Sunday: funding and IPO talk around OpenAI and Anthropic remains the thread, alongside data-center pushback, new agent tools and a Bloomberg report that Apple CEO Ternus is pressing for faster A.I. progress.
OpenAI reportedly raises a further $11 billion at a $1.25 trillion valuation
High Business
Several outlets report that OpenAI has pulled in an extra tranche of about $11 billion, putting its valuation at $1.25 trillion, below the $1.4 trillion floated earlier. The company has not officially confirmed who is taking part.
Anthropic reportedly still eyes a mid-November listing, but conditions have turned tougher
Medium Business
According to Bright, citing CNBC sources, the recent selloff in AI shares is putting pressure on Anthropic's IPO and valuation hopes are lower than before. The listing date has not been officially confirmed.
After the US midterms, Democrats reportedly plan a wave of bills to rein in AI
Medium Policy
Bright reports that Democrats want to file many AI bills once the new year starts. VentureBeat analyzes what a Democratic win could mean for liability and oversight of AI companies. For now these are plans, and it is unclear whether majorities will materialize.
Apple's Ternus reportedly pushes for faster A.I. progress and a Siri that carries out tasks on its own
Medium Business Products
According to Bloomberg's Mark Gurman, citing people familiar, Apple CEO John Ternus is frustrated by the A.I. gap with rivals and is pressing software chief Craig Federighi to move faster, including turning Siri into an agent. Siri lead Mike Rockwell is also said to be weighing a departure next year. Apple has not confirmed any of this.
Nvidia completes its $1 billion investment in Korea's Naver, becoming a top shareholder
High Chips Business
Korean and Taiwanese outlets report the payment closed on November 6, leaving Nvidia with just over 4 percent of Naver. Naver plans to use the compute to build AI infrastructure from its Sejong data center.
OpenAI teams up with security firm Hacktron to hunt for weaknesses in its own services
Medium Safety Business
According to a daily AI newsletter, Hacktron gets broader access to frontier models through the Daybreak program, with a focus on integrations with third-party platforms. I could not yet check an OpenAI post with the details.
Microsoft releases APM, an open package manager for AI agents
Certain Products
APM uses a manifest file and a lockfile to pin an agent's dependencies, under an MIT license. According to the project it works with tools such as Claude Code, Copilot and Cursor, so teams can share the same agent setup.
Google ships Agent Development Kit 2.14.0 with stable agent protocols and evaluation
Certain Products
The new version of Google's open agent toolkit marks parts such as the A2A integration, evaluation and the core runtime as stable. New additions include multi-run evaluation and a runner for teams of agents.
Baidu unveils Apollo World, an AI world model for training robotaxis in simulation
High Research Products
The model recreates traffic scenes, including cyclists and animals, so Baidu's Apollo Go vehicles can practice virtually. Baidu says huge distances can be simulated daily; those figures come from the company itself.
Amazon lets shoppers text Alexa for help, after earlier blocking Meta's Muse
Medium Products
Users can now ask Alexa for shopping help by text message. TechCrunch ties it to Amazon's earlier block of Meta's Muse agent on its shopping site.
Bright: smart headband makers reportedly train AI on users' brainwaves without saying so
Medium Safety
According to Bright, makers of smart headbands use users' brainwave data to train AI without clearly disclosing it. I could only see the headline, so the details are unchecked.
Underdog shows off a speech model of just 32 MB in a demo
Medium Models Research
The young company shows speech running on an extremely small model, which brings on-device uses without a server connection closer. Performance has only been judged from the demo.
Physicl raises $10 million to build digital worlds for training robots
Medium Business Research
The startup builds simulated environments meant to be realistic enough to train robots in. The raise fits the growing focus on world models and physical AI.
Meta warns that election backlash against data centers could slow AI expansion
High Policy Chips
Meta says local political resistance to data centers could make building harder. That matches reports that voters in dozens of US localities are introducing limits and that data-center stocks fell after the election.
Gemini may soon alert parents when children talk about self-harm, app code suggests
Low Safety Products
References found in version 17.73.15 of the Google app point to a feature that notifies parents when a child discusses self-harm with Gemini; Bright adds that Google is also working on daily time limits for kids. Google has not announced it, and it may still change or be dropped.
Next up: Anthropic's possible IPO around mid-November; its tighter usage policy takes effect on November 12; Nvidia's Huang is due for a House briefing on November 18.