As Agent becomes the mainstream form of enterprise AI adoption, security has shifted from a nice-to-have to a must-have — without it, companies simply will not deploy. Volcano Engine has released the AI Trust security product family, a three-layer architecture of "trustworthy models - controllable agents - AI-driven security operations" that puts a security lock on enterprise Agent deployments.

These capabilities come from the accumulated AI security practice inside ByteDance and have been battle-tested by real businesses.

Layer 1: Trustworthy Models —— AICC Confidential Computing

The two big pain points for enterprises using cloud models are compliance and privacy. AICC (AI Confidential Computing) is rooted in chip-level trust, providing end-to-end full-link encryption and complete traceable auditing, helping enterprises achieve security protection equivalent to a private deployment with a lightweight investment.

Core features:

• Multi-source model ecosystem: Supports the full Doubao family plus open-source SOTA models such as DeepSeek, GLM, and Kimi
• Full compatibility with domestic Xinchuang stacks: Adapts to mainstream domestic chips such as Huawei, Cambricon, and Hygon
• Native integration with Agent development platforms: Products like ArkClaw, Agentkit, and HiAgent work out of the box

It now provides confidential inference services for industries including smartphones, finance, automotive, healthcare, central and state-owned enterprises, and semiconductors. China Mobile has also jointly launched a "Mobile Engine confidential model service" zone with Volcano Engine, building a cloud-network-edge-device integrated trusted AI inference architecture.

Layer 2: Controllable Agents —— AI Assistant Security Platform

Agents are the executors of the AI era, and they need full-link security protection before businesses can run them with confidence in enterprise workloads. The AI Assistant Security Platform builds a complete governance system:

• Runtime security: Defends against prompt attacks, prevents data leakage, intercepts high-risk operations, and applies sandboxing and static detection to all kinds of tools
• Identity and permission control: Compatible with mainstream identity protocols, governs Agents with least privilege, and keeps every operation traceable end to end
• Global posture monitoring: Unifies monitoring of all Agent assets, behaviors, and risks, enabling both audit and interception

Over the past two months, the platform has supported 10 billion detection calls per day. In IDC's "China Agent Threat Detection Technology Assessment", Volcano Engine was evaluated among 13 mainstream vendors and took first place in the total score along with several first-place category wins.

In addition, for the compliance filing that enterprises must complete before launching Agents externally, Volcano Engine provides a large-model compliance filing solution, supported by twin engines — large-model security assessment and an application firewall — to help enterprises accelerate commercialization.

Layer 3: AI-Driven Security Operations —— Security Operations Agent

In the AI era, alerts and vulnerabilities are growing explosively. With models at its core, the Security Operations Agent focuses on 7 key enterprise security operations scenarios and closes the loop through multi-agent collaboration and self-evolution.

Core metrics:

• Hundreds of thousands of daily alerts triaged 100% by AI
• Per-alert triage takes under 1 second
• Cold-start accuracy exceeds 95%, reaching above 99% after 1-2 days of self-directed learning

After GAC Group integrated the Security Operations Agent into its own business workflows, alert-handling efficiency improved 10-fold, vulnerability assessment sped up 7-fold, and code fix cycles were cut in half.

Where It Fits

• Enterprise-grade Agent deployment: Scenarios that need compliance auditing and runtime security protection
• Highly sensitive data processing: Confidential inference needs where data must not leave the domain, as in finance, healthcare, and central and state-owned enterprises
• Large-scale security operations: Security teams facing massive alert volumes, replacing manual triage with AI
• Agent compliance filing: Enterprises that need to complete large-model application filing quickly