OpenAI just launched autonomous agents for your team

Major AI model releases span multimodal capabilities including advanced Gemini 3.1 variants for speech, robotics, and complex reasoning, alongside innovations in distributed training, AGI measurement frameworks, and creative AI applications.

OpenAI introduced workspace agents in ChatGPT—shared, autonomous AI agents powered by Codex that handle complex workflows across teams, running continuously in the cloud with enterprise controls. Early users report substantial time savings, with agents automating tasks from lead qualification to financial reporting within existing tools like Slack.

The Verge

Anker has created its own compute-in-memory AI chip called Thus that enables more powerful on-device AI in small form factors like earbuds, first launching in Soundcore products in May to deliver significantly improved call quality and noise cancellation through larger neural networks.

Testing Catalog

Google launched Workspace Intelligence, a semantic layer that enables Gemini AI agents to understand and act across the entire Workspace suite by mapping emails, files, chats, and projects into shared context. This transforms Workspace from disconnected apps into an agentic system with enterprise security controls, challenging Microsoft 365 with deeper cross-app AI integration.

Morgin AI

Language models labeled as 'uncensored' still systematically suppress charged words at the probability level—a 'flinch' effect baked into pretraining that persists even after refusal-removal techniques are applied. Testing across seven models revealed commercial pretrains show significantly more word-level censorship than open-data alternatives, and popular 'abliteration' techniques fail to remove this underlying bias.