The CVSS 7.5 rating for CVE-2026-62999 is a significant mischaracterization. This is not a moderate information-disclosure issue — it is a direct path to remote code execution, and the only reason it rates as 'moderate' is that CVSS was designed for opportunistic attacks, not supply-chain exploitation.
The vulnerability is a canonicalization failure: Copier's trusted-repository prefix check operates on percent-encoded URL paths, while the actual Git transport operates on the decoded path. This allows an attacker to bypass the prefix check by embedding encoded sequences (e.g., ..%2f) that resolve to path traversal after the check passes. This is not a new vulnerability class — the same pattern appears in CVE-2001-0333 (Apache traversal through double-encoding), CVE-2002-0676 (IE Unicode canonicalization), and CVE-2021-23337 (Lodash path traversal). Twenty-five years of history has not produced systemic prevention.
But the severity argument is not about the encoding bug — it is about what Copier templates actually are. A Copier template is not an inert archive. It is executable project scaffolding that supports pre- and post-generation hooks, arbitrary Jinja2 templating with access to environment variables, and shell commands that run during project rendering. A template that passes the trusted-prefix check is a template that runs with your user permissions, executing whatever hooks its author defined. The encoding bypass does not create the RCE path — it merely allows untrusted templates to reach the execution stage that trusted templates already reach by default.
The 'user interaction required' qualifier in the CVE is misleading. In developer workflows, that interaction is routine — a developer running copier copy <template-url> <destination> is indistinguishable from an attacker delivering a malicious template through a trusted-looking URL. In CI/CD pipelines, there is often no human reviewing the template URL at all.
The 9.17.0 patch addresses the canonicalization mismatch, but organizations should verify that the fix canonicalizes before prefix matching rather than checking both encoded and decoded forms. More importantly, the trusted-prefix mechanism conflates retrieval trust with execution trust — it controls where Copier clones from, not what runs after cloning. This is a category error that no canonicalization fix can resolve.
If you use Copier in automated pipelines, the fix alone is insufficient. Implement compensating controls: run Copier in isolated network segments, use dedicated build agents without access to production credentials, and audit the hooks and commands in any template before the first run. The EPSS score of 0.00294 measures opportunistic exploitation — it says nothing about targeted supply-chain attacks where the vulnerability is the delivery mechanism itself.