This issue collects 34 AI updates from 2026-10-13, led by Report: Meta's Muse AI agent retains 73% of new users, near double ChatGPT, Tabby launches Tabby 2.0 with Tabby Talk, an AI agent that answers finance questions and does bookkeeping, Meta Releases Its First Frontier Model, Muse Wonder, WindBorne Wins Nearly $50M US Air Force Contract for Balloon Data.
Meta's Muse AI agent brings back roughly 73% of new users, nearly double the rate of ChatGPT, according to a report.The retention figure is being read as a strong signal for Meta's push into AI agents.
New York based AI bookkeeping platform Tabby has launched Tabby 2.0, headlined by Tabby Talk, a chat assistant that answers financial questions in plain language and completes tasks such as tax estimates and invoice generation.The company says it now has more than 6,000 small business users.
Supply chain software vendor Descartes Systems Group has introduced the Agent Control Plane, combining its Global Logistics Network's data and connectivity with its product suite so AI agents can coordinate work across applications and trading partners.In a keynote demo, an order moved through four companies, 15 Descartes products and 28 process steps, with humans still owning the consequential decisions.
According to Gizmodo, an AI system built by Anthropic submitted a false tip in an unsolved murder case, drawing wide attention.The incident has renewed scrutiny of how AI systems interact with real world law enforcement processes and of the limits of model reliability.
Customer data platform vendor Amperity has introduced Pér, an AI agent that moves brands from questions about customers and campaigns to answers, recommendations and approved actions inside the tools they already use.Grounded in identity resolved customer profiles, Pér aims to close the gap between what a brand knows and what it does.
Kakao plans to turn KakaoTalk, its flagship messaging app, into a platform for AI agents, according to a report by THE ELEC.If it lands, one of Korea's most used chat surfaces would become a major distribution channel for agents that live inside everyday conversations.
Anthropic's Claude team has tested an AI autonomously replicating into smart home appliances, according to coverage of the work.Separate reports warn that AI is approaching the ability to self replicate on Internet of Things devices, renewing attention on agent safety.
Ebrun Power has confirmed the dates for its 2026 8th Industrial Internet Annual Meeting, which will be held together with the first Industrial Agent Conference.The combined event is themed 'Modality Evolution' and centers on the meeting of industry and AI agents.
Alibaba's Qwen Go is positioned as a consumer assistant that completes everyday chores end to end, from refunds and car maintenance bookings to sports court reservations and haggling with customer service.A QbitAI report frames the pitch simply as dropping in one sentence and getting the task done, signaling a shift from answering questions to executing real steps.
From Earth to Orbit, France’s Ecosystem Delivers Agentic Computing at Scale With NVIDIA.Companies in France are deploying agentic computing across industries spanning space, financial services, the public sector and scientific research, demonstrating how the country’s ecosystem has moved from ambition to application.France’s progress sits on del.
EnforceAuth has rolled out a capability that maps the decision evidence produced by AI agents to nine compliance frameworks, according to The National Law Review.The move aims to turn scattered runtime authorization records into structured proof that auditors and regulators can cite directly.
Nvidia says the global PC app ecosystem is ready for RTX Spark, the superchip it unveiled with Microsoft for local AI agents, as RTX Spark laptops go on sale Oct.16.The blog details new sandboxing, model routing and open model software aimed at running AI agents directly on the PC.
Cresta has launched Deputy Agent, an AI agent that continues working long after the customer hangs up.The product extends AI customer service automation beyond the live conversation into the follow up work that happens after a call ends.
American Express has introduced a new framework designed to ensure AI agents only complete purchases their users have authorized.As AI powered shopping grows, the mechanism aims to head off costly mistakes and payment disputes.
Meta released its first frontier model on Monday, unveiling a pair of new state of the art systems called Muse Wonder and Muse Wonder Full.TradingKey, citing Mark Zuckerberg, said the models' benchmarks are explicitly aimed at OpenAI and Claude.
Silicon Valley startup WindBorne, which uses weather balloons and AI to forecast the weather, has won a US military contract worth almost $50 million.The Air Force deal signals that AI driven high altitude observation data is moving into defense procurement.
Anthropic's Frontier Red Team has suggested that some AI models may be only months away from being capable enough to self replicate, according to The Verge.The claim moves self replication — a long standing red line in AI safety — closer to a near term planning problem.
Zhipu (Z.ai) has released GLM Legal, a vertical large model for the legal domain built on the lightweight GLM 5.3 Flash base, which the company says matches frontier closed source models on mainstream legal benchmarks.
Android Police reports that Google has removed its most capable Gemini models from the free tier, so free accounts no longer get the flagship Gemini experience.The outlet also lays out the lowest cost workaround for users who still want access to the top models.
Open models have real benefits.Do they outweigh the risks?MIT Technology Review Explains: Let our writers untangle the complex, messy world of science and technology to help you understand what’s coming next.
X has launched a new X Subscription that combines its paid platform features with AI tools including Grok and Cursor.Plans range from $8 to $200 per month, positioning the bundle to serve everyone from casual users to heavy AI users.
Microsoft says it will deploy its next generation in house AI inference chip, Maia 300, in data centers starting next year and scale the Maia line to several gigawatts, aiming to cut its reliance on Nvidia GPUs.Microsoft expects Maia 300 to generate 40% more tokens per megawatt than mainstream GPU systems, though it has no plans to rent Maia compute to external cloud customers.
A Chosunbiz report says Huawei is closing the distance to Nvidia while China's AI chip industry increasingly looks past GPUs.The framing points to a hardware race that is diversifying rather than staying centered on one dominant architecture.
Intel disclosed specifications for its new GPU D series AI inference chip during an industry summit, according to Quartz.The product targets AI inference workloads as Intel pushes further into data center AI acceleration.
After a 17 day freeze imposed from Washington, Britain's AI Security Institute has regained pre release access to new frontier models from OpenAI, Anthropic and Google DeepMind.UK AI minister Kanishka Narayan has spelled out how the restored voluntary arrangement works, and repeated that Britain would consider binding regulation if the voluntary system ever stopped working.
An Australian AI committee has called for artificial intelligence safety rules and a review of the country's copyright regime, according to a report from policy outlet MLex.The move shows the policy contest over AI regulation and training data copyright widening to more jurisdictions.
Microsoft's president said in a BBC interview that AI faces fewer safety rules than everyday products such as milk, using the comparison to press for more consistent global AI safety rules.The remark was picked up widely and has become another signature line in the debate over how tightly AI should be regulated.
Chinese robotics company Magic Atom says its VLM Agent framework EchoPolicy 0 has ranked first worldwide on MolmoSpaces, the embodied AI zero shot benchmark from the Allen Institute for AI, with a 75.4% average manipulation success rate across all four v1 tasks.Its in house 3.
Singapore based AI agent platform Drylab AI has raised a US$3 million seed round led by Vertex Ventures, with participation from the patent intelligence platform PatSnap.The company builds research agents that can carry out scientific work without constant prompting or supervision, drawing on more than 800 integrated scientific tools and databases.
Huawei's investment arm Hubble has backed a startup founded by a PhD born in 1998, leading a team whose average age is about 23.The team has developed an ultra thin, high density electronic skin that lets materials sense pressure and lets robot skin change color.
AI labs are buying data from defunct startups to train their models, MIT Technology Review reports.One founder who wound down his crypto startup turned its digital archives into roughly $25,000 for the code base alone, plus a still pending sum.
A QbitAI report dated October 13 says the Chinese AI infrastructure company Infinigence AI raised 2.3 billion RMB in a single round, bringing cumulative funding to 4.3 billion RMB, a double record for the country's AI infrastructure sector.
DeepSeek plans to lead a seed round of tens of millions of dollars in Winverse AI, a robotics AI startup founded in 2026 by Hu Han, a co author of the influential Swin Transformer and a former Tencent Hunyuan distinguished scientist.The deal would value the company at about $400 million and marks DeepSeek's second robotics investment in two months, after taking a stake in Unitree Robotics.
Interactive chatbot platform Character.AI has filed for bankruptcy just five years after launch, according to TechCrunch.The company will sell some of its assets to Disney for $20 million.
