Axios Supply Chain Attack, Claude Code Leaked, AI and Security

• Source: Stratechery (Ben Thompson)
• Date: 2026-04-01
• Type: daily-update
• RDCO Relevance: High (Claude Code IP, AI security, supply chain risks, Anthropic moat)

Two security stories with significant AI implications. First, the axios npm package (100M+ weekly downloads) was compromised via a stolen maintainer token, delivering a cross-platform RAT through a phantom dependency. This was the third major npm supply chain attack in a year, following the Shai-Hulud worm and PackageGate vulnerabilities.

Second, Claude Code’s source code was accidentally leaked via an included source map file in an npm release. The 59.8MB JavaScript map file exposed the harness architecture that powers Claude Code’s $2.5B ARR business. Anthropic confirmed it was human error in release packaging, not a security breach. Thompson notes the leak primarily benefits companies like Cursor and Microsoft who are building harnesses for Claude, not OpenAI, because the key differentiator is model+harness integration, not harness alone.

Thompson’s broader security thesis: AI will be bad for security short-term (vibe-coding vulnerabilities, AI-assisted exploit discovery) but transformative long-term. AI can examine entire dependency trees, audit legacy code, navigate complex security workflows, and stress-test continuously in ways humans cannot or will not.

RDCO note: The Claude Code leak confirms that Anthropic’s harness IP is substantial and commercially valuable ($2.5B ARR). But Thompson’s point stands: copying the harness without the model integration is insufficient. The axios attack is a concrete reminder that our own dependency management needs scrutiny. Cross-ref: Jaya Gupta moat thesis (06-reference/2026-04-10-jaya-gupta-anthropic-moat.md).