Deals
New turn in Google military AI deal as DeepMind staff resign
Researcher Andreas Kirsch says Google seems to have no right to veto how the Pentagon uses its AI.
Tag
240 stories carry this tag.
Security
Malware in LiteLLM 1.107.0 steals cloud and AI keys
Teams that operated LiteLLM 1.107.0 must think that attackers have their secrets, and the last safe version is 1.106.2.
Industry
Hassabis says AGI can come in two to three years
Earlier this year he gave three to five years, and he gives no test result for the new date.
the-decoder.comClaimed, not confirmed
Industry
Pichai reports that Gemini 4 Argon used 25 trillion tokens in 72 hours
Google does not have sufficient capacity for the need, and it makes efficient versions of the model to give access to more users.
Models
Google gives Gemini voice help for navigation to Apple CarPlay
Drivers in the US can use it today. Google says the release is global but gave no list of regions.
Industry
Eric Schmidt guesses AI can replace 2,000 of 10,000 employees
Schmidt gave no data, and a friend got death threats after the company cut 20% of its employees.
Agents
Cloudflare helps Commerce.com start WebMCP on 100,000 domains
WebMCP gives a shopping agent tools with set names, and the agent does not guess which button to click.
Compute
Google Cloud opens NVIDIA Vera Rubin A5X instances next month
Google Cloud wants to have as many as 960,000 Rubin GPUs at many sites.
Security
NVIDIA adds Confidential Bare Metal for AI model makers
Model makers can operate sensitive work on rented dedicated servers and keep control of their models and data.
blogs.nvidia.comClaimed, not confirmed
Security
Google opens one site that checks media for AI watermarks
The tool finds media only from SynthID partners, and a no does not show that a person made it.
Agents
Gemini 4 Argon cuts ADK session prep time by about 40% in one night
Google reports that its team rejected 4 of the 6 changes that Argon made, and 2 are in production.
Agents
Google adds a C++ Tensor API to LiteRT for on-device models
Google reports that a model an agent wrote with the API is about 1.5 times as fast as Apple MLX on a MacBook Pro.