Today's briefs
Google Launches Gemini Decisions API for Fast Multimodal Decision-Making
Google has launched the Gemini Decisions API, featuring the Gemini Decision 1 model priced at $0.10 per million tokens, designed specifically for high-throughput, low-latency decision tasks in production AI pipelines. The API is built for scenarios where agents or systems need to make rapid, calibrated choices across multimodal inputs — such as routing, classification, or conditional branching — without the overhead of a full generative model call. For developers building agentic systems, this is a purpose-built inference endpoint that could substantially reduce cost and latency for decision nodes in multi-step pipelines. The aggressive pricing positions it as a practical alternative to running larger models for lightweight but frequent decision tasks. Developers should evaluate it immediately as a building block for autonomous agent orchestration, real-time filtering, and conditional workflow logic.
Google DeepMind
Google DeepMind Formally Verifies Key Parts of Signal's Encryption Protocol
Google DeepMind researchers have completed a formal mathematical verification of critical components of Signal's encryption protocol, using AI-assisted theorem proving to confirm the cryptographic guarantees that billions of users rely on. This represents a meaningful advance in applied formal verification, demonstrating that AI tools can now be productively applied to real-world security-critical codebases at scale. For security engineers and developers building privacy-sensitive systems, this sets a new precedent for using AI in the assurance process — not just code generation, but rigorous correctness proofs. The verification adds an independent layer of confidence to Signal's protocol beyond traditional peer review and audit. It also signals that AI-powered formal methods are maturing to the point where they can tackle production cryptographic infrastructure.
Google DeepMind
TSMC to Invest $1B in Both Anthropic and OpenAI
Taiwan's TSMC, the world's dominant advanced chip manufacturer, is set to invest $1 billion spread across Anthropic and OpenAI, marking an extraordinary move by a hardware supplier to take direct equity positions in frontier AI labs. This signals that chipmakers are no longer content to be passive suppliers — TSMC is positioning itself as a strategic stakeholder in the success of the leading AI model companies it already serves. For developers and engineers, this deepens the already tight integration between semiconductor supply chains and AI model availability, suggesting that compute access and model development will become even more intertwined. The investment likely accelerates both companies' ability to secure preferential access to leading-edge process nodes. It also raises structural questions about independence and prioritization in the chip supply ecosystem as AI demand continues to outpace production.
Ars Technica
MBZUAI's IFM Releases PAN 2, an Open 29B World Model for Physical AI
The Institute for Foundational Models (IFM) at MBZUAI has released PAN 2, an open 29-billion-parameter world model specifically designed for Physical AI — systems that must reason about and interact with the physical world. PAN 2 is intended to serve as a foundation for robotics, embodied agents, and other physical AI applications, providing a large-scale open alternative to proprietary world models being developed by companies like Google and Tesla. For developers building robotics or embodied AI systems, the open release means they can fine-tune, adapt, and deploy a frontier-scale world model without licensing constraints. The 29B parameter scale places it in a competitive range with other leading open models, and its physical-world focus is a meaningful architectural distinction. This is a significant addition to the open physical AI ecosystem and worth evaluating for simulation, planning, and control tasks.
MarkTechPost
AWS Adds Direct Web Search API for Custom Agents and Data Pipelines
Amazon Web Services has introduced a direct web search API designed specifically for integration into custom AI agents and data ingestion pipelines, giving developers a first-party AWS-native way to ground agent responses in live web content. Previously, developers building on AWS had to route through third-party search APIs or build their own retrieval infrastructure to give agents access to real-time information. The new API integrates natively with AWS's agent and pipeline tooling, reducing architectural complexity for teams already invested in the AWS ecosystem. For developers building RAG pipelines, research agents, or any system requiring up-to-date knowledge retrieval, this is a meaningful reduction in integration overhead. It also positions AWS more competitively against Microsoft Azure's Bing-grounded Copilot stack and Google's native search-augmented agent capabilities.
Amazon
Adobe's AI Experiments Enable 3D Photo Navigation and Character Interviews
Adobe revealed a set of experimental AI features at Adobe MAX 2026 Sneaks, including a tool that lets users walk around inside photographs by inferring 3D scene geometry from a single image, and another that allows users to conduct interactive text-based interviews with characters depicted in creative works. These experiments push the boundary of what static visual assets can become — transforming photos and illustrated characters into interactive, navigable experiences using generative AI. For developers working on creative tooling, AR/VR content pipelines, or interactive media, these capabilities hint at an emerging class of AI-powered asset transformation that could reshape content production workflows. Adobe has not announced release timelines, but Sneaks historically previews features that eventually ship in Creative Cloud products. Developers building on Adobe APIs or integrating creative AI tools should watch these experiments closely as signals of where the platform is heading.
The Verge
United Airlines Blocks Unauthorized AI Agents from Booking Flights
United Airlines has taken a formal stance against AI agents attempting to book flights on behalf of users without explicit airline authorization, blocking automated agent traffic that doesn't comply with its terms of service. This is an early but significant indicator of how incumbent service providers are beginning to actively resist the agentic web — where AI assistants autonomously transact on users' behalf without direct human interaction at each step. For developers building travel agents, booking assistants, or any AI system that interacts with third-party services on a user's behalf, this signals a coming wave of API access restrictions and anti-automation enforcement. It raises important questions about the architecture of agent-to-service trust, credentialing, and authorization standards that the industry has yet to solve. Developers should monitor how airlines and other service industries respond to agent traffic, as this may necessitate formal agent identity protocols or OAuth-style agent authorization frameworks.
The Verge
Nvidia CEO Jensen Huang Departure Rumor Surfaces and Is Disputed
Reports circulated that Nvidia's board was preparing to remove Jensen Huang as CEO, a claim that generated significant industry attention given Huang's centrality to Nvidia's AI-driven rise. The rumor was quickly disputed and appears to lack substantive sourcing, with no corroboration from credible insider accounts. Jensen Huang has been the architect of Nvidia's transformation from a gaming GPU company into the dominant AI infrastructure provider, making any leadership change consequential for the entire AI hardware ecosystem. For developers and engineers dependent on Nvidia's roadmap for GPU compute, training infrastructure, and CUDA tooling, leadership continuity at Nvidia directly affects long-term platform confidence. At this stage the reports appear unverified, but the story underscores how sensitive the market and developer community are to any signal of instability at the company most central to AI compute.
NVIDIA