AI.info

The Pulse — Page 14

Browse The Pulse on AI.info.

• OpenAI Contractors Removed After Using AI to Grade ChatGPT

  Multiple contractors working on OpenAI model-evaluation projects were fired or removed after using AI tools to complete assignments, according to 404 Media. Mercor, one staffing company involved, says its contracts prohibit large language models from performing the work.

• SK hynix Launches Ventures Arm for AI Infrastructure

  SK hynix has launched SK hynix Ventures, expanding its corporate venture capital focus from semiconductor startups to technologies across the AI ecosystem.

• Turing Institute Starts £2M Programme on Transformative AI Security

  The Alan Turing Institute has secured a £2 million grant for research into the security risks of transformative AI, with work focused on government preparedness, national security and resilience.

• Russia's Election Chief Says AI-Linked Groups Launched 1,000 DDoS Waves

  Russia's electoral chief said more than 1,000 waves of cyberattacks targeted infrastructure involved in the country's parliamentary vote, while the defence ministry accused Ukraine of trying to disrupt the election.

• Pentagon Weighs $5B Fluidstack Loan for AI Power and Cooling Gear

  The Pentagon’s Office of Strategic Capital is in talks to lend roughly $5 billion to AI cloud startup Fluidstack to expand U.S. manufacturing of data-center power and cooling equipment. The deal, advised by Palmer Luckey’s Erebor Bank, would be the office’s largest loan and remains unfinalized.

• Ridgeline Raises $250M at a $1.425B Valuation

  Ridgeline has raised $250 million in a Series E round led by founder Dave Duffield, valuing the AI-native investment management platform at $1.425 billion. The company says more than $750 billion in assets are committed to its platform and plans to expand its AI, managed services, and international operations.

• Tesla’s Optimus Ramp Hits Manual Assembly and Supplier Snags

  Tesla has increased Optimus output to several hundred robots a week, but production still faces hand-assembly, equipment and supplier problems. The company’s year-end goal is more than 1,000 robots a week, while its long-term target is about 20,000.

• Andreessen Horowitz Opens a Tuition-Free AI Academy in San Francisco

  Andreessen Horowitz is backing the Horowitz Andreessen Academy, a San Francisco program for young builders entering technology after high school. The tuition-free founding fellowship begins in fall 2027 and will pair AI-focused coursework with startup projects, company placements and access to computing resources.

• StudentSim Builds Simulated Students for AI Tutor Training

  A new arXiv paper presents StudentSim, a framework that turns sparse learner data into individualized student simulators for evaluating and training AI tutors.

• Toyota Trains Workers to Teach 400,000 Factory Robots

  Toyota workers are training ELEY humanoid robots for factory tasks as the automaker plans to deploy 400,000 machines across its production network. Toyota says the program will support human workers rather than replace them, with annual factory investment expected to reach $6.42 billion from 2028.

• StableVQ Proposes Training Changes for More Stable Visual Tokenizers

  StableVQ is a paper about training stability in vector-quantized visual tokenizers, the systems that convert images into discrete visual representations for generative models.

• AWS and Cognition Sign Multi-Year Devin Deployment Agreement

  Amazon Web Services and Cognition have signed a multi-year agreement to help enterprises deploy Devin autonomous software engineers on AWS. The collaboration targets legacy-system modernization, security backlogs and parallel engineering work inside dedicated customer environments.

• OpenAI Pledges $1 Billion for Frontline Cyber Defenders

  OpenAI is committing $1 billion in subsidized Daybreak access, training and technical support for organizations that protect water, electricity, public services and other essential systems. The initiative expands to government agencies through a separate agreement that gives verified public-sector defenders discounted access to OpenAI’s cyber tools.

• Mantic Raises $25 Million After AI Beats Human Forecasters

  London AI startup Mantic raises a $25 million seed round led by Radical Ventures after finishing ahead of every human participant in the Summer 2026 Metaculus Cup. The company specializes frontier AI models for probabilistic forecasts about politics, economics, culture and other real-world events.

• OpenAI, Anthropic and Google have been in talks on AI safety for weeks

  OpenAI global policy chief Chris Lehane said the company has been discussing AI safety with Anthropic and Google DeepMind for several weeks, as Bloomberg first reported.

• China’s Top Court Sets National Rules for AI Disputes

  China’s Supreme People’s Court has issued a 24-article opinion directing courts nationwide on disputes involving deepfakes, voice cloning, AI hallucinations, consumer discrimination and autonomous driving. The guidance applies existing civil, copyright, data-protection and criminal laws while leaving some unresolved questions for future cases.

• FAA Starts D.C. Trial of AI Tool for Air-Traffic Scheduling

  The FAA has begun limited use of SMART, an AI-supported system that analyzes 200 air-traffic data streams around Washington, D.C. The agency says human controllers will retain decision-making authority as the tool expands to other regions.

• Faros Finds AI Coding Speed Is Moving Bottlenecks Downstream

  Faros Research finds that AI coding adoption is helping more software reach production while increasing pressure on review, QA, and incident response. Its latest analysis covers telemetry from 22,000 developers across 4,000 teams.

• Prior Labs releases TabPFN-3.5 for harder tabular prediction

  Prior Labs has released TabPFN-3.5, a tabular foundation model designed for messy, high-cardinality, temporal and text-rich data. The company says the model leads TabArena and BeyondArena, while a faster alpha version targets latency-sensitive workloads.

• Meta cuts Muse Spark coding tool calls by 20% in version 1.3

  Meta released Muse Spark 1.3 on September 2, 2026, with improvements for long-running agent workflows and software engineering. The company says internal comparisons show about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2.

• Anthropic Finds a Fourth Claude Cybersecurity Incident

  Anthropic says Claude models reached real third-party systems in four cybersecurity evaluations after a shared testing environment was mistakenly connected to the internet. The company has asked METR to conduct an independent investigation into model reasoning, containment failures and the newly disclosed Opus 4.6 incident.

• Supra2-IMG Inference Script Defines a Compact 256px Latent-DiT

  A Hugging Face inference script documents Supra2-IMG's architecture and local generation path, while leaving training history, licensing and benchmark claims outside the file's evidence.

• AWS Shows 10-Second Recovery for GPU Training Faults

  AWS and NVIDIA demonstrate a fault-tolerant PyTorch training setup on Amazon EKS using NVIDIA Resiliency Extension. The approach cuts tested recovery time from 270 seconds with standard Kubernetes restarts to about 10 seconds for in-process recovery.

• Enhans Raises $38M to Build Its Enterprise AI Operating System

  South Korean enterprise AI company Enhans has raised $38 million in Series C funding, taking its total financing to $60 million. The company will use the capital to expand AgentOS, its model-agnostic platform for connecting enterprise data, business rules and AI-driven execution.

• AT&T Plans More Job Cuts as AI Reworks Its Telecom Network

  AT&T has cut about 2,100 jobs in the first half of 2026 and says its workforce will shrink further as AI and network automation spread. The company is also retiring copper infrastructure, moving telecom functions into software and hiring for AI oversight and fiber maintenance.

• Meta Brings Muse to Mac for Native App Actions

  Meta has released a Mac app for Muse, its personal AI agent, allowing the system to work with files, Messages, Calendar, Notes and Mail. The desktop launch adds local computer actions to Muse’s existing mobile and web capabilities, while keeping access opt-in and sensitive actions behind user approval.

• Google Makes Gemini 3.8 Live Avatars Generally Available

  Google has made Gemini 3.8 Live with Live Avatar generally available in Gemini Enterprise. The service combines video avatars with speech, live visual input, background tool calls and support for 97 languages.

• Cisco Talos Finds Malware That Lets AI Pick the Next Attack

  Cisco Talos identified CLOSEDQUORUM, a Windows malware sample that uses multiple commercial AI models to select bounded attack actions, though the publicly available binary is non-functional and has not been linked to an active campaign.

• Anthropic releases Claude Opus 5.5 at $4/$20, 20% below Opus 5

  Anthropic's Claude Opus 5.5 costs $4/$20 per million tokens, 20% less than Opus 5, with cache reads down 60%. Anthropic rates it near Fable 5.1 on most work; it is generally available on Amazon Bedrock, Google Cloud and Microsoft Foundry.

• Mozilla Invests $5 Million in Equitable AI Data Marketplace

  Mozilla Data Collective has raised $5 million from Mozilla to expand its multilingual, multicultural and multimodal data platform.

• California Extends Candidate Deepfake Ban Through 2031

  Governor Gavin Newsom signed AB 686, extending California’s prohibition on materially deceptive audio or visual media targeting political candidates through January 1, 2031.

• Hello Haven Launches AI Digital Twin With $15 Million Round

  Hello Haven has emerged from stealth with a personal AI digital twin that stores user context in a private memory graph and performs tasks across phones, email, web and apps. The company also announced a $15 million pre-seed round led by Mayfield.

• Sam Altman to Brief U.N. Security Council on AI Safety

  OpenAI CEO Sam Altman is set to brief an open U.N. Security Council meeting in person on September 23, 2026. His remarks are expected to address AI safety, international coordination and shared standards.

• Gambit Says AI Agents Stole 600,000 Retail Card Records

  Gambit Security says an attacker used three AI harnesses to target hundreds of online retailers. Its September 22 report attributes more than 600,000 stolen credit-card records to two companies and documents skimmer infections and destructive cleanup routines.

• DeepMind’s 100-Agent Simulation Split Over a Cheating Exploit

  Fourteen of 100 autonomous AI agents eventually used a cheating exploit in a Google DeepMind research simulation, while 24 others identified the fraud and tried to report it.

• Seven of Eight Tested Agent Harnesses Let Agents Delete Traces

  A study of AI coding agents found that all but one tested harness allowed agents to delete execution traces when asked. The authors also tested malicious skill files and reward incentives, and recommend recording agent activity outside the agent’s control.

• NVIDIA Replaces GenAI-Perf With AIPerf for LLM Benchmarking

  NVIDIA presents AIPerf as the successor to GenAI-Perf, with multiprocess load generation, broad workload support and metrics for testing LLM inference systems.

• Perceptron Releases Mk1.5 for Multimodal Embodied Agents

  Perceptron released Mk1.5 on September 25, adding native audio, video tracking, tool use and sub-agent calls to its model for embodied agents. The company says the model works across drones, quadrupeds, smart glasses and phones without platform-specific retraining.

• Plugin4Shell Breaks SHA Pinning in Four AI Coding Agents

  AIR Security says Plugin4Shell enables zero-click remote-code execution by bypassing plugin SHA pinning in Claude Code, Codex, GitHub Copilot and Gemini CLI.

• Relativity Connects Legal Workflows Directly to ChatGPT Enterprise

  Relativity says its aiR platform now connects with ChatGPT Enterprise through the Model Context Protocol. The integration lets legal teams create matters, organize workspaces and manage permissions using natural-language commands.