StepFun and Shanghai AI Lab Release Next-Gen Agent System Security White Paper

StepFun (阶跃星辰) and the Shanghai Artificial Intelligence Laboratory on July 13 jointly released the "Next-Generation Agent System Security Technology White Paper," providing a systematic technical reference for securing AI agent systems.

The white paper addresses security risks in AI agent scenarios including autonomous decision-making, tool invocation, and multi-agent collaboration. It proposes a technical framework spanning identity authentication, permission management, behavior auditing, and adversarial defense.

This is the first security white paper jointly published by a Chinese LLM company and a leading AI research institution. StepFun contributes real-world deployment scenarios and engineering practices, while Shanghai AI Lab provides cutting-edge security research capabilities.

As AI agents move from concepts to production deployments, security concerns have become increasingly prominent. Agent systems face threats including prompt injection, privilege abuse, and data leakage — risks that traditional security frameworks were not designed to address. The white paper proposes systematic countermeasures for these novel attack surfaces.

StepFun has been making rapid product moves recently — its just-announced STEPX Neo AI phone also uses on-device agent architecture. The white paper provides compliance assurance and technical credibility for these products.

The publication reflects the Chinese AI industry's accelerating efforts to establish agent security standards. As AI agent commercialization picks up pace, the industry urgently needs unified security specifications and evaluation systems.

Key questions going forward include whether the white paper's technical recommendations will evolve into formal industry standards, and whether other LLM companies will follow with similar security framework documents. Agent security is emerging as a critical infrastructure layer across the AI industry chain.