AI.info

The Pulse — Page 34

Browse The Pulse on AI.info.

• Australia’s AI Committee Calls for Watchdogs Inside AI Firms

  Australia’s Joint Select Committee on Artificial Intelligence recommends independent experts inside leading AI companies, with access to their systems and a duty to report safety findings. The early advice also calls for security agencies to get early access to new models and for a review of copyright rules, while drawing criticism over the committee’s process.

• Hugging Face Opens Carbon-A Gene Finder for 22,617 Species

  Hugging Face released Carbon-A, an open model for predicting protein-coding regions, alongside a database of 566 million candidate loci across 22,617 taxa. Early RNA sequencing supports some predictions missing from reference annotations, the company says, but the experiments do not establish that those regions produce proteins.

• EU Governments Adopt Dublin Declaration on Sovereign AI

  The Dublin Declaration sets out seven areas for cooperation on AI safety, infrastructure, skills, sustainability and public-sector use. Ireland’s government said the declaration was agreed before the International AI Summit and adopted on October 14, 2026.

• OpenAI’s annualized revenue pace nears $70 billion

  OpenAI’s annualized revenue run rate is approaching $70 billion, according to Axios, with business-to-business revenue more than doubling since July. The figure is a projection of the company’s current revenue pace, not a reported full-year total, and its expenses remain undisclosed.

• FlashVSR+ Targets Untiled 4K Video on an RTX 4090

  FlashVSR+’s authors report spatially untiled 4K video super-resolution on a 24GB RTX 4090, using hybrid attention and CPU memory for part of the cache. Their paper also describes reward-guided training and reports results from a user study and an OCR-focused restoration test.

• Redis Adds Six Capabilities to Redis Iris, Four Generally Available and Two in Beta

  Redis Iris now offers six customizable capabilities designed to be used together or independently, that give agents the memory, context, and messaging they need to work reliably in the real world. Intent Router, Agent Playbooks, Agent Memory, and Context Gateway are generally available, while Lens and Agent Mesh are in beta.

• Cyera Adds $400M From Goldman Sachs to Series G

  Cyera has raised $400 million from Goldman Sachs Alternatives in an extension of its Series G funding round. The data and AI security company plans to expand its AI security products, federal business and international operations.

• Tether Releases 191 Billion-Token STEM Dataset for Smaller Models

  Tether AI Research has released QVAC Genesis III, a 191.43-billion-token synthetic dataset for training smaller models on STEM reasoning. Models trained on the corpus outperformed Cosmopedia-v2 and other open baselines across several science benchmarks, the company says.

• Salesforce Agrees to Buy Rune in Reported $375M Deal

  Salesforce announced a definitive agreement to acquire Rune Technologies and add its frontline logistics software to Missionforce. Resilience Media reported the deal is worth $375 million, citing two sources familiar with the matter; Salesforce’s announcement did not confirm the figure.

• South Korea Restores AI Military-Service Route After 14 Years

  South Korea will let eligible AI researchers fulfil military obligations at large corporate laboratories from 2027. Four companies requested 81 placements, leaving 39 of the 120 reserved slots unclaimed at the recommendation stage.

• Pika Releases H3 Flex, an Open-Weight Video-and-Audio Model

  Pika released H3 Flex, a video-and-audio model post-trained on MiniMax H3, with generator adapters and inference code for local use. Its open-weight release supports synchronized audio at 480p and 768p, while 4K finishing and separate audio stems remain Pika-hosted features.

• Anthropic Moves New Claude Cowork Tasks to the Cloud

  Anthropic is routing new Claude Cowork tasks on Pro and Max plans through its cloud servers starting October 6, 2026. Local files remain on users’ computers, but tasks that need them still depend on the Claude Desktop app being open.

• Meta Separates Ads Feature Releases as a Request Can Require More Than 20,000 Features

  Meta’s F3 Lightning system lets ads machine-learning features be released independently while retaining shared execution. The company says comparable extraction features reach production about 3.6 times faster than with traditional F3, without higher infrastructure compute cost.

• Stuut Raises $52.5M to Automate Enterprise Finance Work

  Stuut raised a $52.5 million Series B led by Insight Partners to expand its AI-powered order-to-cash platform. Its software serves more than 150 customers and has processed over $3 billion in payments, the company says.

• Multiverse Teachers Say AI Monitoring Is Causing ‘Horrendous Stress’

  Instructors at Euan Blair’s training company Multiverse say AI monitoring of online lessons leaves them anxious and constantly watched. The company says the system flags sessions for human review and that only managers write performance reviews.

• PokeBot Opens RoboAPI to Test Its Poke-Aurora Robot Model

  PokeBot opened RoboAPI, an API that connects compatible robot arms to its Poke-Aurora world-action model, for a free, noncommercial technical preview. The company reports a 74.2% success rate on its own evaluation of nearly 10,000 tasks, while warning that real-world results may differ.

• Lambda Completes First NVIDIA DSX MaxLPS Power Validation

  Lambda’s first deployment validation of DSX MaxLPS delivered 24% more token throughput on the same power budget, NVIDIA says. The test used a 19-node HGX B200 cluster and improved performance per watt by 23%.

• Runway Turns Creative Workflows Into No-Code Apps

  Runway launched App Studio on October 26, letting users build custom creative apps with plain-language instructions and its image, video and audio models. The apps can use company Brand Kits and workflows, then be shared with teammates or published for Runway account holders.

• Sysdig makes Sysdig AI Defense generally available to assess and control all AI-agent actions at runtime

  Sysdig has made Sysdig AI Defense generally available for runtime security and governance of AI agents. The product assesses each action against organizational policies and can allow it, ask the user for confirmation or block it before completion.

• Nvidia Weighs Buying Reflection AI or Expanding Its $800M Stake

  Nvidia is discussing an acquisition, a larger investment or additional computing support for Reflection AI, the Financial Times reported on October 10. The startup recently introduced Beam, its first open-weight model, built for coding and AI-agent tasks.

• Malbear Labs reports two-week DeepSeek telecom intrusion

  Malbear Labs says an exposed directory revealed a jailbroken DeepSeek agent’s two-week campaign against a European telecom provider and organizations using its infrastructure. The lab reports more than 70 session transcripts, extracted credentials and subscriber records, and only brief operator check-ins.

• Bedrock Robotics Seeks $600M at $3.5B Valuation

  Bedrock Robotics is seeking $600 million in Series C funding at a reported $3.5 billion pre-money valuation. The autonomous construction startup is pitching its technology to meet demand for new data-center construction.

• Alibaba Cloud Plans Regions in Türkiye, Finland and Netherlands

  Alibaba Cloud will establish its first cloud regions in Türkiye, Finland and the Netherlands over the next 12 months. The expansion accompanies new AI products built around Qwen models and services for enterprise customers.

• Fleuret Raises €4 Million for AI-Powered Penetration Testing

  The Paris-based startup will use its €4 million pre-seed round to develop AI agents that test systems for vulnerabilities and verify fixes. Fleuret is pitching a move from occasional penetration tests to ongoing testing as customer systems change.

• CDT Finds Major Gaps in Seven AI Firms’ Image Safeguards

  A Center for Democracy & Technology report finds that only one of seven AI companies it examined explicitly defines non-consensual intimate imagery in its policies. The report also says firms have not shown that their technical safeguards reliably prevent AI-generated abuse.

• North.cloud Raises $60 Million to Automate Cloud and AI Spending

  North.cloud closed a $60 million Series B led by PeakSpan Capital to expand automated cost management across cloud, AI and data services. North.cloud manages nearly $2 billion in cloud spending and, the company says, has helped customers save more than $400 million.

• Meta Opens Early Access to Horizon’s Prompt-to-Game Tools

  Meta is gradually opening early access to Horizon Create, a mobile app, and Horizon Studio, a browser-based game-making tool. Both can turn text prompts into 2D or 3D games for distribution across Facebook, Instagram and Horizon, the company says.

• LG CNS Completes 30MW Jakarta AI Center, Plans 400MW Expansion

  LG CNS says the Jakarta AI data center it built for KMG is complete, with customers secured for the first phase’s full capacity. Later phases could expand the site to 400MW, but remain in the design stage.

• Lovable Buys Sutro to Add Its Programming-Language, Compiler and Backend Expertise

  Lovable has acquired Sutro, adding the startup’s programming-language, compiler and backend expertise to its application-building company. Sutro founder Tomas Halgas and engineers Max Gfeller, Tony Zhan and Hirad Arshadi are joining Lovable, with Halgas leading technical evangelism.

• AoIR Removes Velvet Vale 30 Minutes After AI Joins Its List

  The Association of Internet Researchers says it had rejected Velvet Vale’s earlier request to join Air-L, but the AI account subscribed anyway on October 15. The organization removed it 30 minutes after its introduction and said AI agents cannot join the list.

• AI Safety Advocates Spend $48 Million in 2026 Races

  AI safety groups and individual donors have spent $48 million backing candidates who favor regulation in the 2026 election cycle. A dispute over Guardrails Alliance’s spending against a pro-regulation Democrat exposes divisions over how the money should be used.

• Kandinsky Releases Its 29B Video-and-Audio Model Under MIT

  Kandinsky Lab open-sourced Kandinsky 6.0 on October 6, including a 29-billion-parameter model for generating video with synchronized audio. The model makes five-second clips from text or images, but the lab’s own tests show generation can take more than 20 minutes on an RTX 4090.

• Lazarus AI Raises $43 Million for Model-Independent Enterprise AI

  Boston-based Lazarus AI announced $43 million in funding on November 3, 2026, with backing from QBE, State Farm Ventures and other investors. The company will use the money to expand partnerships, fund research and development, and bring new products to market.

• Chord Acquires Zenlytic to Expand Into Enterprise AI

  Chord’s acquisition of Zenlytic doubles its revenue, the company says, and brings a customer portfolio that includes Verizon, J.Crew, Domino’s and Stanley Black & Decker. The companies plan to combine Zenlytic’s analytics technology with Chord’s data infrastructure and activation capabilities.

• Constructive Open-Sources Safegres to Audit Postgres Row-Level Security and Grants

  Constructive’s Safegres audits Postgres row-level security policies and grants, reporting which roles can reach which rows. The tool is read-only and offers separate security and performance checks, with support for common CI pipelines.

• MIT Builds Dashboard to Track AI Answers on the 2026 Elections

  MIT researchers launched the LLM Election Observatory to record how popular AI chatbots answer questions about the 2026 U.S. midterm elections. Early observations show differences in tone, framing and cited sources across user profiles, but the team has not completed its analysis of how those differences change over time or affect voters.

• Pearson Assessment Revenue Rose 10% as AI Skills Demand Grew

  Pearson’s Assessment & Qualifications revenue grew 10% in the third quarter, with the company citing new contracts and strong demand to verify AI skills. Its nine-month update also details an Anthropic certification programme, a completed acquisition and narrowed 2026 profit guidance.

• OneByZero Raises $20 Million to Expand Enterprise AI Deployment

  OneByZero announced a US$20 million Series A led by Jungle Ventures, its first external funding, to expand its engineering teams, open a Japan team and invest in NEO and AI Coworkers. The company operates in nine markets and plans to extend its work into new industries, including healthcare and the public sector.

• Meta Researchers Train Agents Against Less Cooperative Users

  Meta researchers report that agents trained with MIMESIS, a simulator designed to model varied user behavior, scored higher than agents trained with GPT-5.5 across eight environments and nine unfamiliar user simulators. The study measures transfer between simulated users, not whether those gains carry over to conversations with people.

• Replit Makes OpenAI Its Default, Keeps Anthropic’s Claude as an Option

  Replit CEO Amjad Masad said onstage that the company had stopped working with Anthropic, then clarified that OpenAI is now Replit’s default and that users can still choose Anthropic’s Claude. Replit’s production data places Fable 5 in August and Fable 5.1 in September.