Ant Group releases open-source AI safety models for agents and multimodal systems
Ant Group has open-sourced SingGuard-NSFA, a safety guardrail model for autonomous AI agents, and detailed SingGuard, a multimodal safety model. The models aim to protect AI systems from various risks.
Chinese technology firm Ant Group has released SingGuard-NSFA, a safety guardrail model for autonomous AI agents, as open-source software. The company also disclosed details of SingGuard, a safety model designed for multimodal AI systems.
SingGuard-NSFA is engineered to detect and prevent risks such as prompt injection, sensitive data theft, malicious code execution, resource abuse, and permission misuse before AI agents take action. The model encompasses seven major risk categories, 28 subcategories, and 185 scenarios across 133 languages.
Its evaluation system includes nearly 100,000 samples. According to Ant Group, the model is available in versions with 0.8 billion, 2 billion, 4 billion, and 9 billion parameters, and can perform a single risk judgment in approximately 50 milliseconds.
As AI systems become more sophisticated and autonomous, technology companies are increasing their focus on safety mechanisms. The release of these models as open-source aims to accelerate industry-wide development and enhance overall AI security.