Chain of News 24/03/2026

**Top Story** OpenAI is committing at least $1 billion to a new foundation focused on tackling global challenges through AI. This isn’t just philanthropic window dressing; it’s a strategic move to shape the narrative around AI development and address growing concerns about existential risk and societal impact. The foundation will focus on four key areas: curing diseases, expanding economic opportunity, ensuring AI resilience (safety and security), and supporting community programs. This substantial investment signals OpenAI’s intent to be a leader not just in AI *creation*, but also in responsible AI *deployment* and mitigating potential harms, a critical step as the technology becomes increasingly powerful and pervasive. **AI Models & Research** A decade of progress in AI for science is being reflected upon, highlighting the shift from rule-based systems to data-driven approaches. Researchers are streaming experts from Mixture-of-Experts models to overcome RAM limitations, enabling the use of larger models on constrained hardware. Moda is leveraging a multi-agent system built on Deep Agents and LangSmith for non-designers to create professional-grade visuals, demonstrating the power of AI-assisted creative tools. ChatGPT is introducing Agentic Commerce Protocol for richer, visually immersive shopping experiences, including product comparisons and direct merchant integration. Anthropic’s Claude is facing legal challenges as the Pentagon attempts to ban its use, with Anthropic seeking an injunction – a reminder that even “research preview” safeguards aren’t foolproof. **Developer Tools & Frameworks** The Copilot SDK is now available for integration into React Native applications, allowing developers to build AI-powered issue triage systems with production-ready patterns for graceful degradation and caching. Dynamic Workers are being introduced, offering a 100x speed improvement over traditional containers for executing AI-generated code in secure isolates, enabling millisecond startup times for AI agent sandboxing. Datasette-files 0.1a2 has been released, adding the ability to directly upload files into a Datasette instance, expanding its utility for data management. A critical security vulnerability was discovered in LiteLLM v1.82.8, containing a credential stealer hidden within a compromised file – developers should immediately update to a patched version. Finally, Aqua Security’s Trivy vulnerability scanner has been affected by a security incident across multiple distribution channels, requiring users to investigate potential compromises. **Industry & Business** Mirage, the company behind the Captions video editing app, has secured $75 million in growth financing from General Catalyst’s Customer Value Fund. Agile Robots is partnering with Google DeepMind, integrating DeepMind’s robotics foundation models into its robots and providing data back to DeepMind for further research. MSL has “execuhired” the team behind Dreamer, just days after their appearance on a podcast, showcasing the rapid acquisition pace in the AI talent market. The OpenAI foundation’s $1 billion investment represents a significant commitment to responsible AI development and deployment. **Worth Watching** The concept of “slop” – AI-generated output requiring significant human effort to refine – is gaining traction as a critique of the current state of generative AI. The Spanish-speaking world is seeing increased coverage of AI’s impact on healthcare (reducing wait times in dermatology), the future of work (job security/displacement), and the film industry. There’s a growing discussion around the value exchange when receiving raw AI output, with some arguing it demonstrates a lack of respect for the recipient’s time. Finally, a judge is hearing arguments regarding the Pentagon’s attempted ban of Anthropic’s Claude, highlighting the ongoing tension between government regulation and AI innovation.