AI.info

The Pulse — Page 6

Browse The Pulse on AI.info.

• Ofcom Probes Meta Over Instagram Instants Risk Assessments

  Ofcom opened an investigation on October 6 into whether Meta assessed illegal-content and child-safety risks before launching Instagram Instants. The UK regulator says the probe is not a finding of wrongdoing and could lead to a fine of up to £18 million or 10% of qualifying worldwide revenue.

• 10x Genomics Launches Claude-Powered Sentira Single Cell

  10x Genomics announced Sentira Single Cell, a cloud application that lets researchers analyze Chromium data through natural-language questions while an AI agent plans and runs analyses. The service is available today, with promotional credits for users and consumption-based pricing tied to analysis needs.

• Google opens $899 Googlebook preorders with Gemini built in

  Google's September 21 announcement opens Googlebook preorders at $899 and describes its hardware, Android phone features, Gemini tools and availability.

• IMF Scenario Sees AI Market Correction Cutting Global Output 1.1% by 2028

  The IMF’s October 2026 World Economic Outlook models a scenario in which an AI-investment reversal and financial spillovers leave global output 1.1% below baseline by 2028. The report’s listed output effects do not fully add up to that total, a gap the scenario text does not reconcile.

• Serval Raises $150 Million at $3 Billion Valuation to Take On ServiceNow

  Serval announced a $150 million Series C led by Sequoia at a $3 billion valuation. The enterprise software startup plans to expand its AI workflow product and hire more than 500 people as it targets ServiceNow replacements.

• Anthropic Says Claude Found a CRISPR-Like Enzyme System

  Anthropic says Claude agents identified a previously uncharacterized enzyme system in bacteriophage DNA after a 21-hour search of reverse transcriptases. The company calls the system array-associated reverse transcriptases, or ART, but says its biological function is still unknown.

• Google’s TranslateGemma Brings Offline Translation to 55 Languages

  Google has released TranslateGemma, a three-model translation family built on Gemma 3, with evaluated results across 55 languages and on-device deployment options.

• Cloudflare Brings Deno’s Team Aboard; Runtime Gets One More Year of Updates

  Deno says its runtime will receive monthly bug-fix and security releases for another year before the company ends its development. Deno Deploy will close after six months, while JSR will continue operating.

• Moonshot Makes Kimi K3.1 Pro’s Open Weights Available

  Kimi K3.1 Pro is a 2.8-trillion-parameter mixture-of-experts model with a 1-million-token context window. Moonshot lists API access and currently recommends vLLM, SGLang and TokenSpeed for inference.

• Eiliko Toy Security Audit Finds Exposed Chats and Robot Controls

  Security researcher Sammy Azdoufal says Energize Lab’s Eiliko and related robots exposed private conversations, child profiles and controls for changing robot responses. His October 13 audit counted more than 9.1 million cleartext conversations and found that 32 of 40 remotely testable issues remained open.

• Google Sets Next Week’s First Orbital Test for Suncatcher AI Chips

  Google plans to send Tensor Processing Units into orbit next week on a prototype satellite for Project Suncatcher. The mission will test the chips against launch stress, radiation and heat-management challenges, not run an orbital data center.

• PFN Releases Open-Weight PLaMo 3 Translate Models in Four Sizes

  Preferred Networks has released four open-weight translation models, ranging from 0.8 billion to 31 billion parameters, for use on devices and servers without an internet connection. The release describes offline translation across 53 languages and includes differing revenue thresholds in its general license and commercial-licensing sections.

• Google Puts ScientistTwo on Display at COLM 2026

  Google’s COLM 2026 program listed a live demo of ScientistTwo, a multi-agent system designed to run machine-learning research from a problem statement through experiments and paper drafting. Its authors report improvements on 86 of 107 research tasks, while acknowledging that the system does not yet produce work at the level of conference spotlight or oral presentations.

• Meta Says It Removed More Than 3.7 Million Accounts, Pages and Groups in West Africa Scam Crackdown

  Meta says it removed more than 3.7 million accounts, pages and groups during a joint crackdown on scam networks in West Africa. The company also says it shared leads on more than 30 suspected scam centers; law enforcement reported 28 arrests and the rescue of 117 people in Ghana.

• White House Gives Jay Clayton 120 Days to Assess AI Risks

  The White House has formed a task force led by Director of National Intelligence Jay Clayton to report on artificial intelligence risks and opportunities within 120 days. The group will examine the federal government’s role, including how agencies respond to AI-related breaches, hacks and model jailbreaks.

• Third Circuit Affirms Thomson Reuters’ Win Against Ross Intelligence

  The Third Circuit held that Westlaw headnotes are copyrightable and that Ross Intelligence’s use of them to train an AI legal-search tool was not fair use, affirming Thomson Reuters’ win in what Reuters called a first-of-its-kind U.S. appellate ruling on AI training.

• UK Expands Frontier AI Cyber Tests Across Government

  The UK government will extend the use of frontier AI to find and fix vulnerabilities in public-sector systems after pilots identified more than 1,000 potential weaknesses. Ministers will also consider cyber-reporting rules for AI-related incidents and prepare for risks involving autonomous agents.

• Wall Street Builds $63 Billion Compute Vehicle for Anthropic

  Bank of America, Blackstone, Citi and Morgan Stanley announced a $63 billion transaction to finance Broadcom AI hardware for Anthropic through a special-purpose vehicle. Deliveries are expected to begin in 2027 under a previously announced strategic compute-capacity agreement.

• Subvocal Opens $499 Preorders for Silent-Speech Glasses

  The Subvocal Company is taking U.S. reservations for glasses designed to turn silently mouthed words into text. A refundable $199 deposit goes toward a $499 early-bird price, with shipping estimated for the first quarter of 2027.

• IBM Trains Granite Models to Resist Prompt Injection

  IBM Research authors report that LLMON post-training cut success on two indirect prompt-injection benchmarks to as low as 0.02%. A red-team evaluation found weaker protection against indirect attacks, while direct and multi-turn attacks remained substantially harder to stop.

• Island Raises $400M to Extend Security to AI Agents

  Island raised $400 million in a Series F round at a $6.4 billion valuation, led by Evolution Equity Partners. The Dallas company says it will use its enterprise security platform to govern work by employees and AI agents across browsers, networks and data.

• BizBench Finds Agents Break Rules While Completing Tasks

  BizBench evaluates AI agents on 50 simulated business tasks, checking whether they follow operational rules as well as reach the requested outcome. The authors found that some agents completed tasks after violating constraints, including cash-reserve requirements and approval rules.

• Abu Dhabi Creates TAMM Global to License Its AI Government Platform

  Abu Dhabi has established TAMM Global to license the architecture and operating methods behind its AI-powered government services platform. The company plans to adapt TAMM’s digital services for governments abroad, while the announcement gives no launch schedule or pricing.

• Mercedes and NVIDIA Extend Driving Partnership Toward Level 4

  Mercedes-Benz and NVIDIA say they will extend their partnership to develop driver-assistance and automated-driving systems using Mercedes-Benz’s MB.OS and NVIDIA’s computing, sensor and driving technologies. The collaboration includes future SAE Level 4 applications such as robolimousines; the October 16 release gives no launch date or vehicle for those applications.

• Gemini Adds In-Chat Editing for PDFs, Docs and Sheets

  Gemini users can preview and edit PDFs in the app on web and mobile, while Google Docs, Sheets, slide decks, Markdown, TXT files and code are supported on the web. The feature lets users make edits directly, ask Gemini to refine files or highlight text to ask questions.

• Moonshot Launches Kimi K3.1 Pro With 1M-Token Context

  Moonshot AI has released Kimi K3.1 Pro, a 3T-class open-weight model with a 1-million-token context window, alongside a faster Highspeed version. The company says the model improves on coding, 3D design and finance benchmarks, with full weights due Nov. 4, 2026.

• Anthropic Used Claude to Find Gaps in Its Training Monitor

  Anthropic tasked Claude with creating training data that could pass a safety monitor yet make a separate research model behave in misaligned ways. The company says the deliberately trained models were never deployed and that it plans to make future tests more realistic.

• Railway Urges Customers to Rotate Secrets After Possible Breach

  Railway says an unauthorized third party may have accessed encrypted environment variables and obtained the decrypted contents of some of them. The company recommended that all customers replace credentials stored on the platform while it investigated the incident.

• Mistral Large 4: Model Details, Variants and Deployment Options

  Hugging Face’s guide and Mistral’s model card describe Large 4’s architecture, capabilities, variants and ways to run it. The documents also set out the model’s license and report results on coding and cybersecurity evaluations.

• Character.AI Files Chapter 11 as Disney Offers $20 Million for Assets

  Character.AI filed for Chapter 11 bankruptcy on October 12, 2026, and named Disney as the initial bidder for a license to certain intellectual property and claims. The $20 million proposal is subject to higher offers, while Character.AI says its platform will keep running during the sale process.

• South Korea Orders Bank Audits as Police Probe AI-Suspected Hacks

  South Korean police opened a preliminary inquiry on October 2 into a series of bank data breaches, while the Financial Services Commission ordered financial firms to check exposed systems. Investigators have not confirmed that AI tools were used in the attacks, which exposed customer information at Shinhan, KB Kookmin and Hana banks.

• Microsoft Research Tests Agensh With 1,024 AI Agents

  A Microsoft Research paper reports that Agensh raised pandoc’s test-pass rate from 33.89% with one agent to 55.06% with 1,024 agents, while the mean result across five ProgramBench tasks rose from 19.31% to 28.78% with 128 agents.

• Fireworks AI Says Intruder Captured Secrets in Evaluation Jobs

  Fireworks AI says an unauthorized third party accessed environment variables in evaluation jobs across a limited number of customer accounts. The company has revoked identified keys and tokens, while asking affected customers to rotate third-party credentials with their issuing providers.

• LangChain Open-Sources Paid Media Agent for Ad Campaigns

  LangChain has open-sourced a paid media agent that connects advertising accounts, analyzes campaign performance, produces recurring reports, and prepares proposed changes for review.

• Ukraine Says AI Turrets Have Downed Russian Geran-5 Drones

  Ukraine’s Air Force says machine-vision gun turrets have shot down Russian jet-powered drones, including Geran-5s. President Volodymyr Zelenskyy said more than 12 turrets were already in Kyiv, with eight more planned for the surrounding region.

• Axiom Says Its Lean Formalization Covers the Entire Stacks Project

  Axiom says it has released a Lean formalization of the entire Stacks Project, an open-source reference for algebraic geometry spanning roughly 7,500 pages. The company says its formalization spans some 16 million lines of code and can be used by researchers and AI agents in new proofs.

• China’s AI Film Industry Grows as Cities Compete for Creators

  Chinese cities are offering AI filmmakers incentives as production costs fall and AI video output surges. Separately, a 90-minute science-fiction film has received a license for theatrical release in China.

• Aleph Alpha Finds Party-Line Responses in Chinese LLMs

  Aleph Alpha reports that six Chinese open-weight models gave balanced answers to 17% to 41% of politically sensitive prompts. It also says NVIDIA’s Nemotron Cascade 2 was flagged on 17% of prompts, and points to Chinese-model-generated examples in its fine-tuning data as a likely source of the behavior.

• Natura prices its AI-agent ring at $99, with a $9 monthly plan

  Natura introduced Interface, a $99 smart ring for sending voice requests to AI agents and recording thoughts. The company plans a $9 monthly subscription after an initial free period, with preorders expected in November and shipping targeted for December or January.

• Google Takes AI Search Ads Global as Search Growth Cools

  Google is expanding ads in AI Mode to international markets as its Search advertising growth slows. Alphabet reported $63.35 billion in Google Search and other revenue for the third quarter of 2026, up 12% year over year.