The last 24 hours brought a flurry of frontier-model news as Google wrapped up its I/O keynote, Anthropic completed a strategic tooling acquisition, and regulators on both sides of the Atlantic tightened their grip on how advanced AI is tested and shipped.
Google ships Gemini 3.5 Flash, Omni, and the Spark agent
At Google I/O 2026, the company made Gemini 3.5 Flash generally available as the new default model in the Gemini app and AI Mode in Search, advertising frontier-level performance on coding and agent benchmarks at a fraction of competitors’ prices. Google also unveiled Gemini Omni, a cinematic video generation and editing model, and Gemini Spark, a proactive agent that runs multi-step workflows in the background for AI Ultra subscribers. CNBC has the keynote rundown.
Anthropic absorbs Stainless and winds down hosted SDK tooling
Anthropic confirmed its acquisition of Stainless, the developer-tools startup whose SDK generator powers official client libraries for OpenAI, Google, Cloudflare and Anthropic itself. The company is shutting down Stainless’s hosted services, forcing rival labs to internalize SDK and MCP tooling that they had effectively outsourced. Read Anthropic’s announcement.
EU Commission publishes draft high-risk AI guidelines
On May 19, the European Commission released its long-awaited draft guidelines on the classification of high-risk AI systems under the AI Act and opened a public consultation through June 23. The text arrives just days after EU institutions agreed a Digital Omnibus deal that defers most Annex III obligations to December 2027 while extending prohibitions to non-consensual “nudifier” applications.
OpenAI pushes content provenance standards
OpenAI used a May 20 research post to detail its expanded work on content provenance, outlining how watermarking, cryptographic signing and C2PA-style metadata can travel with AI-generated images, audio and video across platforms. The move comes as U.S. and EU regulators converge on disclosure requirements for synthetic media.
Pre-deployment testing pacts gain a third partner
The Center for AI Standards and Innovation (CAISI) confirmed that Google DeepMind, Microsoft and xAI have now all signed agreements to submit frontier models for government red-teaming before public release, formalising a pre-deployment evaluation framework that had been voluntary since 2023. CNN reports on the expanded program.
What to watch: developer reaction to the Stainless wind-down — and whether OpenAI or Google answer Gemini Spark’s agentic push within the week.