Coding agents can wipe prod: what actually stops them | The Platform Radar cover art

Coding agents can wipe prod: what actually stops them | The Platform Radar

Coding agents can wipe prod: what actually stops them | The Platform Radar

Listen for free

View show details

This Season's Biggest Deal | $0.99/mo for 3 months + $20 Audible bonus

Auto-renews at $8.99/mo. after 3 months - terms apply.
Summary

Your permission rules and hooks check the command text. They never check what the command will actually delete. This episode shows which safeguards hold when an agent runs a recursive delete, and which ones only look like they do.


Two real deletions followed the same pattern, and this episode walks through it. You learn why permission rules and pattern-matching hooks are weaker than they look, and how to sort every safeguard around your coding agent into advisory or control. Then we cover sandboxing, how much access an agent's credentials should have, and what is left when every other layer fails. You finish with a checklist to run against your own setup this week.


Chapters

00:00 Intro

01:35 Two deletions, one pattern

04:32 Rules and hooks read text

07:54 The sandbox the OS enforces

11:24 Credentials sized to the task

14:44 When everything else fails

18:06 What to do Monday morning


Sources (retrieved 2026-10-21)

- Claude Code deleted ~600GB: unprompted `rm -rf` on a substitution that resolved to a drive root, GitHub / anthropics/claude-code (report by the affected user) (2026-09-18): https://github.com/anthropics/claude-code/issues/95426

- Configure the sandboxed Bash tool, Anthropic (Claude Code docs) (2026-10-21): https://code.claude.com/docs/en/sandboxing

- Configure permissions, Anthropic (Claude Code docs) (2026-10-21): https://code.claude.com/docs/en/permissions

- Hooks reference, Anthropic (Claude Code docs) (2026-10-21): https://code.claude.com/docs/en/hooks

- Careful Adoption of Agentic AI Services, CISA with ASD's ACSC and international partners (2026-05-01): https://www.cisa.gov/resources-tools/resources/careful-adoption-agentic-si-services

- OWASP Top 10 for Agentic Applications for 2026, OWASP GenAI Security Project (2025-12-09): https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/

- AI Incident Database, Incident 1469: PocketOS production database reportedly deleted by Cursor AI agent running Claude Opus 4.6, Responsible AI Collaborative (AI Incident Database) (2026-04-24): https://incidentdatabase.ai/entities/pocketos/


Follow the show: https://theplatformradar.com

Episode page: https://theplatformradar.com/episodes/deep-dive-1-what-actually-stops-a-coding


The host's voice is an AI clone of Mat Caniglia's own voice, used with his permission.

Hosted by Mat Caniglia, in an AI clone of his own voice.

No reviews yet
In the spirit of reconciliation, Audible acknowledges the Traditional Custodians of country throughout Australia and their connections to land, sea and community. We pay our respect to their elders past and present and extend that respect to all Aboriginal and Torres Strait Islander peoples today.