All AI updates

Amazon says it will stop using NDAs when negotiating data center deals with local governments, following a similar move by Microsoft earlier this year. The report says secrecy around AI infrastructure has fueled community backlash, with opposition leading to hundreds of proposed and enacted moratoriums from New York to San Francisco.

Instinct launched its invite-only AI agent in August. Users can text it to handle chores such as booking DMV appointments and sending follow-up emails. Muse and Dots arrived afterward; the report describes them as broadly similar products from more powerful companies.

TeleTune infers each session’s goal from usage logs and uses a text skill library to predict users’ next actions. It keeps skill edits only when they improve accuracy on held-out logs. The source says next-action accuracy on old logs tracked success in a live environment, but gives no experimental figures.

A blogger used voice conversations in the Codex tab of the ChatGPT desktop app while cooking dinner to build a newsletter archive for his blog. Voice work covered most of the imports, archives and search integration; before deployment, he spent another half hour using typed prompts to revise an import and review the code.

The Financial Times reports that HSBC plans to cut about 70% of its UK financial advisers and shift wealth clients toward an AI-backed digital service. People familiar with the plans say about half of the unit’s management and specialist roles would also go.

Doubao Work’s task mode now has an unlimited canvas for viewing and comparing materials, plans and results in one place, with options to adjust text, colors and layout after generation. Doubao 2.1 Lite is available for tasks including document writing, spreadsheet work and PPT creation; image creation also supports Seedream 5.0 Flash.

Arthur Holland Michel argues that AI companies use model training and additional classifiers to block harmful requests, but those safeguards can be bypassed and can reject harmless questions. The article cites a test in which five widely used models were less willing to produce a pamphlet criticizing Thailand’s king than one criticizing Britain’s king.

In a new Q&A, Huawei said Qiankun ADS does not immediately disengage during encounters on narrow roads and can reverse to yield when space is insufficient. It also said CAS 5.0 can detect and respond to risks while a person is driving without ADS enabled. The system remains driver assistance and cannot replace the driver.

Anthropic has announced two Claude features in testing. Users can connect a company data platform and ask in natural language for an automatically updating dashboard, or turn text, charts and other materials into a short animation. Dashboards are available to all paid-plan users, while Motion is limited to Team and Enterprise users.

Google Cloud announced Gemini agent as a single interface for questions, knowledge work, media creation and coding. The report says it currently routes tasks across Gemini and Anthropic Claude models. Teams can set a project spending cap; when reached, the agent pauses until someone resumes it.

The account says the attacker's Claude Code session histories, Claude memory files and ARTEX configurations were exposed in open directories. Officials believe an open-source Chinese agent helped breach at least seven South Korean financial institutions. It says ARTEX ran on DeepSeek v4.1-flash, but describes its API access and the host's role in the attacks as uncertain.

Odyssey-3 predicts how a generated environment changes as users move the camera, take actions or trigger events. The preview supports first-person and third-person navigation as well as independent camera movement. According to Odyssey’s announcement, the Pro version scored 66.1 on the Physics-IQ Verified video-to-video benchmark, a record for that leaderboard.

NVIDIA announced plans to spend $1 billion over five years to advance science in the US. Its stated areas of support include US higher education research institutions, quantum computing, and cloud service providers serving US government missions.

In the study, 98 patients used the research-grade diagnostic AI chatbot AMIE before their visits. Supervising doctors monitored every interaction in real time, and none had to be stopped under preset safety criteria. Clinicians said its summaries helped them prepare in 75% of cases. Larger clinical trials are still needed to assess its performance for patients.

Alexandr Wang says each Muse runs in its own secure virtual machine, where data shared with it is stored. A separate agent checks attempts to send information out and blocks sensitive data such as credit card numbers; users must also approve each new website Muse wants to contact. A more encrypted virtual machine is planned for later and may be slower.

Alexandr Wang says each Muse runs in its own secure virtual machine, where the data shared with it is stored. A separate agent checks outgoing information and blocks sensitive data such as credit card numbers; Muse also asks users to approve each new website it wants to contact or send data to.

China’s intellectual property authority published an embodied AI robot design patent held by BYD on October 9. The patent shows the appearance of a humanoid robot intended for walking, balancing and interaction, with potential uses in service, education and industrial settings. The report concerns a design patent, not a product launch.

Meta has begun rejecting ByteDance ads and ads from any advertiser linking to TikTok in seven countries. The restriction covers ads and paid marketing messages on Meta’s apps, including third-party campaigns. Meta says ByteDance is a competitor trying to draw users away from Facebook and Instagram.

AI model evaluation platform Arena announced a $200 million Series B round at a $3.1 billion valuation, up from its $1.7 billion post-Series A valuation in January. Individuals can use its free platform to compare and rate models’ results, while its commercial service uses community feedback to analyze model performance.

Bloomberg reports that Oracle is trucking natural gas to meet data center energy needs before supporting pipelines are built, aiming to bring computing capacity online sooner. The approach has been used at a Utah data center for more than a year and is being introduced at the Shackelford data center in Texas, which Oracle is developing for OpenAI. The reported end-to-end cost is about four times that of delivery through a major pipeline hub.