Back to News
SecurityAI Understanding briefing

Runjian unveils Zhiyu AI security system to protect enterprise AI deployments

Runjian Co. introduced its Zhiyu suite – an AI‑focused security platform that includes an autonomous penetration‑testing agent, model‑shielding, and continuous monitoring tools – aiming to redefine protection boundaries for AI‑driven enterprises.

4 min readRead the linked source
Source-provided image accompanying Runjian unveils Zhiyu AI security system to protect enterprise AI deployments
Source referenceSource recorded
Publisher
eu.36kr.com
Source link
eu.36kr.comhttps://eu.36kr.com/en/p/4002530176110469
Source type
Linked source — primary-source status has not been established.
ContextUnderstand this in 60 seconds

Start here

Key terms

Prompt Injection
An attack pattern where malicious instructions are inserted into model inputs or retrieved content.
Guardrails
Rules, checks, and controls that limit unsafe or undesired model behavior.
Benchmark
A standardized test or dataset used to measure and compare model performance.
Test yourselfAI Ethics Quiz

What happened

Runjian Co., Ltd. announced the launch of its Zhiyu intelligent security product system at the 2026 World AI Conference (WAIC). The suite comprises four integrated components: Hive, an AI‑driven penetration‑testing system that uses multi‑agent swarm collaboration to automate full‑stack attacks; ModelShield, which intercepts malicious prompts and outputs at model ingress and egress; ModelInsight, a continuous evaluation engine that generates attack samples to uncover model blind spots; and a monitoring layer that ties the tools together into a closed‑loop defense workflow. In a live demonstration, Hive completed an end‑to‑end attack chain without human intervention and scored 98.44 points on Tencent Security’s TSecBench evaluation, ranking first among tested solutions. Runjian positioned Zhiyu as a way for overseas customers to build trust in Chinese AI services by addressing token‑invocation leakage, cross‑jurisdictional compliance, and AI‑specific threat vectors such as and agent privilege abuse.

At the WAIC, Runjian’s Chief AI Security Scientist Dr. Chu Ge described the shift from protecting systems against hackers to preventing AI from being deceived, abused, or prompt‑injected. The Zhiyu suite was presented as a response to this new threat model.

Hive, the suite’s penetration‑testing component, employs a swarm of autonomous agents that divide tasks such as reconnaissance, exploitation, verification, and support. The agents share intelligence on a unified operation map, enabling rapid progression through complex environments without manual oversight.

ModelShield sits at the model’s input and output interfaces, applying prompt‑enhancement techniques and bidirectional detection to block malicious payloads. ModelInsight continuously generates synthetic attack samples to probe the model’s behavior before and after deployment, feeding findings back into the defense loop.

The four components are integrated through a monitoring layer that logs events, enforces policy, and provides alerts to customers, forming a “lock + security guard + monitoring system” metaphor described by Dr. Chu.

Source details: eu.36kr.com ↗

Why it matters

The announcement marks one of the first commercially packaged systems that treats AI itself as a security perimeter rather than an after‑thought. As enterprises embed large language models into customer‑service bots, contract‑review pipelines, and internal decision‑making, traditional vulnerability‑patching approaches no longer cover the new attack surface where malicious prompts can coerce models into disallowed behavior. Runjian’s claim that its suite can autonomously discover and mitigate such threats addresses a gap that regulators and industry analysts have flagged as a critical risk for AI adoption. By offering a full‑lifecycle solution—from automated red‑team testing to real‑time and post‑deployment monitoring—the platform could become a reference point for companies seeking to meet divergent data‑security laws in China, ASEAN, and other regions. If the performance claims hold up in broader deployments, Zhiyu may accelerate the commercial viability of AI‑driven services by reducing the perceived security liability for multinational firms.

Traditional cybersecurity tools focus on static vulnerabilities, firewalls, and human‑in‑the‑loop analysis. AI‑centric attacks can bypass these controls by manipulating model inputs, making a vector that bypasses network perimeters.

Runjian’s claim of a 98.44 score on TSecBench suggests a high level of automated attack capability, which, if verified, could set a new for AI red‑team tools and push competitors to develop comparable defenses.

The system addresses compliance challenges for Chinese AI firms operating abroad, where data must satisfy both local regulations and China’s Data Security Law and Personal Information Protection Law. By securing token invocation and providing audit trails, Zhiyu could help firms navigate these overlapping legal regimes.

If widely adopted, the platform could reduce the cost and expertise barrier for AI security, encouraging more enterprises to integrate large models into critical workflows while maintaining regulatory compliance.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Agent Lifecycle Stage:
1
User Intent & Planning: "Audit customer refund request #4092 and settle payment."
2
Tool Calling: Emits structured JSON call crm_get_transaction(id='4092').
3
Guardrail & Verification:🛡️ Paused: High-value action requires human operator sign-off.
4
Final Settlement: Refund recorded, email receipt dispatched, and audit log stored.
Core takeaway: An AI agent is not just a language model—it is a closed loop of planning, tool invocation, and environment feedback. Production systems require self-healing retries and strict human approval guardrails.
Interactive Concept Check+10 Points
AI Ethics Quiz

Impossibility results in algorithmic fairness (e.g. Kleinberg et al., Chouldechova) show what?

What to watch next

Key indicators to monitor include: (1) whether Runjian publishes independent third‑party audit results confirming Hive’s TSecBench score; (2) pricing and licensing terms, which the company has not disclosed, to see if the solution is accessible to midsize firms; (3) adoption by Chinese AI vendors expanding into ASEAN markets, especially any pilot projects that reveal real‑world effectiveness; and (4) regulatory responses in jurisdictions where token‑invocation data flows across borders, which could shape compliance requirements for AI security products.

Third‑party validation: Independent security labs may test Hive and ModelShield to confirm the reported performance and identify any false‑positive or false‑negative rates.

Commercial rollout: Pricing, licensing models, and cloud‑hosting options will determine whether the suite is limited to large vendors or accessible to smaller AI adopters.

Regulatory impact: Governments in ASEAN and beyond may reference Zhiyu’s capabilities when drafting AI‑specific security standards, especially concerning cross‑border data token handling.

Competitive response: Other security vendors may launch rival AI‑focused products, leading to a nascent market for AI lifecycle protection tools.

Related guides & quizzes

AI EthicsAI AgentsPrompt EngineeringAI Models ExplainedTest what you know — try a free AI quizLook up an AI term in our glossaryFollow the AI regulation tracker
Found this useful?