This issue collects 22 AI updates from 2026-10-05, led by Volantis Raises $88M for Photonic AI Interconnect, Google reshuffles Gemini tiers: free users drop to Flash Lite as the $5 plan loses Pro, Bessent Says China's Kimi Passed Information to Anthropic and 'Helped a Lot', OpenAI CEO Sam Altman warns against treating AI as a religious authority.
Photonics company Volantis has raised $88 million to advance photonic interconnect technology for AI computing.The funding signals continued investor appetite for the physical layer of AI infrastructure, where optical links are seen as a way past the bandwidth and power limits of copper wiring.
Singapore's Singtel has launched an AI Token as a Service offering aimed at local enterprises, according to Retail News Asia.The product provides AI capability on a token metered basis, positioning the telecom operator as a distribution channel for enterprise grade AI.
Coforge has introduced an intent engineering framework that adds a governance layer on top of prompt and context engineering for enterprises deploying AI agents.The framework defines what agents should optimise for, how their performance is measured, and when they must hand decisions back to humans.
Safeworld, founded by Carnegie Mellon Safe AI lab director Ding Zhao, emerged from stealth on October 5 with a seed round of more than $12 million led by Shine Capital and a16z Speedrun.The company stress tests robotic control systems inside simulations populated with realistic human models to show that generative AI driven robots are safe before they deploy.
According to the decoder, from October 2026 Google will restrict which Gemini models personal account users can reach: without a subscription, users get only the smallest model, Flash Lite, while the stronger Flash and Pro are excluded.The $4.99 AI Plus plan also loses Pro, leaving all three models to AI Pro at $19.
OpenAI CEO Sam Altman has warned that giving AI religious force, or treating models as a religious authority, is a real safety issue, and said the idea makes him very uncomfortable.The remarks land as public debate over AI's role in faith and society intensifies.
OpenAI's head of safety systems, David Robinson, has left the company, according to Sina Finance, following a run of reported safety problems.Reports say he had publicly argued that the way AI companies currently develop their systems is unacceptable and called for nuclear plant level regulation of large models.
According to a Chejiahao post on Autohome, the Doubao large model has been integrated into Seres' new AIVA ME7, becoming part of the vehicle's smart cockpit capabilities.The move pushes large model deployment from general chat scenarios into production car cabins and keeps the race between automakers and model providers heating up.
智谱股价涨近5%，市值3206亿港元|Cursor|GLM 5|高盛|AI代码生成|Qwen 手机新浪网 新浪财经.智谱股价涨近5%，市值3206亿港元|Cursor|GLM 5|高盛|AI代码生成|Qwen 手机新浪网 新浪财经
A new report says the AI capability gap between the US and China has narrowed to a record low 3 percent, with DeepSeek named as a leading driver.The report also places DeepSeek's V4.1 Flash sixth in its September ranking.
OpenAI announced a new visual ad format in ChatGPT and expanded its measurement tools, attribution partnerships, and brand suitability options for advertisers.The update extends the company's advertising push inside its conversational AI product.
Chinese news aggregators reported that US Treasury Secretary Scott Bessent said China's Kimi passed Chinese information to Anthropic and that it 'helped a lot'.The remark ties Moonshot AI and Anthropic into a single, sensitive claim about US China AI technology flows.
OpenAI's Sam Altman has said publicly that the benefits of artificial intelligence justify accepting a portion of its risks, according to Business Standard.The remark puts the trade off between the pace of AI progress and safety back at the center of the debate.
Chosun Ilbo reports that seven of the 11 AI models already deployed on South Korea's government wide public sector AI platform are foreign made, and only 12 of the 31 introduced models are domestic.The list includes Bge M3, an embedding model from Beijing's BAAI, which the U.S.
Google has frozen its open source bug bounty program after a "significant rise" in AI generated submissions, TechCrunch reports.The move highlights how automated reports are straining bug bounty triage and raising new questions for open source security collaboration.
NASA and IBM have jointly open sourced a lunar AI model trained on 17 years of Moon data.The release turns long accumulated planetary observations into a reusable foundation model that researchers and developers can adapt for tasks like crater detection and terrain modeling.
EFutures is expanding into AI consulting and AI agent development, according to a report from Daily FT, extending its service business deeper into enterprise artificial intelligence.The move reflects how traditional technology service providers are repositioning around AI strategy work and the engineering of deployable AI agents.
OpenAI Codex lead Tibo Sottiaux said on X that for 28 days he will either ship a clearly useful improvement for most Codex and Work users or hand out a full quota reset, as reported by QbitAI.The pledge follows continued user complaints about crowded DevDay updates, unpredictable resets and mid task model disconnects.
Inferenz has launched Risk Analyzer, an AI agent that turns patient risk flags into a concrete next best action plan for care teams.Positioned above existing risk scoring platforms, it is designed to explain why a patient is at risk and route a prioritised intervention to the right clinical team within seconds.
Logistics Reply has introduced the LEA AI Agent Authority Model, bringing governed agentic AI into warehouse execution alongside its LEA Dynamic Intelligence capability.The offering centers on keeping AI agents' permissions and actions inside a controlled, auditable framework for logistics operations.
Snaplii has introduced a dedicated wallet for AI agent spending, pitching it as a way to keep agent transactions separate from a user's main account while offering cashback.The same release frames the launch against two industry data points, Muse topping the app store and Instinct raising $1 billion, as signs of continued momentum in the AI app and agent market.
According to Business Today, Anthropic's Claude models are now available with local inference in India via Amazon Bedrock.The move lets Indian customers run the models while keeping data within the country, a key consideration for regulated industries and cross border compliance.
