COMPANY
Apiiro
companytopic-notesecurity
Overview
Apiiro is a security research firm covering application-security posture, coding-agent trust surfaces, and secret-leak telemetry across public source-code hosts. The firm’s 2026-11 write-up on persisted permissions.allow entries in Claude Code‘s "Don't Ask Again" flow is the corpus’s first entry pointing to the firm as a primary source on agent-permission-state leakage.
Timeline
- 2026-11-08-AI-Digest — Apiiro’s blog post claims that Claude Code‘s persisted
permissions.allowentries (“Don’t Ask Again”) have leaked credentials across1,122public GitHub files. Load-bearing softener carried in the digest: the1,122figure could not be independently corroborated in the verification pass (apiiro.com DNS-blocked; no secondary outlet replicated the exact count), so attribution sits with Apiiro’s own post until a second source replicates it. The closest confirmed prior is GitGuardian’s April2026report of30npm packages leaking.claude/settings.local.jsoncredentials. Carry today’s finding as “latest instance of a known class — agent/dev-tool permission state leaking through packaging boundaries — not a novel vector”. Pairs with Claude Codev2.1.319’s Auto-mode server-side-gating flip from 2026-11-07-AI-Digest as the permission-surface the Auto mode hardens and thepermissions.allowleak exposes — same substrate, opposite sides. Log against MOC - Agent Security.
Key Developments
"Don't Ask Again"secret-leak claim (November 2026): Apiiro’s post claims1,122public GitHub files have leaked credentials via Claude Code‘s persistedpermissions.allowentries. Carried hedged — Apiiro’s own figure, not independently replicated, with GitGuardian’s April202630-npm-package leak as the closest prior.
Related
See also: Claude Code, Anthropic, MOC - Agent Security.