frontier models

Cisco is preparing infrastructure and security strategies for widespread deployment of AI agent coworkers and managing risks from frontier models. This reflects enterprise recognition that AI agents will become embedded operational tools, requiring new architectural and threat-management approaches.

Meta launched Muse Spark, its first frontier AI model that breaks from the company's open-source tradition by not releasing the model weights publicly. This shift signals Meta's strategic pivot toward protecting its most advanced AI capabilities from competitors, potentially reshaping the open-source AI landscape.

The new ARC-AGI-3 benchmark offers $2M to any AI that matches untrained humans, yet every frontier model scores below 1%. This matters to technologists and investors as it highlights the significant gap between current AI capabilities and human-level general intelligence, potentially affecting AI investment valuations.