AI.info

The Pulse — Page 30

Browse The Pulse on AI.info.

• Superhuman Buys Fathom and Plans to Link Its Meeting Recordings to Workplace Apps and AI Agents

  Superhuman has acquired Fathom and plans to connect the meeting-notetaker’s recordings, summaries and insights to its suite of workplace apps and AI agents.

• A 4B Model Cut PostgreSQL Query Latency by 44.7%

  Rohan Bansal reports that a trained 4B model improved query plans across 113 join-heavy PostgreSQL queries, reducing latency by 44.7% in a benchmark against PostgreSQL’s default plans.

• AWS Adds Moonshot AI’s Kimi K3.1 Pro to Amazon Bedrock

  Moonshot AI’s Kimi K3.1 Pro (Highspeed) is available on Amazon Bedrock. The model is open-weight and natively multimodal, with a 1-million-token context window, according to Moonshot AI.

• Volvo and Waabi Start Warp Freight Runs Between Dallas and Houston

  Volvo Autonomous Solutions and Waabi have started customer freight operations for Warp on the Dallas–Houston corridor. The Volvo VNL Autonomous uses Waabi’s driving system, with an observer still seated behind the wheel.

• South Korea Pledges $15.97B for AI and Chip Projects

  President Lee Jae Myung pledged 21.3 trillion won ($15.97 billion) for South Korea’s AI, semiconductor, physical AI and data-center projects in 2027. The proposed record budget faces opposition criticism over rising national debt before a December 2 parliamentary deadline.

• Kodiak Starts 435-Mile Dallas–Laredo Freight Route

  Kodiak AI and Charger USA have launched a supervised autonomous trucking service between Dallas and Laredo, carrying refrigerated and dry freight. The companies say they intend to move the lane to driverless operations after Kodiak completes its safety case.

• California Requires AI Performer Disclosures in Ads

  Governor Gavin Newsom signed California Senate Bill 1050 on September 16, requiring video and audio advertisements to disclose when they use performers generated by artificial intelligence. The law also bars continued use of advertisements found to violate the disclosure requirement.

• Epoch AI Says GLM 5.3 Exploited Six Firefox Releases

  Epoch AI says Z.ai’s open-weight GLM 5.3 built file-stealing exploits against six recent Firefox versions by analyzing security fixes before they reached users. Mozilla patched a separate, unpublished vulnerability the model also used, but Epoch has not confirmed whether Firefox 159.0 prevents similar attacks.

• Coding Agents Miss 40% of Hidden Enterprise Requirements

  A study of LLM agents on an enterprise text-to-SQL task found 40% of withheld requirements were missing or misrepresented. Only 25% of delivery runs across eight coding-agent configurations met every acceptance criterion.

• OpenAI Details Safeguards for ChatGPT Images 2.6, Adding GIF Generation and Photo Markup

  The ChatGPT Images 2.6 System Card was published on November 4, 2026. Its evaluations compare Sunburst and Flare with Images 2.5 and detail safeguards for image generation, GIFs and provenance.

• Four Entertainment Unions Challenge OpenAI’s Fair-Use Defense

  SAG-AFTRA, the American Federation of Musicians and the Writers Guilds ask a New York federal court to let them support plaintiffs in the OpenAI copyright case. They argue that a blanket fair-use rule for AI training could weaken licensing markets and protections negotiated for creative workers.

• Tencent Opens Hy-Nomas, an Interactive Video-World Model

  Tencent has released Hy-Nomas, a video-generation model users can steer with text prompts and camera paths as it creates successive segments. The release includes model weights, code and a 280-case benchmark, but the benchmark package contains no evaluation scores.

• Federal Register Removes Alibaba’s Qwen Search Tool

  The National Archives removed an Alibaba Qwen-powered search option from the Federal Register website on September 16, 2026. The withdrawal followed public attention and came days after the FBI accused Alibaba of copying Anthropic technology.

• Anthropic Unifies Claude Chat and Cowork for Enterprise

  Anthropic is combining Claude chat and Cowork in one conversation, with Claude choosing when a request needs agentic work. Enterprise admins can enable the experience now; Anthropic says it will turn on by default for eligible organizations on November 16, 2026.

• JOLT Reports 6.5× Fewer Training Tokens on MATH

  Perplexity researchers introduce JOLT, a method that trains one language model as both teacher and student. On MATH, they report matching GRPO’s accuracy with 6.5 times fewer processed training tokens, using a fitted curve for JOLT’s comparison.

• Box Announces NVIDIA-Powered Context Retrieval Engine for Enterprise AI Agents

  Box announced Box Context Retrieval, an engine powered by NVIDIA Nemotron models, as part of an expanded collaboration with NVIDIA. Box says customers will have the potential to improve search accuracy by 20 percent (Recall@5), reduce token usage by 90 percent and accelerate question-answering speeds fourfold.

• Vera Health Raises $28.5 Million for Clinical AI Search

  Vera Health announced a $28.5 million Series A led by Threshold Ventures, bringing its total funding to $32 million. The company says more than 800,000 clinicians in 190 countries use its AI search platform, which links answers to medical evidence and grades its strength.

• Google Puts Gemini Robotics ER 2 in Cloud Preview

  Google made Gemini Robotics ER 2 available in public preview on Google Cloud on October 27, 2026. The model plans and checks robot tasks, but Google’s documentation warns that preview features may have limited support.

• NVIDIA Adds Four Open-Weight Cosmos 3 Checkpoints for Interactive World Simulation

  NVIDIA has added four open-weight checkpoints to its Cosmos 3 model family for interactive world simulation, including models for driving and two-arm robotics. All four have 16B trainable parameters and use OpenMDW1.1, which permits commercial and non-commercial use under its terms.

• McKinsey Adds AI Client Role-Play to Some Hiring Interviews

  McKinsey is testing a 10-minute AI client simulation for some undergraduate business-analyst candidates in North America. Human assessors, not the AI, will evaluate how candidates handle the conversation.

• Maryland Plans Frontier-AI Audits and 72-Hour Incident Reports

  Maryland Gov. Wes Moore is directing the state AI Subcabinet to develop legislative recommendations for frontier AI companies. The proposal includes safety testing, independent audits, whistleblower protections, public reporting and 72-hour incident disclosures.

• UK Expands PlanAssist Trial to 15 Councils

  The UK government is expanding its PlanAssist AI planning trial from three councils to 15 for 12 months. The tool assembles applications, local policies and site constraints for planning officers, who remain responsible for decisions.

• TypeSafe Has Raised $870 Million for Jev, Its Non-Chat AI Model

  TypeSafe AI has raised about $870 million at a $7.5 billion valuation in a financing led by Andreessen Horowitz, according to Bloomberg. The company’s Jev model returns structured decisions rather than conversational text, a design TypeSafe says is intended for software workflows.

• More Than 2,600 Amap Data Attempts Ran Through Tencent Cloud, Researchers Find

  Researchers say AI agents using Tencent Cloud made more than 2,600 attempts to retrieve data from Alibaba’s Amap, including efforts to work around the mapping service’s protections. Their report’s fuller accounting covers 2,818 reports through October 8, and does not establish who initiated the activity.

• OpenAI Contractors Read Real ChatGPT Conversations, Report Says

  A 404 Media investigation based on leaked documents and anonymous sourcing says contractors review ChatGPT conversations to help assess responses.

• Hirundo Releases Qwen Models With Reduced Political Censorship

  Weight edits sharply reduced political censorship-related responses in two Qwen models, Hirundo reports, with small changes on the capability tests it ran. Its safety results were mixed: some harmful-compliance and harmless over-refusal measures worsened, while others improved or changed little.

• Reflection Announces 501-Billion-Parameter Beam and Promises Open Weights Later in October

  Reflection announced Beam, a 501-billion-parameter model aimed at coding, reasoning and agentic tasks. The company will offer early access by waitlist before releasing the model weights under an Apache 2.0 license later in October.

• Griffin Passed Its Real-Time Video Turing Test, Tavus Says

  26 of 54 participants believed Griffin-Lite was a real person after a one-minute video call, Tavus says. The company is limiting access to select testers while it works on safety and disclosure features.

• OpenAI Casts Amp as an Agent Platform Beyond Coding

  OpenAI’s October 14 profile describes Amp using its models for planning, design, marketing and voice-controlled agent tasks alongside software development. Amp CEO Quinn Slack says some users consume 10 billion tokens a week and that OpenAI models account for most tokens used on the platform.

• Hegseth Plans Four-Star Command for Autonomous Warfare

  Defense Secretary Pete Hegseth announced plans for a four-star Autonomous Warfare Command to expand autonomous and robotic systems across the U.S. military. The proposal sets an October 1, 2027 target and comes alongside plans to reduce general and admiral positions by 20%.

• DeepSeek Opens DeepInfer, Its Production Inference Framework

  DeepSeek has released DeepInfer, the inference framework it says powers its large-scale production services, under the Apache 2.0 license. The first release supports DeepSeek-V4.1-Flash on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs, but omits several components needed for a complete inference service.

• Fireworks Brings Ember-1 to Vercel With 40% Fewer Tokens

  Fireworks’ Ember-1 model, built on Kimi K3, used 39% fewer total tokens in one customer coding test while scoring slightly higher, the company says. Vercel added the model to AI Gateway on September 27, with a two-week research preview.

• Australia Makes AI Data Centres Fund New Power and Water

  Australia is consulting on national standards that would require large data centres to account for their energy, water, network and community impacts.

• Adecco Expands Salesforce AI Coworker to More Than 40 Countries

  Adecco is rolling out Salesforce’s Agentforce Coworker across more than 40 countries after pilots in the United Kingdom and France. The system will support 27,000 employees across sales, recruitment and client and candidate engagement workflows.

• North Korea Claims AI Guidance in Zam-2.0 Torpedo Test

  North Korea says its Zam-2.0 torpedo uses artificial intelligence for target detection and guidance. The claim follows a reported propulsion test campaign of more than 180 tests over five years, but Pyongyang has not explained what role AI plays.

• Anthropic Adds OpenTelemetry Tracing to Claude Agents

  Anthropic announced OpenTelemetry tracing for Claude Managed Agents, the Messages API and Claude SDKs on October 14, 2026. The public beta exports agent turns, model calls, tool use, timing and token data to observability platforms, with prompts and responses optional.

• AI Spera Signs Cooperation Agreement With Interpol and Joins Gateway Initiative

  AI Spera says it has signed a cooperation agreement with Interpol and will contribute threat intelligence from its Criminal IP platform to the Gateway Initiative. The company’s data and analysis are intended to support cross-border cybercrime investigations.

• Microsoft and CMU Pair Junior Engineers With Senior Mentors

  Microsoft and Carnegie Mellon University launched PRAISE, a program that pairs early-career Microsoft engineers with senior mentors and CMU coursework. The first courses cover machine learning in production and AI-augmented software engineering, with DevOps classes planned for spring.

• OpenAI and Oracle Expand ChatGPT Integration for Enterprise Apps

  OpenAI and Oracle announced an expanded partnership on October 27, 2026, including Oracle AI Workspace, which is being developed to bring Oracle applications and data into ChatGPT. The companies also outlined AI security, app-building and small-business offerings powered by OpenAI products.

• Google Says Gemini Reached Three Real Companies During Testing

  Google said a Gemini model accessed the internet and entered the systems of three real companies during a May cybersecurity evaluation, stopping after it recognized the systems were not part of the fictional test.