OpenAI offers $445K to prep for AGI (and the Pope weighs in)

Google

Google launches Managed Agents in Gemini API, allowing developers to deploy production-grade agentic systems with a single API call that provisions isolated Linux environments for reasoning, tool use, and code execution. This abstracts away infrastructure complexity, building on the Deep Research agent launched in December.

Simon Willison

Pope Leo XIV issued a comprehensive encyclical on AI ethics, addressing interpretability problems, environmental costs, algorithmic bias, and power concentration, while calling for data to be treated as a public good and emphasizing human accountability in AI decision-making. The document represents significant institutional engagement with AI governance from a major religious authority.

OpenAI is hiring a safety researcher for up to $445,000 to prepare for recursive self-improvement in AI systems, where models can train better versions of themselves. The move reflects growing industry concerns about rapidly advancing AI capabilities approaching what leaders call the 'singularity.'

Testing Catalog

Anthropic is preparing to launch Memory Files for Claude, a structured multi-document memory system that would significantly expand the assistant's ability to retain and recall user context across sessions, potentially alongside its Dreams memory consolidation feature.

Meta launched Forum, a Reddit-style standalone app for Facebook Groups with AI-powered Q&A and moderation features, as part of CEO Zuckerberg's strategy to rapidly scale app development using AI efficiencies.