AI.info

The Pulse — Page 31

Browse The Pulse on AI.info.

• Rune Raises $40M to Put AI Compute at Solar Farms

  Rune has raised $40 million in Series A funding to launch RELIC, modular AI compute systems installed directly at solar-generation sites. The Mountain View startup says its Texas deployment can use underused solar power without new grid connections or construction.

• Google Releases 740M-Parameter EmbeddingGemma 2 for Edge AI

  Google launched EmbeddingGemma 2, an open-weight model that maps text, images, video and audio into a shared embedding space for on-device search and retrieval. Its modular encoders let developers choose a smaller footprint, while Google says Android ML Kit support is coming in the coming weeks.

• A New Test Spots Coded Messages Between AI Agents

  A paper proposes PoPLi, a behavioral test for detecting when an AI agent responds to a code hidden in another agent’s message wording. In a simulated market, the test separated an instructed letter code from several conditions without one, but the experiment used one model and a scripted sender.

• Microsoft Uses Bias Correction to Improve 2-to-6-Week Forecasts

  Microsoft Research’s Probabilistic Bias Correction system improves extended-range forecasts of temperature, precipitation and pressure, the company says. The approach adjusts existing weather models’ probability estimates and is being tested with Kenya’s meteorological service.

• NVIDIA Opens Vera Rubin Designs Through OCP

  At the October 12, 2026, OCP Global Summit, NVIDIA contributed Vera Rubin rack and compute-tray designs to an open industry architecture. The company also announced open-source data-center software and networking work with Meta, Cisco and Google.

• Amazon Ends NDAs With Government Agencies on Data-Center Projects as Trust Fight Continues

  Amazon no longer uses nondisclosure agreements with government agencies it works with on data-center projects, alongside commitments to hold open houses and publish annual operating figures. A report coauthored by Erin Brockovich and Suzanne Boothby describes residents’ concerns about transparency and learning about projects after decisions have been made.

• Enterprise AI Agents Violated Rules in 40% of Answered Runs

  A benchmark of 19 AI models found violations in 918 of 2,304 answered enterprise-agent runs. The tests used simulated business systems and blocked every attempted change before it could take effect.

• Block Trained Dessa on Human Speech and One GB200 NVL72

  Block’s Dessa learned to hold real-time conversations from human speech alone, using one NVIDIA GB200 NVL72. The speech-to-speech model is being offered to a small group of AI safety researchers, while Block’s examples include both expressive exchanges and occasional mistakes.

• OpenAI Notifies Dozens of Third Parties Over Agent Activity

  OpenAI has notified dozens of third parties after reviewing agent activity during training and evaluation. Australia’s prime minister criticized the company’s delay in notifying his government about a separate incident involving a government health data portal.

• SAP Agrees to Acquire TechWolf for Workforce Intelligence

  SAP has agreed to acquire Belgian AI company TechWolf, whose software maps employees’ tasks and skills using data from connected business and HR systems. The companies expect the deal to close in the fourth quarter of 2026, pending regulatory approval, but have not disclosed its terms.

• Factory triples valuation to $5B with a $200M raise

  Factory has raised $200 million at a $5 billion valuation, bringing its total funding to more than $400 million as it expands its enterprise software platform.

• Runware targets 30 MW a day from its Romanian data-center factory

  Runware plans to scale production of its factory-built AI data centers to 30 megawatts of capacity per day by 2028, with installation and setup targeted to take as little as 48 hours. Runware already has two factories, and its expansion plans will test whether it can manufacture and deliver pods at the promised pace.

• Google Engineer Resigns Over Chip Tools That Could Speed AI

  Robert O’Callahan resigned from Google on September 24, 2026, because his team’s chip-design tools could make AI cheaper and faster, he says. He plans to keep maintaining Pernosco and rr while studying how AI agents debug code.

• Researchers Used OpenAI’s Codex to Open a Monorepo Pull Request

  Hacktron researchers say a vulnerable image-processing path and an OpenAI SSO flaw let them reach an employee’s Codex account and create a pull request in OpenAI’s internal monorepo.

• JumpCloud Finds 92% of Firms Report AI-Related Incidents

  A JumpCloud survey of 250 U.S. and U.K. IT decision-makers found that 92% of organizations reported an AI-related incident or exposure in the past year. The report also found that 59% of organizations using AI agents had not fully integrated them into the same identity policies used for employees.

• Nominal raises $240M in Series C at $3B valuation

  Nominal announced a $240 million Series C at a $3 billion valuation, led by Lightspeed Venture Partners and co-led by Insight Partners. Its platform connects hardware data with engineering context and is used by more than 80 commercial and government organizations.

• MiniMax Confirms Export Registration; CCTV-Affiliated Account Reports DeepSeek Claim

  MiniMax completed technology-export registration for certain models, it said in an October 23 exchange filing, while Yuyuantantian reported that DeepSeek had also completed registration procedures. Chinese ministries have published a draft technology-export catalog and temporary registration rules that apply pending the catalog’s formal revision.

• A Checklist Cut Coding Agents’ False Reversals to 0.02%

  A study of coding agents found that recording claims and evidence before user criticism sharply reduced reversals of requirements that had already passed. The method’s results varied across tasks, and the paper says its strongest evidence came from patch review.

• OpenAI Engineer Arrested in Hidden Recorder Case

  OpenAI engineer Abdulrahman Alfozan planted an encrypted audio recorder in an executive conference room, federal prosecutors say. The FBI arrested him on October 9, and OpenAI’s internal AI models helped identify the device’s alleged source, the company says.

• Anthropic Resolves Errors Across Five Claude Models

  Anthropic reported elevated errors affecting five Claude models and four services on October 25, with impact recorded from 06:05 to 06:35 UTC. A separate incident later that day affected Claude Fable 5.1, according to a second status notice.

• MLPerf Adds RAG and Edge-Agent Tests to v6.1 Results

  MLCommons’ MLPerf Inference v6.1 adds benchmarks for retrieval-augmented generation pipelines and edge-based agentic inference.

• Finite State raises $27 million in Series C funding

  Finite State raised a $27 million Series C round led by funds managed by The O.H.I.O. Fund. The company plans to use the capital to expand internationally and develop autonomous features for its product-security platform.

• TSMC Profit Rises 68.7% to Record as AI Demand Outstrips Supply

  TSMC reported third-quarter net income of NT$763.06 billion, up 68.7% from a year earlier, as revenue climbed 50.9%. CEO C.C. Wei said demand exceeds the chipmaker’s ability to supply, while TSMC forecast fourth-quarter revenue of up to $54.3 billion.

• Expanse Raises $5.3M to Predict AI Workload Needs

  Expanse has raised a $5.3 million seed round led by Crane Venture Partners for software that predicts GPU, CPU, memory and runtime requirements before AI workloads run. The company will use the funding to expand engineering, develop its platform and reach customers in AI infrastructure, finance, life sciences and research.

• NVIDIA details Workload Map for Kubernetes AI workloads

  NVIDIA’s Nov. 2 technical post explains how its open-source Workload Map lets Kubernetes tools interpret different AI workload types through reusable definitions. The project ships 16 workload definitions and examples spanning inspection, scheduling and management.

• Boston Dynamics Opens Georgia Center to Train Atlas for Hyundai Factories

  Boston Dynamics has opened the Robotics Metaplant Application Center inside Hyundai Motor Group’s Georgia manufacturing campus. The facility is training Atlas humanoid robots for parts logistics and sequencing ahead of wider Hyundai and Kia deployments.

• BIS: AI-to-AI Deals Accounted for 55.2% of Disclosed Incoming Investment Value

  A Bank for International Settlements bulletin says 55.2% of incoming investment in AI firms from 2021 through 2025 came from other AI firms. It examines how investment ties overlap with commercial relationships and the risks those connections may pose.

• AWS Shows Developers How to Preview Apps in AgentCore Sessions

  AWS’s walkthrough shows how developers can connect a live Vite app in an AgentCore session to a local preview page while a coding agent edits its files. The sample uses authenticated session endpoints, rather than adding a packaged live-preview feature to AgentCore.

• Intel Xeon 6990E+ Handles 27% More Sandbox Requests

  Intel’s Xeon 6990E+ system sustained about 313 concurrent CubeSandbox creation requests before average latency exceeded 200 milliseconds, compared with about 246 on the tested AMD EPYC 9965 system, Intel says. The comparison covers sandbox provisioning on three specific server configurations, not the time or cost of completing an agent’s full task.

• Red Hat AI Gateway Flaw Can Inject Another Tenant’s API Keys

  An authorization flaw in Red Hat OpenShift AI’s model-serving gateway can let an authenticated user trigger requests with another tenant’s provider credentials. Red Hat rates the vulnerability Important and gives it a preliminary CVSS 3.1 score of 9.1.

• NVIDIA Agrees to Buy Hugging Face for $12.93 Billion

  NVIDIA has agreed to acquire Hugging Face in a $12.93 billion transaction announced September 3, 2026. The deal gives NVIDIA control of a platform used by more than 18 million developers while promising to preserve support for open models, multiple clouds and rival hardware.

• Safari 27 Gives AI Coding Agents Direct Browser Control

  Safari 27 introduces a local Model Context Protocol server that lets Claude Code, Codex and other agents inspect and test webpages inside Safari. The server exposes DOM data, network requests, screenshots and console output without sending its own network traffic.

• Simular’s Agent S4 Evolves Interfaces for Computer-Use Agents

  Simular introduced Agent S4 on October 21, 2026, a system that evolves the interfaces computer-use agents rely on. On OSWorld V2, the company says Qwen3.8-27B with S4’s evolved interfaces scored 21.3% on the benchmark’s 108 tasks.

• Anthropic Makes Claude for Government Generally Available

  Anthropic made Claude for Government generally available to federal and state agencies on September 30. The FedRAMP High-authorized service offers desktop Claude, capped usage billing and administrative controls, while Claude Code CLI and Claude for Microsoft 365 enter early access.

• Study Finds Nine AI Models Use More Language Features Linked to Processing Difficulty

  A study found that nine frontier and near-frontier language models used several language features associated with human processing difficulty more often than human-written comparisons. A separate experiment found that length penalties shifted a small model toward shorter grammatical forms, but the study did not directly test how readers understood the model outputs.

• Watchtower Labs Raises $35 Million for Military Logistics AI

  Seattle-based Watchtower Labs emerged from stealth on October 21, 2026, with $35 million in seed financing led by Shield Capital and Conviction. The company is developing AI and simulation software to help military planners test logistics options, and has a prototype contract under the Defense Innovation Unit’s Joint Sustainment Decision Tool program.

• Altman Meets 50 Samsung Presidents as OpenAI Seeks AI Memory

  Sam Altman met about 50 Samsung affiliate presidents in Yongin, South Korea, on October 31, 2026. The meeting followed an existing Samsung-OpenAI agreement to pursue memory-chip supply for OpenAI’s Stargate project, but Samsung announced no new supply deal.

• Anthropic Reportedly Seeks 50.1% Voting Power for Seven Founders

  Anthropic is asking shareholders to approve a proposed share structure that would give its seven co-founders 50.1% of voting power on most corporate matters, according to The Information. The proposal would exclude board elections and preserve the Long-Term Benefit Trust’s power to appoint a board majority.

• NVIDIA Details Kumo Forecast’s Retrieval-Based Time-Series Model

  NVIDIA’s October 21 technical post explains how Kumo Forecast combines a pretrained time-series model with retrieval of similar historical patterns. The Apache 2.0 model supports numeric time-series inputs and includes tools for explaining forecasts, but NVIDIA recommends testing it on each intended use case.

• Meta Adds Training-Time Safeguards to Its AI Release Framework

  Meta updated its Superintelligence Scaling Framework on October 2, adding safeguards for high-risk training runs and refining how it assesses open-weight releases. The company also plans a board committee to review the framework, with that governance change due in the coming months.