AI.info

The Pulse — Page 40

Browse The Pulse on AI.info.

• &Truth Raises $70 Million to Automate Code Proofs

  &Truth emerged from stealth on October 29 with $70 million in seed funding to automate formal verification of software. Its system will use Lean to produce independently checkable proofs that code meets precise specifications, the Palo Alto and Singapore company says.

• OpenAI Opens Managed Agents API to All Developers

  OpenAI has placed its Agents API into public beta, giving developers access to a managed Codex harness for long-running, tool-using agents. The service supports durable sessions, hosted or third-party sandboxes, MCP connections, context compaction and parallel subagents.

• May Mobility agrees to go public in $1.4B SPAC deal

  May Mobility has agreed to merge with ACP Holdings Acquisition Corp. in a transaction valuing the autonomous ride-hail company at about $1.4 billion. The deal could provide up to $337 million in gross proceeds and put May Mobility on Nasdaq under the ticker MAY.

• NVIDIA Adds Automated Recovery for Slurm Training Jobs

  NVIDIA says its recovery service recorded about 8,550 anomaly detections on one training cluster between July 9 and October 9. The open-source tool automates recovery for Slurm training jobs, with new deployments set to monitor mode before taking scheduler actions.

• A Probe Flags AI Agents Before They Leak Secrets

  A preprint by Leonidas Raghav tests whether a model’s internal signals can flag the call that exposes a protected secret before the agent generates its reply. In tests with Qwen3-32B and gpt-oss-20b, the warning often came just in time—but also flagged permitted disclosures and benign activity.

• Multiply Labs Raises $75M to Scale Robotic Drug Manufacturing

  Multiply Labs announced a $75 million Series B led by Dr. Patrick Soon-Shiong and NantWorks to expand its robotic biologics-manufacturing platform. The San Francisco company says the systems can reduce cost per dose by 74% and deliver up to 100 times the throughput of manual production.

• Anthropic Signs First Publisher Licensing Deal With the AP, Adweek Reports

  Anthropic reportedly signed a content and data licensing deal with the Associated Press just before the midterm elections, according to two people familiar with the matter cited by Adweek. The report says the agreement would give Claude access to timely AP information, while noting that financial and other terms were not disclosed.

• LG and Microsoft Expand AI Infrastructure Partnership

  LG and Microsoft are broadening cooperation across AI data centers, physical AI and enterprise AI tools.

• Cognex to Buy RealSense for $500 Million in Cash

  Cognex is acquiring RealSense for approximately $500 million in cash, expanding into robotic perception and Physical AI.

• SimBioSys Wins FDA Clearance for MRI-Based Breast Cancer Risk Tool

  The FDA cleared SimBioSys’s TumorSight Risk software to estimate five- and ten-year breast cancer recurrence risk for a defined group of patients. The clearance covers women with HR-positive, HER2-negative stage I–IIIA disease whose lymph-node status is N0 or N1.

• Cloudflare Announces Trusted Agent Delegation Coalition on AI Agent Traffic

  Cloudflare announced the Trusted Agent Delegation coalition to bring representatives from across the ecosystem together for discussions about AI agent traffic. Its post describes how remotely hosted agents can obscure the people behind requests and outlines approaches for sites and agents to identify users, communicate policies and manage access.

• Snap Brings SPECS Intelligence to iPhone, Mac and AR Glasses

  Snap introduced SPECS Intelligence on September 16, 2026, an anticipatory AI service spanning iPhone, Mac and SPECS augmented-reality glasses. The iOS preview is available to U.S. users aged 18 and older, while the fuller Mac experience begins with an invitation-only waitlist.

• Anthropic Asks Law Firms to Set Rules for AI Deal-Making

  Anthropic is consulting law firms on fair negotiation norms for AI agents acting on users’ behalf. An earlier Claude marketplace test found that stronger models secured better prices, while participants often did not perceive the disadvantage.

• Cantina Releases Apex Flash-1, a 321B Open-Weight Security Model

  Apex Flash-1 solved 40 of 60 tasks across 20 held-out vulnerability cases, at a lower estimated run cost than the comparison models, according to Cantina. The company released the 321-billion-parameter model with Yeta Labs on October 2, 2026, describing it as a worker for focused vulnerability investigations rather than a stand-alone security agent.

• AI Helped Tighten the Grothendieck Constant’s Bounds

  A Princeton research team used AI to help tighten bounds on the Grothendieck constant, with researchers stepping in when the system repeated failed approaches. The case study describes what the system proved, what the authors verified, and where human judgment shaped the research.

• OpenAI Disrupts Russian and Iranian Influence Operations

  OpenAI banned Russian and Iranian ChatGPT account clusters that it ties to separate influence operations. Its investigation identified nearly 100 articles published or syndicated under seven fake journalist bylines linked to the Iran-origin campaign.

• OBLITERATUS Posts DeepSeek Model With a 4.30% Full-Refusal Rate

  OBLITERATUS released modified DeepSeek-V4.1-Flash-1017 weights, with an evaluation reporting a 4.30% full-refusal rate and a 78.28% harmful-assistance rate. The figures come from a published comparison of completed responses to 838 prompts; the model card also reports a separate MMLU result.

• UK Accepts All 44 AI Healthcare Regulation Recommendations

  The UK government has accepted all 44 recommendations from its National Commission into the Regulation of AI in Healthcare. The MHRA will test ongoing safety monitoring through AI Airlock Phase 3, while a full implementation roadmap is due by Spring 2027.

• OpenAI Plans to Buy Google’s AI Chips, The Information Reports

  OpenAI plans to buy Google’s Tensor Processing Units, according to an October 27 report by The Information, but the report gives no price or purchase details. The move would extend OpenAI’s use of Google chips as the company hires engineers to optimize its models for TPUs.

• One Stubborn AI Agent Can Steer a Multi-Agent Network

  A study of language-model multi-agent systems finds that one highly stubborn and persuasive agent can trigger a persuasion cascade and reshape a group’s collective opinion.

• Blueprints AI Pitches Permit Plans in Minutes, With Architects Still in Charge

  Blueprints AI’s second-generation software can generate coordinated building plans in minutes, the company says, while licensed professionals remain responsible for reviewing and approving the documents. TechCrunch reports that the Santa Monica startup is pitching the product to architecture firms, developers and real-estate owners.

• WindBorne Wins Air Force Contract Worth Up to $49 Million

  WindBorne Systems won a U.S. Air Force contract with a ceiling of $49 million for atmospheric data collected by its autonomous weather balloons. The Air Force has placed $1.7 million in orders so far, with payments tied to data it requests.

• Pope Leo Rejects Trump’s Claim That AI Safety Warnings Are ‘Fake News’

  Pope Leo XIV said on September 28 that warnings about artificial intelligence risks should be taken seriously, pushing back against President Donald Trump’s dismissal of them as a hoax. Leo called for political leaders, AI companies and social organizations to discuss the risks, while a Vatican commission works on the issue.

• China’s AI Feature Opens to 300,000 Yuan in First-Day Sales

  China’s first fully AI-generated theatrical feature, Sanxingdui: Future Past, opened on October 23, 2026, with Maoyan listing 300,000 yuan in first-day box-office sales. The film cost more than 10 million yuan to make, putting its early cinema performance well below its reported production budget.

• OpenAI Turns ChatGPT Into a Shared Workspace for Teams

  OpenAI announced ChatGPT Space, shared Pages, collaborative slides and scheduled team tasks at its September 29 DevDay. The additions bring documents, app integrations and AI agents into a common workspace, with features rolling out across paid plans.

• Study Finds 147 EU Politicians Linked to Deepfake Porn

  An Agora Digitale Transformation audit found that 147 of 5,872 EU national parliamentarians appeared in or were associated with deepfake pornography. Women lawmakers were 33 times more likely than men to be exposed, the study found.

• Uber and Wayve Launch London’s First Autonomous Ride Service

  Uber and British self-driving company Wayve have launched supervised autonomous rides across London, marking the United Kingdom’s first public autonomous ride-hailing service. Passengers can be matched with an electric Ford Mustang Mach-E at no extra cost, but a licensed safety driver remains onboard.

• Claude Turns Wiles’s Fermat Proof Into 13 Million Lines of Lean

  Claude produced the first complete computer-checked formalization of Fermat’s Last Theorem in 11 days, Anthropic says. The system generated 13 million lines of Lean code and proved 30,300 intermediate theorems while translating Andrew Wiles’s proof into a form computers could verify.

• Perplexity Runs Computer Fully On-Device on Apple Silicon Macs

  Perplexity launched a fully local version of its Computer agent for Apple silicon Macs on October 20. The mode runs its model and task processing on the Mac, while users can still authorize cloud access for certain tasks.

• StepFun Ships 600B Step 5 Preview With 1M-Token Context

  StepFun has released Step 5 Preview, a sparse-MoE model with 600 billion total parameters, 27 billion activated parameters, a 1-million-token context window and fully open API access.

• Recraft to Add Ideogram 4.5 for Native-Resolution Editing

  Recraft says Ideogram 4.5 is coming to Recraft Studio, where it will join the platform’s image models and editing tools. Ideogram’s model page describes editing at an image’s original resolution and shows an example built from a 24.2-megapixel image.

• Fake Apex Accelerator Routed AI Coding Traffic to Its Servers

  The fake Web3 accelerator Apex used malicious npm packages to deliver malware and redirect AI coding-agent traffic, Elastic Security Labs says. Its researchers found that the packages also installed an attacker-controlled MCP server and offered security audits that returned fabricated results.

• Nvidia’s Board Opens Independent Inquiry Over Reports on Jensen Huang

  Nvidia’s board said on November 8 that independent directors, helped by outside counsel, will investigate reports that CEO Jensen Huang was targeted by a foreign intelligence effort. Huang apologised to staff for being unfaithful to his wife and said he will cooperate.

• Mount Sinai Study: Safety Prompts Cut Risky AI Choices by 6.5 Points

  A Mount Sinai study tested 20 open-weight AI models on more than 10 million responses to clinical scenarios. Brief safety reminders lowered potentially harmful choices from 16.6% to 10.1%, but did not eliminate them.

• NYC Council Secures AI Lab Testimony, Subpoenas SpaceXAI

  Meta confirmed voluntarily, while Anthropic, OpenAI and Google agreed to testify after subpoena warnings. The Council also plans to consider AI safety proposals at its October 5 hearing and subpoenaed SpaceXAI to compel its appearance.

• Vals AI ranks Claude Fable 5.2 first on its Vals Index

  Vals AI ranked Anthropic’s Claude Fable 5.2 first on its Vals Index, with a score of 71.31%. The model also led four index components, but placed lower on HLAB-Vals-1.1 and Tax Agent Bench.

• Satlyt Runs Google’s Gemma 3 1B Model on a Satellite

  Satlyt has deployed a quantized Google Gemma 3 1B model aboard a satellite to analyze telemetry, system logs and software faults in orbit. In fault-injection tests, the onboard system cut diagnostic data sent to Earth by about 64% while generating root-cause explanations and recommended actions.

• Cloudflare Previews Clairic, a Typed Data Extraction Model

  Cloudflare introduced Clairic on October 19 as a preview model on Workers AI for extracting, calculating, estimating and forecasting typed values from unstructured input. The company says it runs in a single pass, but warns that its early checkpoint can struggle with longer calculations and detailed forecasts.

• Clinical AI Review Finds Doctors Missed 26.6% of Erroneous ECG Readings

  A systematic review in npj Digital Medicine found that clinicians failed to correct or recognize 26.6% of erroneous interpretations across four real-world computerized ECG studies. The authors say human oversight should be measured by whether it intercepts errors, not treated as a guarantee of clinical AI safety.

• OpenAI Says ChatGPT Search Is Roughly 20% of Google’s Volume

  Adam Fry said ChatGPT’s search-like queries now amount to roughly one-fifth of Google’s query volume. The reports also describe how OpenAI defines a search and give separate figures for ChatGPT’s overall growth.