Meta announced the release of its updated Advanced AI Scaling Framework and Safety & Preparedness Report for Muse Spark on April 8, 2026. The new framework expands the types of risks evaluated, strengthens deployment decision-making, and introduces new Safety & Preparedness Reports. These reports detail risk assessments, evaluation results, and the rationale behind deployment decisions. The framework applies to all frontier deployments, including open, controlled API access, and closed models. The company emphasized that models are only deployed when they meet the standards set by the framework. Muse Spark, which is more capable than previous models, underwent extensive safety evaluations before deployment. The evaluation process included testing against thousands of scenarios designed to find weaknesses and monitoring live traffic for unexpected issues. The results showed strong safeguards across all measured risk categories. The report also confirmed that Muse Spark does not possess the level of autonomous capability needed to pose control risks. Safety protections are integrated at every stage, from data filtering to training and product-level guardrails. The approach evolved from rule-based training to a more comprehensive system that includes reasoning about safety principles. This work complements human oversight and improves as the model’s reasoning capabilities advance. *Source: [metaai](https://ai.meta.com/blog/scaling-how-we-build-test-advanced-ai/)*