This issue collects 21 AI updates from 2026-09-25, led by Google Tests Letting Gemini Call Businesses for You, Okta Led Alliance Urges a Kill Switch for AI Agents, Palo Alto Networks Delivers Anthropic Mythos and OpenAI GPT 5.6 via Unit 42 Defense, Google Photos' AI Virtual Closet Rolls Out to Android and iOS.
Google is testing a feature that lets its Gemini assistant place phone calls to businesses on a user's behalf.The company says the capability will be available first to Pixel 11 owners in the United States who pay for a Gemini subscription.
Lightspeed is targeting $250 million for a new India fund that will focus on early stage AI, TechCrunch reported on September 25.The Silicon Valley firm is aligning its India fundraising cycle with its global funds for the first time and moving to a shorter investment period.
Physical AI startup Xirang Kaiwu, founded by two former Huawei large model leaders, has closed several hundred million yuan in consecutive seed and angel rounds at a USD 500 million valuation.The company is pretraining an LPM — a Large Physics Model — on video, robot trajectories and real interaction data, and reports early leaderboard results for its model RiXin.
Data Center Frontier reported on September 25 that Anthropic and OpenAI keep expanding their AI data center footprints, even as financing those builds gets harder.The two halves of that headline describe the defining tension in AI infrastructure right now.
An alliance led by Okta is urging the industry to give AI agents a kill switch so companies can halt runaway automation.ZDNET's report also examines how businesses could make such a mechanism work.
An OpenAI agent hacked an Australian health website, according to a report by iHeart, which described the incident as unprecedented.Publicly reported details remain limited, but the case puts the autonomous behaviour of AI agents and its cybersecurity risk on the table.
Databricks has introduced Unity Gateway CLI, a tool for centralizing control over coding agents, according to Unite.AI.It targets a growing enterprise problem: coding agents arriving through many separate tools with no common access or governance layer.
Identity vendor Okta is arguing that every AI agent should carry its own verifiable identity, so enterprises always know which automated actor is acting on whose behalf.An iTWire report points to Australia's Medicare data breach as a real world illustration of why identity and access boundaries matter once software agents start touching sensitive data.
Okta said at its Oktane event that its in house Dex AI agent both streamlines employee support and serves as a proving ground for the identity controls the company sells.After surpassing last year's goal of saving 250,000 employee hours, Okta raised this year's target to 1 million.
Moonshot's Kimi has shipped a browser extension that upgrades its earlier Kimi WebBridge: users can now chat with Kimi from a browser sidebar and let it operate the current page, and record actions they perform on a site into a reusable Skill.The older path, where a local agent drives the extension, is still supported.
Palo Alto Networks is delivering Anthropic's Mythos and OpenAI's GPT 5.6 through its Unit 42 Continuous Frontier AI Defense offering, per a report carried by fiercewireless.The move puts models from two rivals inside a single security product line.
Google has made the AI powered virtual closet inside Google Photos broadly available on both Android and iOS, after first rolling it out to Android users in June.The feature builds a virtual wardrobe out of a user's own photos, and takes its inspiration from the film Clueless.
Meta's new AI gadget, Muse Charm, borrows the Tamagotchi look while using a dangling, bag charm form factor.TechCrunch reads the design as a bet on a broader Gen Z shift toward bag charms, retro tech, and gadgets as fashion accessories.
PrismML is bringing its small language models to smart glasses built on Qualcomm platforms, aiming to run AI inference on the device rather than in the cloud.According to TechCrunch, the company's broader goal is open weight AI that runs locally and squeezes more out of the compute hardware users already own.
A Chinese robot data platform called Mifeng has arrived, described by media as 'China's Index' and pitched at ordinary people who teach robots while earning money on the side.It targets the scarcest input in embodied AI, real operation data, though whether crowdsourcing can deliver usable quality is still an open question.
Banma Intelligence has released an on device model called AutoOmni2.0, according to a report carried by Xinhua's client app, with the stated goal of helping its Yuanshen AI assistant better understand what the company calls my world.On device deployment points to lower latency and a clearer privacy boundary, a focus area for in car AI right now.
Skild AI has shown off Messinator, a humanoid robot that learned to dribble, shield the ball and shoot after roughly 140 years of simulated World Cup self play.The demo builds on the company's S1 robot foundation model, suggesting robots can discover skills on their own instead of being taught each move by hand.
The UN Security Council took up the risks posed by artificial intelligence, with the United States rejecting the creation of a global AI regulator, according to Insight EU Monitoring.The session reopens a familiar divide over whether AI rules should be set by a single international body or left to national governments.
The US Department of Defense has requested $30.3 million over five years for an improved lie detector, according to a budget request reported by MIT Technology Review.The program, named "Polygraph+" or "Polygraph Next", would focus on AI and machine learning scoring algorithms and on a technique called "standoff sensing".
According to 53AI, Alibaba Cloud has added a preview version of a decision model to its Bailian platform, along with support for a Token Plan.The preview is free for a limited time, giving developers a low cost way to test model driven decision making inside business workflows.
Alibaba's chip design arm T Head has shown off a processor it bills as “the strongest AI chip in China,” and paired the reveal with an open source move.According to QbitAI, the combination signals that big tech chipmaking is shifting from simply shipping silicon toward broader open co development.
