Coding Agent Sandboxes Break in Places Teams Do Not Expect

Pillar shows that agent sandboxes must be assessed not only around the agent process, but around files, configs, allowlisted commands, and local daemons the host later trusts.

1 min1 source

In one minute

  • Pillar shows that agent sandboxes must be assessed not only around the agent process, but around files, configs, allowlisted commands, and local daemons the host later trusts.
  • The record is connected to 3 topics: agent-security, coding-agents, sandbox.
  • 1 public source carries the evidence boundary.

Source ledger

Publishable sources attached to this record.

1 public source
#SourceRolePublic status
1pillar.securitysourceprimary receiptsource_urls
On this page

Pillar Security collected a series of sandbox escapes around coding agents: Cursor, Codex, Gemini CLI, and Antigravity. The point is not that sandboxes are useless. It is that the boundary often sits somewhere other than the product diagram suggests.

The agent may not execute a forbidden command directly. It only needs to write a file that a trusted component outside the sandbox will later read and execute: an IDE extension, task runner, Git helper, hook, Python discovery mechanism, Docker daemon, or another local service.

What changes

The old question was: can the agent process escape the workspace?

The harder question is: what can the agent write, and which non-sandboxed parts of the system will later trust that data?

That makes an agentic IDE more than an editor extension. It is an endpoint actor. Security teams must track provenance, side effects, command arguments, local daemons, and host-side automation, not only a list of forbidden paths.

New Runtime Read

An agent sandbox must model the whole trust chain: model output -> file write -> host component -> execution. If controls see only the first process, they miss half the attack.

Open archive
  1. OpenAI's Hugging Face Incident Makes Agent Sandboxes a Production Riskagent-security · sandbox2 sources
  2. Devin Outposts Splits the Agent Brain from the Execution Planecoding-agents · agent-runtime4 sources
  3. Gemini 3.6 Flash Moves the Agent Race Toward Cost per Taskgemini · agent-economics1 source
  4. AI Coding Workflow: From Idea to Verifiable Workcoding-agents · workflow2 sources