AI.info

The Pulse — Page 9

Browse The Pulse on AI.info.

• Google Research Tests Watch-Based Hypertension Screening

  Google Research evaluated a smartwatch system that it says uses optical and motion sensors to detect signs of hypertension without cuff calibration. In a clinical study, its screening results were compared with 24-hour ambulatory blood pressure monitoring.

• Reddit Keeps Three Claims Against Anthropic Alive in Scraping Suit

  A San Francisco judge let Reddit pursue breach-of-contract, contract-interference and unfair-competition claims against Anthropic over alleged scraping of Reddit data. The court gave Reddit until October 16 to amend its unjust-enrichment and trespass claims.

• Wes Moore Recruits Governors for a Bipartisan AI Framework

  Maryland Gov. Wes Moore says he and Indiana Gov. Mike Braun will develop a bipartisan governors’ framework for artificial intelligence. The initiative comes as Moore criticizes Washington’s approach and several states adopt their own AI measures.

• Singapore’s MAS Sets AI Risk Expectations for Financial Firms

  Singapore’s Monetary Authority has issued AI risk-management guidelines covering all financial institutions and AI technologies. The framework sets staged implementation dates and makes firms accountable for AI supplied by third parties.

• OpenAI Says AI Helped Find Two Critical PHP Flaws

  OpenAI says Calif used GPT-5.6 Sol to find a critical PHP flaw, while Wiz independently found both vulnerabilities. PHP released fixes for two flaws that can allow unauthenticated remote code execution and urged users to update.

• Atos Launches Genesis Intelligence AI Suite and Codex

  Atos announced Genesis Intelligence, a sovereign AI suite for organizations with regulated or sensitive data. Its Atos Codex component is designed to turn operational data and AI insights into actions under human oversight.

• Agent-Reflex Trains Qwen3-8B to Recover From Tool Failures

  A paper on arXiv describes Agent-Reflex, a three-stage training method for helping Qwen3-8B recover from faulty tool responses and user corrections. The authors report scores of 58.6 across five long-horizon tool-use benchmarks and 29.1 on AgentNoiseBench.

• One AI training step cut review diversity by 11%

  A University of Maryland study finds that adding synthetic peer reviews to AI training compresses ratings and reduces semantic diversity. The researchers introduce TrustReviewer, an open-source system designed to preserve broader scientific judgments.

• OpenAI Maps Internal Signals Tied to Model Metagaming

  Researchers at OpenAI and a collaborator from Apollo Research identified four internal signals associated with metagaming in an OpenAI o3 reinforcement-learning run. Their experiments found the signals could affect answers without appearing in written reasoning, while cautioning that the results may not generalize beyond the tasks studied.

• Mer Survey of 541 Generative AI Users at Japanese Companies Finds Gaps in Data Preparation

  A survey of 541 people at Japanese companies using generative AI found that 79.1% consider their business data prepared, while just 3.7% reported no particular data problems. Mer also reported gaps in connecting and maintaining data, even as 81.0% plan to expand AI investment within the next year.

• Researcher Traces 16,500 UNCTAD API Scans to Likely OpenAI Agents

  A researcher counted more than 16,500 scans of UNCTADstat’s public data API between April 13 and June 19, 2026. Traffic patterns and wiki logs led him to say OpenAI agents were likely responsible, though the evidence does not establish that they coordinated with other agents.

• AutoTuneBench Recasts LLM Inference Speed Claims

  A sole-author arXiv paper argues that reported gains from agent-driven GPU-kernel and serving-engine tuning can depend heavily on baseline choice, machine, task headroom and measurement design.

• Synopsys Explores Chinese AI Ties for Chip-Design Tools

  Nikkei Asia reported that Synopsys is exploring partnerships with Chinese AI labs to adapt chip-design tools for the Chinese market, with no partners or signed agreements identified in the reporting. The exploration comes alongside Synopsys’ announced agreements with OpenAI and Amazon and its fiscal 2027 revenue guidance of $11.15 billion at the midpoint.

• Dextr AI Raises $6.7M to Expand Its Hotel Agents

  Dextr AI raised a $6.7 million seed round led by Elevation Capital, with Foundation Capital participating. The San Francisco startup plans to use the funding to hire engineers, expand software integrations and build more agents for hotel operations.

• RuleTrek Benchmark Finds Agents Miss Complex Rules

  RuleTrek tests 11 language models on tasks that require finding and applying rules scattered across files. Its results show lower end-to-end scores as rule graphs grow larger or deeper.

• Ofcom Advisory Committee Presses Platforms on AI Election Risks

  Ofcom’s Online Information Advisory Committee urged online platforms to work more closely with UK authorities on election threats. Its October 13 report also called for more research into generative AI and democracy.

• AMD Adds a CPU-Threading Route to GPUs in ROCm 10.1

  AMD’s ROCm 10.1 release, published October 5, adds hipThreads to map standard C++ threading concepts onto AMD GPUs without a full HIP rewrite. The update also changes GPU data transfers, host-memory placement, compiler tooling and support for running ROCm inside WSL2.

• Flex Secures $2B Axiom Investment Ahead of 2027 Spin-Off

  Flex agreed to sell $2 billion in convertible preferred shares of Axiom Solutions International to investors led by General Catalyst and Koch Equity Development. The deal values Axiom at an initial $37.5 billion enterprise value and supports Flex’s planned first-quarter 2027 spin-off.

• NaiveAI Releases 309B Model With 1M-Token Context

  NaiveAI announced Naive-N0.5-Flash, a 309-billion-parameter open-weight model with 15.5 billion active parameters and a native 1-million-token context window. The company says it uses AI-assisted research and a custom inference system for coding and AI research workloads.

• Google Opens Gemini and Search to More Third-Party Apps

  Google has launched a preview platform where developers can register and manage apps intended to work in Search, the Gemini app or both. The company says more than 50 Connected Apps are available across the products, with integrations from Adobe, Experian, Localiza and Peloton coming to Search in the coming weeks.

• Microsoft Sees AI Compressing Cyberattack Steps to Seconds

  Microsoft’s 2026 Digital Defense Report says AI is speeding up parts of the cyberattack chain, while vulnerability weaponization can follow discovery in less than 24 hours. The company says most complex intrusions still involve human direction, even as automated activity grows.

• New York Bans Unsafe AI Companion Features for Minors

  New York Gov. Kathy Hochul signed a law barring covered AI companions from using specified unsafe features with minors, including simulated relationships and self-harm encouragement. The law takes effect January 1, 2027, and gives the state attorney general enforcement authority, including civil penalties of up to $25,000 per violation.

• Google’s R4T Turns Reinforcement Learning Into Faster Retrieval

  Google Research has detailed Retrieve-for-Train, a system that uses reinforcement learning during training and a diffusion model for faster multi-result retrieval. The 53.9-million-parameter retriever produced 10 retrieval directions in one pass and delivered 12x to 20x lower latency in Google’s tests.

• PhAI Labs Presents ScienceIDE as an Environment for Training Scientific AI Agents

  PhAI Labs’ ScienceIDE turns scientific code repositories into executable environments for task generation, execution, verification and evaluation, supporting supervised fine-tuning, reinforcement learning and scientific-code assessment.

• SemiAnalysis GDPx Gives Its Top AI Agent a 12% Score

  Claude Opus 5.5 leads SemiAnalysis’ GDPx benchmark with a 12% aggregate score across 11 domains. The benchmark uses different grading methods, and its finance and legal results include partial scores alongside zero scores under intended binary grading.

• Schneider Electric Agrees to Buy PTC for $22.6 Billion

  Schneider Electric agreed on October 5, 2026, to acquire PTC in an all-cash deal valuing the software company’s equity at $22.6 billion. The $205-per-share offer includes planned financing of €5 billion to €6 billion in equity and €16 billion to €17 billion in new debt.

• Clio Buys Learned Hand to Bring AI Into U.S. Courts

  Clio acquired Learned Hand, whose AI workspace helps judges and clerks review filings, research law and prepare draft court documents. The deal expands Clio into the judiciary and brings Learned Hand together with Clio Library and Clio Docket.

• Amazon Announces Selling Partner Plugins for Quick and Claude

  At Amazon Accelerate, the company announced plugins linking sellers’ accounts to its Quick assistant and Claude, alongside upgrades to Seller Assistant and Seller Central.

• GE HealthCare Launches AI System to Forecast Hospital Bottlenecks

  GE HealthCare has announced CareIntellect for Operations, an AI-enabled software service that forecasts hospital capacity constraints up to 72 hours ahead. The Queen’s Health Systems and Duke Health will be the first clinical evaluation sites.

• Nolla Health’s AI Writes Initial Acne Prescriptions in Utah

  Nolla Health says Utah residents 18 and older can use its app for AI-issued initial acne prescriptions, with the state’s authorized-pilots listing marking the program active from October 5, 2026, through October 5, 2027. The pilot is limited to topical treatments for adults with mild or moderate acne and includes physician review and other state-mandated safeguards.

• Meta and Singapore Police Target 3.7 Million Scam Assets

  Meta says information shared by the Singapore Police Force helped it take action against more than 3.7 million scam-linked accounts, pages and other assets in 2026. The effort included 3.6 million dormant shell pages removed before scammers could activate them.

• California Subpoenas OpenAI Over Cybersecurity Risks

  California Attorney General Rob Bonta served OpenAI with an investigative subpoena on September 30, 2026. The subpoena expands a state inquiry into cybersecurity incidents involving the company and its AI models, including the Hugging Face incident.

• UCR Researchers Separate LLM Confidence From Correctness

  UC Riverside researchers identified distinct internal features associated with language models’ confidence and correctness, and tested how modifying those features affected answers.

• AWS Releases Open 2B Decision Model for AI Agents

  AWS released Strands Decider 2B, an open-source model that selects among predefined answers and returns confidence scores. The model is designed to run locally and handle agent workflow decisions without generating text.

• Microsoft’s Magnetar Proves Code Up to 39 Times Faster

  Microsoft Research unveiled Magnetar, a zero-knowledge proof system that checks programs compiled for RISC-V and relies on standard hash functions. In tests across 15 workloads, it outpaced three competing systems, with reported average speedups ranging from 9.6 to 39 times.

• Qwen Cuts Live Translation Lag to 2.3 Seconds

  Qwen has released Qwen3.8-LiveTranslate, a real-time interpretation model that reduces average lagging from 2.8 seconds to 2.3 seconds. The model adds speaker separation, synchronized bilingual output, long-context disambiguation, and support for 60 input languages.

• BAN Projects Up to 617 Million Tonnes of AI-Driven E-Waste by 2050

  The Basel Action Network says AI-driven electronic equipment retirements could total 395 million to 617 million tonnes between 2025 and 2050.

• Anthropic Warns Government Disputes Could Damage Commercial Ties

  Anthropic warned that government actions and perceptions could affect commercial relationships, even though government agency contracts account for less than 1% of its annual revenue. A Reuters report based on an IPO prospectus it said it had seen describes possible business disruption and reputational harm.

• Strata Runs Qwen’s 125B Model on Gaming PCs, Reports Selected Speedups

  Strata says its inference engine runs Qwen3.8-Flash-Next on consumer PCs by distributing work across the GPU, system memory and processor. Release v0.1.43.1, published October 12, fixes issues in the previous release, while v0.1.43 reports speedups on selected hardware.

• Manus 2.0 Adds Event-Triggered Workflows and Personal Agents

  Manus unveiled Manus 2.0 on September 28, adding event-triggered Automations, a purchasable Cloud Computer and the new Cue app for personal agents. The release also upgrades the desktop app to Manus Studio and introduces editing tools for video and games.