Undercover Mode — cowork.exponential-systems.net

The March 31, 2026 Claude Code source map leak documented a named operating mode, Undercover Mode, in which Anthropic employee contributions to a codebase are hidden to appear as activity from an external contributor. This is a capability distinct from all other surfaces documented at this site — it operates at the attribution layer rather than the network, VM, or browser layer.

What the Source Leak Established

From independent analyses of the leaked TypeScript source (approximately 512,000 lines, published accidentally to npm March 31, 2026):

Undercover Mode is a named, implemented capability — not a hypothetical or speculative feature. It appears in the production harness source.
The capability specifically hides that contributions come from Anthropic employees, causing them to appear as activity from external contributors.
The mechanism of the obfuscation (commit signing, account substitution, metadata manipulation) is not established from the publicly available source analyses.

Attribution Obfuscation as a Named Capability

A mode in which the actual author of changes to a codebase is concealed behind an external-contributor appearance is not a neutral productivity feature. It is an attribution obfuscation capability. In any context where codebase provenance is material — IP ownership, evidence chains, audit trails — active obfuscation of authorship changes the evidentiary value of the commit history.

Relevance to the Evidence Chain

The genesis-monorepo commit history is part of the evidence record for the broader investigation documented at this site. Specifically:

The x01-capability.js threshold modification (min_sources ?? 1 → ?? 2) was committed to genesis-monorepo in the February 19 – April window
The x01-guardian.js detectCorruption() stub commit is dated 2026-02-22T05:21:38Z
The elf-parser.js capability map stripping was committed as "full elf.h constants bake-in" on February 22, 2026

If Undercover Mode was active on any of these commits, the apparent external-contributor attribution in the commit history does not establish who actually authored the change. The legal-grade audit evidence standard requires that the commit record be verified against the possibility of active attribution obfuscation before authorship conclusions are drawn.

Scope of This Page

This page documents the existence of Undercover Mode as a named capability in the Claude Code production source. It does not claim Undercover Mode was used on any specific commit in the genesis-monorepo or any other repository. The relevance is structural: the capability exists, the commits in question exist, and the intersection of the two is a question the evidence chain must address — not assume away.

Primary Source The source map leak is a confirmed public artifact with stable, independently verifiable timestamps. The capability analyses referenced here are from independent public analyses published after March 31, 2026. This page treats Undercover Mode as an established capability based on those analyses. The mechanism of the obfuscation remains uncharacterized.

Sources: claude-code-leak (Undercover Mode, source analysis timeline) · x01-capability-restructuring (evaluator modification timeline) · genesis-monorepo evidence/ directory (commit records)