As AI agent technology accelerates its penetration into core business scenarios of enterprises, autonomous execution-type AI represented by OpenClaw has become a key engine for the new round of productivity transformation. However, its powerful system access and task execution capabilities also bring unprecedented security risks—such as privilege overreach, memory contamination, malicious induction, and computing resource abuse are becoming increasingly prominent. Some enterprises have underestimated the potential governance challenges while embracing efficiency.

image.png

To address these issues, Ant Data officially launched the "Yitianjian 2.0 - Lobster Guardian" AI security protection system and simultaneously initiated the "Lobster AI Security Protection Plan," aiming to provide enterprises deploying OpenClaw with comprehensive, multi-layered security guarantees throughout the entire lifecycle.

The newly released "Claw Security Kit 1.0" was officially launched on March 19, focusing on three core capabilities: first, "Counteracting Ideological Mutation," which identifies input and output content accurately to intercept prohibited information and sources of malicious inducement, preventing the model from being "hacked"; second, "Cleaning Skills Repository," based on Ant Data's CALIR five-dimensional model framework, it conducts compliance scans and behavioral audits on Skills plugins, effectively identifying and blocking risks such as privilege overreach, sensitive data leakage, and abnormal calls; third, "Risk Public Opinion Report," which generates real-time risk emergency reports and actively pushes them, helping enterprises grasp the security situation in a timely manner, achieving transparency and immediate awareness of risks, turning passive defense into active response.

To help more enterprises achieve secure baseline construction at low cost, Ant Data also launched the "Lobster AI Security Protection Plan." This plan will provide the first 100 partner enterprises with 200,000 free Lobster security protection calls over a period of three months, aiming to allow customers to quickly experience enterprise-level AI security assurance.

In terms of large model security technology, Ant Data has successfully been listed in IDC's "China Large Model Security Market Map" due to its leading technical capabilities and solutions. Based on real-world attack and defense experiences, Ant Data's large model security technology has achieved SOTA (State Of The Art) effects in complex scenarios such as financial compliance and jailbreak attacks, and is capable of generalizing and fixing zero-day risks.

Ant Data's continuous innovation in the field of AI security has become a leading security practice in the industry, ensuring that customers can still enjoy controllable and traceable security guarantees while benefiting from AI agents. "AI agents are not 'black boxes,' nor should they be 'blind boxes,'" said the AI Security team at Ant Data. "We hope that through 'Lobster Guardian,' enterprises can enjoy the benefits of AI agents while having a foundation of explainable, controllable, and traceable security."