Snapchat Blocks AI-Generated Videos From Spotlight Monetization

Google DeepMind has launched a new institute designed to surface and debate differing perspectives on artificial general intelligence (AGI) between Google, Google DeepMind, and the global research community. The institute acknowledges that stakeholders will not always agree and may change positions as new data emerges in the rapidly evolving AGI field. The move signals an effort to broaden the conversation around AGI development beyond internal company views.

Base Labs, the research group spun up by Baseten earlier this year, has launched a partnership with Hugging Face and Goodfire to develop and publish methods for training and monitoring open-weight AI models. The collaboration focuses on AI safety practices for open models, addressing a gap in standardized approaches to model development and oversight. The partnership will produce publicly available methods and tools for the open-source AI community.

OpenAI has published a framework for tracking, investigating, and disclosing instances of model misalignment, along with six reports documenting unexpected or concerning model behaviors. The framework establishes a systematic approach to identifying and communicating when AI models behave in ways that diverge from intended design. This represents a step toward greater transparency in how AI developers handle safety issues.

AWS released 38 open-source agent skills across 11 healthcare and life sciences domains to fix a critical failure mode in AI agents: they cite correct frameworks but misapply them in practice, leading to silent errors in variant interpretation, claims processing, and clinical workflows. The skills encode structured decision procedures as markdown documents that agents consume at inference time, improving head-to-head performance by 70-86 percent and critical thinking by 78-85 percent. All skills are released under MIT-0 license and available on GitHub.