The pilot attacks a persistent evaluation trade-off: labs do not want to reveal frontier-model internals, while evaluators do not want benchmark prompts leaking back to the model provider. DeepMind says a Singapore AI Safety Institute pilot kept both sides’ sensitive assets hidden during execution.
Vercel KMS gives Functions OIDC-authenticated access to managed RSA, ECDSA and EdDSA signing keys. Builders can scope grants by project and environment, constrain JWT claims with JSON Schema, rotate keys centrally and publish standard OIDC/JWKS metadata for verification outside Vercel.
GLiNER2.5-Decide attacks the same bounded-decision layer as Jev and CLM from a much smaller encoder architecture. Its strongest benchmark claims are vendor-produced, but CPU deployment and constrained joint decoding make it a materially different option for software-facing AI decisions.
The architecture matters as much as the voice quality: developers can replace a chained speech-to-text → LLM → text-to-speech loop with one full-duplex conversational model while keeping their own choice of backend reasoning model, tools and agent harness.
Playground’s legacy-runtime work turns the browser sandbox into a version-spanning compatibility lab. Maintainers can inspect old WordPress behavior without keeping obsolete PHP stacks alive, although the browser runtime is not a faithful recreation of every historical host.
Android Bench 2.0 moves coding-agent evaluation away from small repository fixes toward dependency upgrades, app builds, migrations and other jobs that can take a human engineer days. The results expose a much larger reliability gap than short-task benchmarks—and show that the agent harness can materially change cost and outcome.
The study moves the AI-search traffic debate beyond observational correlations: participants were randomly assigned to current Google Search, a version with AI features hidden, or AI Mode-only search during ordinary browsing. It is still a preprint and does not establish effects for every query or publisher.
Amplitude’s new consent gating separates the visible experiment experience from persistence and measurement. Sites can avoid experiment flicker before a consent decision, but denied users’ buffered impressions are discarded and Analytics consent still has to be handled separately.
Google Ads' new channel-priority controls offer a way to test Search partners and Display without giving them unrestricted budget, but the limits are not guarantees and Demand Gen has no allocation caps.
Bounded decision models are turning into a real model category. Cloudflare's entry is open-weight, multimodal and Jev-API compatible, while its fastest variant is aimed at latency-sensitive agent routing.
The new program creates a specialized Gmail deliverability path for verified political senders, but it is not an inbox guarantee: recipients can still mark mail as spam, block senders or unsubscribe, and non-compliant domains can be suspended or removed.
The limits themselves were already documented; the material change is enforcement. Free-tier D1 workloads that previously relied on soft overage behavior now need query-cost awareness, indexes and a plan for temporary failures or paid migration.
Google is changing Gemini Notebook’s packaging from feature-style quotas toward a compute budget. That gives users more flexibility but makes the effective cost of one request less predictable and ties premium upgrades more directly to computational intensity.
Google is tying licensed commercial content directly to an AI workspace: book ownership becomes the access control for grounded AI use. That gives publishers a new distribution path while keeping paid-source entitlement inside the AI experience.
For some queries, Google is no longer waiting for a user to open a longer AI Overview. The system can expand it automatically, creating a larger zero-click answer surface and moving traditional results further down the page.
The integration brings Amazon’s catalog into YouTube’s native shopping layer. It reduces the gap between product recommendation and purchase compared with description links, but access currently requires both YouTube and Amazon affiliate eligibility and is still limited to a select group of creators.
Google’s new agent FinOps model combines hard monthly spend caps that pause agent API calls, Flexible Savings Plans with one- or three-year commitments, pay-as-you-go Gemini Enterprise usage and planned deferred execution at up to half normal inference cost. The controls are useful, but commitment economics and task eligibility need to be modeled carefully.
Google Cloud’s Developer Device Platform is now in public preview with remote physical-device streaming, parallel emulator testing, smart sharding and an agent skill that can drive multi-step journeys, inspect visual issues and feed fixes back into coding agents. It is billed per active device minute and remains a pre-GA service.
Railway Cloud Agents are managed, persistent development machines rather than a new model or harness. They reuse developers’ existing agent credentials, sleep when disconnected by default, retain disk state, and live inside Railway project environments—blurring the boundary between remote coding workspace and deployment platform.
Legora’s Agent Pro pricing illustrates a concrete AI SaaS shift: base platform economics can remain seat-oriented while high-variable-cost agent work is metered separately. The model is notable for its controls as much as its pricing—and for what it does not disclose publicly.