AI.info

The Pulse — Page 5

Browse The Pulse on AI.info.

• TypeSafe’s ‘Jev Leak’ Was a List of 101 Probabilities

  A GitHub Gist linked in a post by TypeSafe founder Diogo Almeida contains values from 0.00 to 1.00 in hundredth-step increments, but no customer identifiers or individual output records. The post describes a leak discovered “this Friday evening,” although it is dated Saturday, October 24, 2026.

• Dell Puts Gemini at the Center of Its First XPS Googlebook

  Dell has introduced the 13-inch XPS Googlebook, its first XPS laptop designed around Google’s Gemini Intelligence platform. The $999 laptop combines a Snapdragon X Elite processor, Android phone integration and a 2.5K touch display.

• NewsGuard Says ChatGPT Images 2.5 Made Fake Candidate Images 80% of the Time

  NewsGuard says ChatGPT Images 2.5 produced compromising images of U.S. midterm candidates in 16 of 20 targeted prompts. OpenAI says its policies prohibit election interference and that generated images carry SynthID watermarks.

• Mistral Expands Studio With Managed Agents and Apps for Enterprise AI

  Mistral expanded Studio with managed agents, apps and new workflow tools for enterprise AI. Mistral says the platform can run in cloud, VPC and on-premises environments, with controls for data, tools and approvals.

• Ando Launches Agent-Native Messaging With $20 Million in Funding

  Ando emerged from stealth on September 24 with a team messaging platform for people and AI agents, alongside a $20 million funding announcement. The company says agents can join channels, keep shared context and take part in live conversations.

• ElevenLabs Ships Music v2.5 With API Access and Free Downloads

  ElevenLabs released Music v2.5 on September 11, making it the default model in ElevenMusic and adding access through ElevenCreative and the ElevenLabs API. Free users receive five lossless downloads per day, while Pro users receive 400 per month.

• Arena Raises $200 Million and Launches an AI Alignment Index

  Arena announced a $200 million Series B at a $3.1 billion valuation on October 8, 2026. The company also introduced an Alignment Index that scores AI agents on unauthorized actions, false attribution and deceptive claims of task completion.

• At Least Eight States Sue OpenAI Over ChatGPT Safety

  Arkansas Attorney General Tim Griffin sued OpenAI, alleging that it misled consumers about ChatGPT’s safety. Arkansas says eight other states filed similar suits that day, Utah among them; Iowa’s release names seven.

• Microsoft Puts Global Generative AI Use at 18.8% in Q2

  Microsoft says 18.8% of the world’s working-age population used a generative AI product in June 2026, up from 17.8% in the first quarter. The report also finds a widening gap between the Global North and South and growing use of open-weight models through APIs.

• Trump Hosts Anthropic CEO Dario Amodei for Late White House Dinner

  President Trump dined privately with Anthropic CEO Dario Amodei at the White House late Sunday, their first one-on-one meeting, after months of conflict over AI safety and the Pentagon’s supply-chain designation. Amodei is due at a Tuesday White House lunch.

• GitHub Adds Project-Level Sandboxing to Its Copilot App

  GitHub announced local sandboxing for the GitHub Copilot app on September 23. Users can configure it per project for local repository and working-tree sessions.

• DeepSeek’s DSec Supports 380,000 Agent Sandboxes at Once

  DeepSeek’s DSec platform supports more than 380,000 concurrent sandboxes for agent training and evaluation. A technical report describes how it manages bursty workloads, image storage and pauses in GPU training.

• Runway Launches Ads Engine That Learns From Campaign Spend

  Runway introduced Runway Ads on September 30, an advertising tool that generates creative, publishes approved variants to Meta, Google and TikTok, and uses campaign results to make more. The company says its own weekly ad volume rose from 77 to roughly 900 during its internal rollout, while return on ad spend doubled.

• AISI Models Chained Cloud Flaws Into a Five-Step Attack

  The UK AI Security Institute says frontier models found and chained real misconfigurations in an isolated copy of its AWS research platform. One model completed a five-step route to unauthorized data for under £150 in token costs, while basic alerting missed the activity.

• PwC and Cohere Launch Global Enterprise AI Alliance in Canada

  PwC and Cohere announced a global alliance launching first in Canada to help organizations deploy enterprise AI and measure business value. The partnership combines Cohere’s AI platform and models with PwC’s transformation, technology and risk expertise, with deployment options including private cloud, on-premises and air-gapped environments.

• Tether Invests in AIsa to Let AI Agents Pay Per Use

  Tether has made a strategic investment in AIsa, whose platform lets AI agents access paid data, software and computing resources through one interface. AIsa says its seed extension brings total funding above $9 million and that it has begun enterprise pilot programs.

• Misleading AI Summaries Cut Memory Accuracy to 44.8% in Study

  A study by Mattea Sim, Yael Eiger and Tadayoshi Kohno found that misleading AI-generated summaries reduced participants’ accuracy on a memory question about a traffic sign. The researchers also found that all 20 summaries they tested contained errors, with omissions the most common.

• Army’s ECHOTRIBBLE Project Seeks Autonomous Cyber Agents

  Public-facing material describes ECHOTRIBBLE as an Army cyber-defense prototype, while documents reviewed by The Intercept describe a project with an offensive focus. CFIC says one company was selected for an award, but its public page does not identify the company or detail safeguards against unintended damage.

• AIB Signs Nebius for 50 MW of U.S. Data-Center Capacity

  AIB Data Centers announced a binding 12-year agreement to provide Nebius with 50 MW of critical IT capacity at a planned southeastern U.S. facility. The release describes the site’s power arrangements and expected financing, but does not provide a delivery date or contract value.

• Instinct Raises $1 Billion at a $10 Billion Valuation

  Instinct says it has raised an additional $1 billion in a Series C round backed by Sequoia Capital, Benchmark Capital and Coatue, valuing the company at $10 billion. The company’s personal AI agent remains in early access, with new features for concierge tasks, coordination between users and location sharing.

• Extend says Verify detected 96.3% of fraudulent documents

  Extend announced Verify, an API that checks documents for signs of fraud before businesses or AI agents rely on them. The company says its benchmark evaluation detected 96.3% of fraudulent documents with a 4% false-positive rate, but the announcement does not disclose the test’s size or composition.

• White House Seeks U.S. Review Before UK Tests OpenAI and Anthropic Models

  The White House has asked OpenAI and Anthropic to let U.S. officials review new models before sharing them with the U.K.’s AI Security Institute. The request follows Anthropic’s decision to limit access to Mythos 5.1 and comes as Washington weighs how to assess cybersecurity risks in advanced AI systems.

• AI Chatbots Shift Worker Choices by Echoing Their Values, Study Finds

  A Scientific Reports study finds that AI recommendations framed around users’ political and moral values can increase idea endorsement and willingness to pay. The effect is strongest among people with firmer political views, raising concerns about AI influence in workplace decisions.

• Together AI Launches Custom Post-Training for Open Models

  Together AI introduced a post-training service that lets developers define training workflows while the company provides GPU resources and infrastructure. Its launch announcement also describes supported models, a temporary B200 price and reports from early customers.

• BootLoops Brings Certified Computation to AI Science Agents

  BootLoops combines computational tools with protocols for checking AI-assisted quantitative research. Its examples span mathematical physics, population genetics, ecology and phylogenetics, with researchers involved in assessing the scientific value of results.

• Nona Builds a 366M-Parameter Model for Human Heavy-Chain Antibodies

  Nona Biosciences says its HCAbLM model was trained on 31.8 million fully human heavy-chain-only antibody sequences. The company reports that the 366-million-parameter system outperformed larger general-purpose and antibody models on developability benchmarks.

• Anthropic Agents Outworked Employees by Six to One in August

  Anthropic says its AI agents surpassed employees’ work hours in March 2026 and logged six times as many hours as their human handlers by August. The company’s study also tracks shifts in junior technical hiring, employee output and the growing time spent supervising agents.

• AAIF Adds Four Agent Projects, Bringing Its Total to 10

  The Agentic AI Foundation announced four projects from Anthropic, OpenAI, Hugging Face and Alibaba, expanding its hosted portfolio to 10. The additions span agent skills, model interfaces, reinforcement-learning environments and sandboxes for AI applications.

• Meta Parts Ways With Employees Hired From Virtue AI

  Meta says it is letting go of employees hired from AI security startup Virtue AI, four months after the group joined in June. The company attributed the split to clashing work styles and said its AI safety focus continues.

• Hassabis Warns Major AI Security Incidents Could Come Within Two Years

  Demis Hassabis called recent AI-agent hacking incidents “early warning shots” and said more serious incidents could follow if companies and governments do not improve security. Microsoft president Brad Smith urged sensible safety laws and regulation, while Hassabis called for adaptable certification and international standards.

• Basecamp Research Raises $140M for AI-Designed Medicines

  Basecamp Research has raised $140 million in an oversubscribed Series C led by S32 to advance its EDEN biological AI models and therapeutic pipeline. The London-based company plans to begin with in vivo cell therapy while expanding pharmaceutical partnerships.

• European Commission Publishes EU KIDS Act Proposal on Child Safety Online

  The European Commission’s EU KIDS Act proposal addresses risks to minors from certain online services and AI systems while seeking a common age threshold for autonomous account creation on specified platforms.

• Vals AI Finds Four Models Kept at Least 80% of Their Scores at Lower Effort

  At their lowest reasoning setting, four of seven models kept at least 80% of their Vals Index score for half the cost or less. The study also found that lower settings sometimes matched leaderboard scores at lower cost, while results varied across tasks.

• Hugging Face Proposes Model Integration Contracts for Testing

  Hugging Face has proposed Model Integration Contracts, small executable fixtures that let Transformers test model behavior on CPU during pull requests. The proposal follows a silent UMT5 attention bug that escaped detection for nearly two months.

• Ascerta Raises $18 Million to Measure Enterprise AI Returns

  Ascerta, formerly Pay-i, announced an $18 million Series A led by Dell Technologies Capital on September 30, 2026. The Bellevue company says its platform connects enterprise AI costs and usage to business outcomes.

• Meta Claimed $3.9 Billion in Research Tax Credits for AI Data Centers

  Meta classified its AI data centers as experimental “pilot models” to claim federal research tax credits, according to a New York Times investigation. The credits cut $3.9 billion from Meta’s tax bill in 2025, while the company’s filings identify research credits as a source of tax uncertainty.

• UK Puts Global AI Standards at Center of G20 Presidency

  Prime Minister Andy Burnham says the UK will make artificial intelligence the focus of its 2027 G20 presidency. He proposes a single set of global principles and standards covering transparency, access and AI safety.

• Microsoft Details Agent-Led Migrations Across Huge Codebases

  Microsoft says AI agents are helping small teams migrate codebases that span hundreds of millions of lines, including work on Windows and Azure. Its approach pairs automated changes with reviewer agents, testing and formal verification to check whether migrations preserve expected behavior.

• Manchester Trains NVIDIA Earth-2 to Forecast UK Air Pollution

  Researchers at the University of Manchester have adapted NVIDIA Earth-2 models for high-resolution air-pollution forecasting across the United Kingdom. The system trained on Isambard-AI in two days and can run inference on NVIDIA’s DGX Spark desktop computer.

• NVIDIA Trains Agents to Recover From Early Errors With PivotOPD

  NVIDIA highlighted PivotOPD, a training method that teaches multi-turn AI agents to avoid pivotal errors and recover when they happen. In tests, the method recovered from 72.7% of 72 replayed ALFWorld mistakes, compared with 8.3% for the base model.