INDEX / 10 ENTRIES

Blog

All my blog posts

  1. The Real Frontier of Agent Security: The Four-Pillar Control Plane Behind Anthropic's Finance Agents

    Anthropic released 10 financial-services agent templates powered by Opus 4.7. The real story isn't model performance—it's that per-tool permissions, managed credential vaults, full audit logs, and human-in-the-loop are now default. This signals a shift in vendor security focus from model alignment to deployment-time control.

    • AI Safety
    • Agent
    • Anthropic
    • FinTech
    • Compliance

  2. Decoding Claude Security: How Anthropic Productized the 'AI Security Researcher'

    Claude Security beta, powered by Opus 4.7, is currently exclusive to Enterprise customers. This deep-dive analyzes Anthropic's new product across six dimensions: positioning, architecture, workflow, capability boundaries, pricing, and ecosystem strategy.

    • AI安全
    • Claude
    • Anthropic
    • 代码安全
    • Agent

  3. X-Raying AI Coding Agents with eBPF: How ClawXRay Works

    When Claude Code makes 200 API calls on your machine, do you know what it sent or how many tokens it burned? ClawXRay uses eBPF to intercept TLS-encrypted traffic at the kernel level, making every LLM call from AI Agents fully transparent.

    • AI Security
    • eBPF
    • Agent Observability
    • LLM Security

  4. Protecting Prompts Like Program Control Flow: A Deep Dive into the PCFI Paper

    A paper that brings the concept of Control-Flow Integrity (CFI) from software security to prompt defense. Core insight: Prompt Injection is fundamentally a structural problem — low-priority inputs hijacking the execution flow of high-priority instructions.

    • AI Security
    • Prompt Injection
    • Paper Review
    • LLM Security

  5. Not Bound to Any Agent Framework: How We Built a Security Agent Platform with Dual-Contract Design

    While everyone debates whether to choose LangChain or Claude Code, we took a different path — defining two universal contracts that allow any Coding Agent to serve as the platform's execution engine.

    • AI Security
    • Agent Architecture
    • Agent Platform
    • Context Engineering

  6. Agent Observability Overview 4: From Data to Judgment

    This is the final article in the Agent Observability series, exploring how to go from data to judgment - cost attribution, behavioral baselines, anomaly detection, and making observability truly operational.

    • AI
    • Agent
    • Observability
    • Cloud Native
    • Cost Optimization

  7. Agent Observability Overview 3: Practical Choices for Collection and Reconstruction

    This is the third article in the Agent Observability series, exploring three paradigms of data collection, the misalignment between audit and runtime, and the design principle that reconstructability is better than full collection.

    • AI
    • Agent
    • Observability
    • Cloud Native
    • eBPF

  8. Agent Observability Overview 2: How I Understand Agent Key Objects

    This is the second article in the Agent Observability series, exploring four core objects in the Agent world - Agent, Model Source, Tool, and Agent-to-Agent Link.

    • AI
    • Agent
    • Observability
    • Cloud Native

  9. Agent Observability Overview 1: Why Agents Make "Visibility" Harder

    This is the first article in the Agent Observability series, exploring the core challenges of observability in the Agent era - uncertainty, semantic gaps, and data silos.

    • AI
    • Agent
    • Observability
    • Cloud Native

  10. Runc Container Escape Vulnerabilities (CVE-2025-31133 et al.) In-Depth Analysis

    An in-depth analysis of three critical runc vulnerabilities disclosed in November 2025, exploring their attack mechanisms, exploitation methods, and remediation recommendations.

    • Container Security
    • Kubernetes
    • Vulnerability Analysis
    • runc
    • Container Escape