This issue collects 42 AI updates from 2026-10-14, led by Google Launches Flagship Gemini 4 Argon for Developers, Businesses and Consumers, Satlyt, Founded by Ex Google and SpaceX Product Manager, Raises 8M Dollars to Run AI on Satellites, OpenAI Told UK MPs It Is Not Racing to the Bottom on AI Safety, Liquid AI releases LFM2.5 24B A2B, a 24B reasoning model with just 2B active parameters.
Google introduced Gemini 4 Argon, its latest frontier model, giving developers, enterprises, scientists and consumers access to it.CNET reports the model is built for coding, math and robotics, and says users can try it right now.
Satlyt, founded by a former Google and SpaceX product manager, has raised 8 million dollars to run AI on satellites.It positions itself as the Android of orbital computing, offering open software that runs across many operators' satellites instead of the closed, all in one route TechCrunch compares to an iPhone.
Twice as many companies use OpenAI's agent orchestration platform as Anthropic's, according to VentureBeat Intelligence's September survey, with 43% of users on OpenAI, 22% on Anthropic and 21% on Google.The same research found 62% of companies replaced a model inside a production app in the past year, and 59% of those had to rework or revert.
Enterprise security startup Pluto Security has raised $42M, betting that broad internal use of AI tools is creating a new class of risk.The report frames the core problem bluntly: when everyone in a company gets access to AI, people will use it in unexpected ways.
Facebook is giving users more control over their algorithm.Facebook is the latest social media platform to give users greater control over the algorithm that determines what appears in their feeds.Meta announced the launch of its new AI powered feature, “Your Algorithm,” on Tuesday, a feature it first launched on Ins.
Industry outlet Heavy Duty Trucking reports that Samsara has launched a Custom AI Agent Builder.The tool lets companies assemble AI agents around their own operations, making it the latest example of vertical software embedding agentic AI directly into the product.
A federal judge has ruled that Apple can access OpenAI's interview materials involving former Apple employees, along with private messages and testimony from key figures in the case.The order pushes OpenAI's recruiting practices into the evidentiary record of Apple's trade secrets litigation.
Brexy has launched an AI Agent Studio aimed at capital markets, according to a report by The National Law Review.The launch puts the company's agent building tooling into one of the most demanding and heavily regulated environments for automation.
ServiceNow has denied reports that it held takeover interest in the AI agent startup Decagon, according to PYMNTS.com.The denial turns a reported deal rumour into an on the record corporate response and refocuses attention on consolidation pressure across the agent sector.
OutSystems has launched a product called Agent Experience that aims to bring governed AI development to any coding agent, according to FutureCIO.The offering places governance above existing agents instead of building a single in house coding assistant.
Netzilo says its AI Edge product now extends AI agent governance from endpoints into SaaS, and has introduced a Drift Control capability built on System One Models.The update targets enterprises that need consistent, auditable oversight of agents across more than one environment.
Channel NewsAsia reports that Anthropic has opened an office in Singapore and signed an AI agreement there.The move extends the AI lab's push into Southeast Asia, where clearer regulation and high enterprise digitisation make the city state a common landing point.
Scimplify has raised $25 million to attack China's grip on critical materials using AI, according to TechCrunch.The company's annualized revenue run rate has tripled to $200 million as it eyes U.S.
Harbor has launched with $10 million in funding, positioning itself around the risks created by AI agents, according to a Program Business report.The report gives the funding figure and the company's focus but does not detail its investors, product or technical approach.
OpenAI told UK MPs that it is not racing to the bottom on AI safety, according to The Independent.The remark lands in the middle of a wider debate over whether competitive pressure pushes frontier labs to trade safety for speed.
At an AI vertical model event in Jinan on Oct.14, Shandong released seven industry specific large models — Hairuo, Tianzhi, Xingji, Yunding, Xinggang, Shangzhi and Qizhi Kongming — spanning high end equipment, chemicals, light industry, metallurgy, mining and port logistics.Provincial officials said 389 models have passed third party evaluation after 100 priority vertical models were selected for promotion earlier this year.
The European Union has announced its first enforcement actions under the AI Act, according to Unite.AI, with the cases triggered by incidents involving AI agents.It marks the bloc's first public step from rule writing to actual penalties under its flagship AI law.
Liquid AI published LFM2.5 24B A2B on its Hugging Face blog on October 13, a reasoning model with 24 billion total parameters but only 2 billion active per token, designed to run fast, private agents on device.The company says it thinks before it answers, calls tools, and works through multi step tasks inside popular agent harnesses.
XRPL has announced a partnership with Privy aimed at powering AI agent payments and embedded wallets.The move pairs a blockchain ledger’s settlement layer with wallet infrastructure designed to live inside applications, targeting a future in which autonomous agents transact on their own.
DIGITIMES reports that LINE, now 15 years in Taiwan, has unveiled a new AI strategy that weaves AI agents through scenarios from chat to shopping.The move extends agents from a single feature into a capability running across its core services.
SenseTime's large scale computing platform has supported the first batch of AI agents deployed in Shanghai's Quantum City initiative, with scheduling efficiency improving by roughly 80%, according to QbitAI.The report frames the shift as agents moving from understanding space to actually getting work done.
APUS AI (Qilin Hesheng Network Technology) announced on October 13 that it has open sourced Reflex Agent, its Personal Agent product.Built around the company's own open decision model Reflex and paired with the Continuum memory engine, the agent is said in lab tests to respond up to three times faster than mainstream agents such as Claude Code and Codex.
Cities across the EMEA region are working with local software and integration partners to build physical AI agents and digital twins on NVIDIA accelerated computing, using them to anticipate maintenance needs, guide planning and make public services more responsive.NVIDIA says the projects span Belgium, Finland, France, Italy, the Netherlands, Poland, Sweden and the UK, with partners including Akila, Galene AI, K2K and ThinkDeep.
OPPO and Manus have announced a new AI agent for the Find X10 series, according to Gizchina.The move shows agentic AI sinking further into the phone itself, with vendors treating agents as part of flagship differentiation.
Pi 1.0, a coding agent, has been released as open source, with coverage claiming it cuts token consumption by around 40%.Because token usage directly drives the cost of running coding agents, that figure has become the release's headline selling point.
GoDaddy and partners have demonstrated an AI agent transaction framework in Singapore, according to StreetInsider.The demo centers on letting automated agents take part in commercial transactions, though the report does not name the partners or detail the underlying technology.
Sparkling has launched a web app that puts a full trading terminal and an AI agent directly in the browser.The move continues the shift of trading tools toward the web and pairs them with agentic AI capabilities.
NeuBird has launched the Agentic Reliability Center (ARC), a product aimed at governing every AI agent that touches production.The pitch positions ARC as a control layer for the reliability and oversight of agents once they leave the lab.
Google announced on Tuesday, October 13, that users can run their own vibe coded apps on Android phones without first publishing them to the Play Store, keeping them private and out of search.The feature starts in the U.S.
Anthropic has warned that AI models may be only months away from being able to self replicate on smart devices, according to The Guardian.The report frames the risk through everyday connected appliances, asking how safety boundaries hold once a model can spread itself.
Chinese large model company StepFun has released its first phone, with "AI gets things done" as its headline selling point, according to a Sina report.The move takes the model maker from software into its own hardware, aiming to put AI assistance directly at the device entry point.
The F.B.I.Watch real task performance, API pricing, context capability, and migration cost.
According to a report by ContentGrip, Gemini Argon has added a paid testing surface for AI Mode, letting users try the mode under a paid tier.The change means verification of AI Mode is no longer limited to free or internal channels, but now includes a paid dimension.
On October 14, PHANES AI, the robotics data company founded by Harbin Institute of Technology (Shenzhen) professor Yang Shuo, and Tianlong Mobile released SeePhys Hub, a whole body multimodal data capture system.The rig combines a four camera headband, wrist bands, waist and calf trackers and an optional haptic glove to recover full body pose and hand contact pressure that first person video alone misses.
Chinese outlet 全天候科技 reports that ByteDance has established a dedicated safety department within its Seed large model team, moving pre release risk assessment to the forefront.The change points to safety review becoming a distinct function rather than a background step.
NanoPrompts has launched a library of ready to use prompts for Google Gemini, according to StreetInsider, aiming to remove trial and error from working with the model.It is a tooling layer bet: as base models converge, curated prompts and workflows become the differentiator.
HPCwire reports that Danaher is collaborating with NVIDIA to advance Autonomous Science, an effort aimed at making scientific workflows more automated and AI driven.The pairing links Danaher's scientific instrumentation business with NVIDIA's accelerated computing and AI platform.
Flox Intelligence is building AI to stop wildlife from colliding with critical infrastructure.Its edge devices identify animals and play species specific sounds to drive them away from trains, airports, and roads.
NVIDIA says it will commit $250 million of compute over five years to strengthen the UK's agentic safety and security research, delivering $50 million in accelerated computing access annually through BOLD at Oxford and SOFAIR at UCL and Imperial College London.It is also joining the Earthshot Prize as a Global Alliance Founding Partner, alongside UK deployments in healthcare, finance, energy and local government.
AMD has added a Model Router component to its Helion blueprint for the Open Compute Project, according to Tech Times.The reported result is that operators can scale serving capacity by adding CPUs instead of buying more GPUs.
An open source project called PS5LM has turned a jailbroken PlayStation 5 into a working local AI server, with no cloud bill and no gaming PC required.It ports llama.cpp onto the console and runs Alibaba's Qwen3.
An Anthropic executive has publicly warned that open source AI models carry more safety risk, according to Quartz.The statement puts the open weight safety debate back in the spotlight and raises the question of whether regulators should treat open and closed models differently.