Observations, analysis and practice on software supply chain security, open-source governance and AI security.
No matching articles
Most teams start agent adoption by granting everything to get it working, and never come back to tighten it. This article proposes an authorization model centered on capability tags, standardizing what a tool can do, authorizing by tag rather than by tool, and handling capability combinations, auditing and revocation. It also explains why conventional IAM thinking needs adaptation for agent scenarios.
Poisoning incidents in public registries have become routine, and attacker evasion has evolved alongside them. This article breaks down four mainstream hiding techniques — obfuscation, delayed triggering, environment detection and staged loading — explains why static scanning alone misses them, and sets out a three-layer defense combining static review, behavioral analysis and egress monitoring.
Source scanning came back clean, so why is the image full of vulnerabilities? Because a container image is a layered set of components — base image, system packages and language dependencies each carry their own risk surface, and source scanning only covers the top one. This article breaks down the three layers, explains why the two scans disagree, and offers practical guidance on base image governance and where to place the gate in CI.
MCP is becoming the standard interface through which AI agents reach the outside world, and with that it has become a new class of supply chain component — installed from public registries, granted real credentials, running with process privileges, yet rarely subjected to the scrutiny applied to open source dependencies. This article lays out eight checks to complete before adoption, covering provenance, tool description auditing, permission scope, version pinning and network egress, and explains why malware detection alone cannot support an admission decision.
AI-assisted coding has moved from novelty to daily practice, but the security discipline around it is often still designed for an era when humans wrote every line. This article organizes ten actionable rules across five stages — prompting, generation, dependency intake, review and merge — covering hallucinated package defense, dependency allowlists, snippet-level license detection and provenance tracking, so teams can stop risk before merge without sacrificing velocity.
Process isolation, dynamic linking, network interfaces, and separate distribution — the four dominant GPL isolation patterns each draw a different line around copyleft contagion. This article systematically breaks down the engineering implementation and legal logic behind each pattern, dispels common misconceptions, and offers a practical checklist for architecture reviews, helping engineering teams find a workable balance between compliance and delivery velocity.
The NTIA's seven minimum elements are the entry bar for SBOM, not the finish line. This guide dissects the most common field-level pitfalls—from hash algorithm selection and supplier name ambiguity to dependency depth and VEX linkage—helping both producers and consumers elevate SBOM from a compliance checkbox into a genuinely actionable security asset.
Chip companies are both heavy consumers of open source and distributors of software at scale — firmware, drivers and SDKs ship with the silicon to customers worldwide, every layer carrying open-source obligations. Four ways the industry is different, and the governance moves that matter.
Expert review committees buckle under the pressure of high-volume component intake, while automated gates often become theater when the underlying policies are too coarse. This piece breaks down the core evaluation dimensions of an open-source intake policy, maps out an engineering path for the policy matrix, and clarifies how exception workflows should interface with procurement and legal.
Since 2023 the FDA refuses premarket submissions for connected devices without an SBOM, and China's NMPA cybersecurity review requires an off-the-shelf software inventory. Ultra-long lifecycles, legacy systems and the fear of patching give medical open-source governance a shape all its own.
Open source governance in the securities sector may look similar to banking on the surface—both are financial services, after all—but the real-time constraints of trading systems, the cadence of Xinchuang migration, the specifics of regulatory reporting, and the mechanics of emergency drills together produce a fundamentally different set of implementation requirements. This article dissects those differences through a RegTech lens and offers actionable control strategies.
An SCA gate is not the same as a blanket block. This article breaks down a production-ready CI gate configuration methodology — covering threshold design, tiered blocking strategies, incremental scanning, false-positive handling, and Jenkins/GitLab CI integration — to help teams find a genuinely sustainable balance between security and delivery velocity.
A software-defined vehicle carries over 100 million lines of code, most of it open source. UN R155, ISO/SAE 21434 and China's mandatory GB 44495 have written supply-chain security into market access, and OEM pressure is cascading down the tiers. The four ways automotive is different, and what to do.
Industry reports keep finding that the vast majority of codebases contain open-source components years out of date. Abandoned components get no patches and no one answers vulnerability reports — the most overlooked supply-chain risk. How to identify, grade and remediate.
The hard part of license risk is not any single license but combinations — GPL inside proprietary code, Apache-2.0 meeting GPLv2, declarations that don't match the files. Five common conflict patterns and a four-tier remediation playbook, all automatable.
China's five-regulator joint opinion made open-source governance a compliance obligation for financial institutions. We unpack the four regulatory keywords and lay out a working framework of inventory, admission, monitoring and incident response.
Of the two-thousand-plus open-source licenses, which ones actually deserve enterprise caution? A four-tier map by contagion — permissive, weak copyleft, strong copyleft, and commercial-restriction clauses — plus a three-step governance rollout.
Supply-chain poisoning is spreading from npm packages to MCP servers and Agent Skills. A review of xz, Shai-Hulud and postmark-mcp — and why malware detection is no longer enough.
Studies keep finding that roughly 40% of AI-generated code contains exploitable flaws — while developers using AI assistants feel more confident. The data, four mechanisms, and a governance path.
The CRA is in force: vulnerability-reporting obligations start September 2026, full application in December 2027, with fines up to €15M or 2.5% of global turnover. A six-point checklist.
The gap between SCA tools isn't the feature list — it's the detection principle. Manifest-level trusts what you declare; snippet-level verifies what you actually shipped. Five blind spots and an evaluation checklist.
SPDX, CycloneDX or SWID? US EO 14028, FDA, the EU CRA and China's financial-sector rules — what they require, and a four-step path from generation to operations.
As code is mass-produced by AI and dependencies are pulled in automatically by agents, the old "scan-and-inventory" paradigm starts to break. Security has to move from after-the-fact to the moment of production.
Many teams treat an SBOM as a document to hand in. But a valuable SBOM drives decisions — which vulns are exploitable, which dependency to fix first, which license carries risk.
商务合作
微信公众号