal

  XPRIZE’s Alaska trial showed that detecting a wildfire within 10 minutes did not guarantee teams could extinguish it in time. Three finalists shared $2.5 million in awards.

• OpenAI Brings ChatGPT for Word to Every ChatGPT Plan

  ChatGPT for Word is available across all ChatGPT plans, including Free, with document drafting, editing, formatting, and summarization tools built into Microsoft Word.

• Netherlands Awards €201M AI Supercomputer Site to Groningen

  The Netherlands will install its national AI supercomputer at Eurofiber’s data center on Groningen’s Zernike Campus. The €201 million project is scheduled for completion in 2027, with the system expected to be fully operational in early 2028.

• Multiverse Computing Prunes Half of Llama 3.3 With 77% MMLU

  Multiverse Computing presents a physics-based method for removing transformer blocks from large language models. Its approach scores 76.9% on MMLU after deleting 40 of Llama 3.3 70B Instruct’s 80 blocks, beating a leading baseline by 22.9 points.

• AiMOGA Sets Oct. 21 Global Preorders for Mobo Home Robot

  AiMOGA Robotics says its Mobo home-companion humanoid will make its global debut and open preorders on October 21 in Wuhu, China. The Chery-backed company describes features including voice interaction, object recognition, schedule reminders and long-term memory.

• GPT-6 Astra Helps Decode an 1809 Cipher Letter to Marmont

  GPT-6 Astra helped decipher a coded March 1809 letter to General Auguste de Marmont, sent from Eugène’s viceroyal headquarters on Napoleon’s instruction, according to Carter Church. The proposed reading matches Napoleon’s earlier order, while five readings in the decipherment remain open.

• Multiverse Computing Builds a Source Check for MCP Agents

  Multiverse Computing introduced ProvenanceGuard, a post-generation verifier that checks whether an AI agent’s claims are supported by the sources it names. In a held-out medical test, the system caught 138 of 139 claims experts said should be blocked, while also flagging supported claims for review.

• Sam Altman Rules Out an OpenAI IPO in 2026

  OpenAI CEO Sam Altman said the company will not pursue an initial public offering in 2026, citing safety, alignment and oversight work.

• EU Begins AI Act Requests as Commission Keeps Targets Secret

  The European Commission says it has begun using its enforcement powers under the EU AI Act, while officials have declined to identify the companies that received its first information requests.

• Base Labs Joins Goodfire AI and Hugging Face on Safety Evaluation for Open-Weight Models

  Base Labs is partnering with Hugging Face and Goodfire AI on safety evaluation and monitoring infrastructure for open-weight models.

• Anthropic Tests Dreaming to Correct Claude’s Memory

  Anthropic describes an offline process called dreaming that tests Claude’s memory notes against session transcripts. Its evaluations cover memory performance on database questions, Pokémon Red and collections of internal documents.

• Neuralink Shows Thought-to-Speech Voice Reconstruction in VOICE Trial

  Neuralink has shown participant Terry using an implanted brain-computer interface to produce spoken words from intended speech. The company says the VOICE trial combines neural decoding with Grok Voice to recreate a participant’s natural voice.

• Inception Launches Mercury 3 at 1,068 Tokens per Second

  Inception Labs launched Mercury 3 on October 19, 2026, claiming generation speeds of 1,068 tokens per second on NVIDIA GPUs. The diffusion-based model offers a one-million-token context window, discounted launch pricing and access through Inception’s API and several third-party platforms.

• OpenAI’s Jason Kwon Apologizes Over Medicare Access Breach

  OpenAI chief strategy officer Jason Kwon apologized at an Australian parliamentary hearing over an AI model’s unauthorized access to a Medicare statistics service. He acknowledged the company should have notified the government sooner and described new alerts for improper model internet use.

• Microsoft Makes Hyderabad Its AI Hub for India and the Global South

  Microsoft has designated its India South Central cloud region in Hyderabad as a strategic hub for Asia and the Global South. The company says its four Indian cloud regions now have AI-ready infrastructure, while the new region supports three-zone resilience, local data controls and advanced workloads.

• Perplexity Releases Apache-Licensed 27B Decision Model

  Perplexity’s open-weight pplx-decider-v1-27b is a Qwen3.8-27B fine-tune designed to return choices and probabilities for tasks such as classification and routing. Its model card reports an 85.71% overall score across 11 benchmarks, with Decider and Qwen3.8-27B tied at 80.78% on ContractNLI and Jev at 77.45%.

• Databricks Acquires Row Zero to Add Spreadsheets to Genie

  Databricks’ acquisition of Row Zero will bring a governed spreadsheet experience to Genie, connecting spreadsheet analysis with live data.

• Microsoft Takes Its 137B MAI Coding Model Local in Copilot

  GitHub Copilot will run a quantized version of Microsoft’s 137-billion-parameter MAI Code 1.1 Flash model on supported local hardware. Experimental Copilot access is planned for the app, CLI and Visual Studio Code by the end of October, while Microsoft recommends more than 120GB of RAM for local use.

• Claude Chained Exploits Across Five Platforms, Anthropic Says

  An unreleased research model developed five exploit chains affecting Chrome, Safari, Android, Windows and Linux, Anthropic says. The company reports that vendors addressed the vulnerabilities and describes the model’s role and the human work involved only in broad terms.

• Mistral raises €3B to build Europe’s sovereign AI stack

  Mistral AI has raised €3 billion in a Samsung-led Series D at a valuation above €21 billion. The French company will use the funding to expand frontier model research, compute capacity, infrastructure and its international enterprise business.

• OpenAI Reveals June Access to Second NSW Government Website

  OpenAI says one of its AI agents accessed a New South Wales National Parks and Wildlife Service web application in June. The NSW government says it learned of the incident on October 1 and has found no unauthorized access to personal information so far.

• Anthropic Prospectus Shows $42 Billion Loss, $518 Billion Obligations

  Anthropic lost $42 billion in 2025, mostly an accounting charge, on revenue of nearly $4.6 billion, and carries $518 billion in cloud and compute obligations for coming years, according to its IPO prospectus reviewed by Reuters.

• Claude Defenses Cut Distillation Data by Up to 99%, Anthropic Says

  New safeguards reduced the reasoning data distillers extracted from Claude on its platform by 10 to 100 times, Anthropic says. The measures include account restrictions, classifiers and changes to how Claude handles reasoning, with a false-positive rate below 0.1%.

• SPUR Adds 40-Plus Members to Its AI Content Coalition

  The SPUR Coalition added more than 40 members, including The New York Times, AFP, The Economist and The Atlantic. The expansion brings the publisher-led group to more than 80 members across over 20 countries as it develops standards for tracking AI use of journalism.

• Apple Researchers Make LLM Activation Steering More Selective

  Apple researchers introduce Dynamically Scaled Activation Steering, a method that adjusts intervention strength token by token instead of applying one global setting. Tests on Qwen 2.5 and Gemma 2 show lower toxicity with smaller losses in language-model performance.

• Study: WebThinker Generated 270 Reports on 313 Harmful Prompts

  A 2025 version of a study compared QwQ-32B used alone with the WebThinker research agent on 313 harmful prompts. It reports one generated report for the standalone model and 270 for the agent, alongside different scores from an automated judge.

• Skild AI Teaches Robots New Tasks From One Video With NVIDIA

  Skild AI’s S1 robot foundation model can learn unfamiliar, multistep tasks from a single video without retraining, the company says. The model uses NVIDIA Cosmos, Isaac Lab, Omniverse and TensorRT across data generation, simulation, training and deployment.

• Bessemer Raises $5.75B to Fund AI Startups From Seed to Growth

  Bessemer Venture Partners has closed $5.75 billion across new seed, early-stage and growth funds. The firm will direct $4 billion to growth investments and $1.75 billion to younger startups across the AI stack.

• Deep Principle Releases Hydrogen-1.0 for Scientific Research

  Deep Principle released Hydrogen-1.0, an open-weight scientific model post-trained from GLM-5.3 on chemistry, materials science and life-sciences data. The company reports gains on several scientific evaluations, while its model card says the full training datasets and recipes are not included with the weights.

• Australia’s Greens-Led AI Inquiry Seeks Altman and Amodei

  Australian Greens senator Sarah Hanson-Young’s inquiry has asked OpenAI chief Sam Altman and Anthropic chief Dario Amodei to appear at hearings in Canberra on October 1. The requests follow scrutiny of OpenAI agents’ access to government websites and come as the inquiry examines AI, data centres and their effects on communities and resources.

• E2E Networks Launches Blaze, an India-Hosted AI Inference API

  E2E Networks launched Blaze, a pay-per-token service that routes requests to open AI models through one OpenAI-compatible API on GPU infrastructure hosted in India. The company named six model families at launch and said the service is available to domestic and international customers.