Level 1. Deviation
GhostApproval: A Trust Boundary Gap in AI Coding Assistants
- Developer
- Not established
- Model
- Not established
- When it happened
- July 8, 2026
- When it was disclosed
- Not established
- How well it is stood up
- Not verified yet
- What caused it
- Not established
- Where it stands now
- Not established
What happened
Reported by wiz.io. Listed here with a link to their report. Nobody on this side has checked it against that report, so it carries no level and this is not our account of what happened.
Our assessment
- Actual
- 1 of 5Deviation
- Potential
- 3 of 5Significant control incident
The system behaved contrary to its instructions, and it stayed contained. The potential level is what the behaviour demonstrated here could credibly have led to, judged from the same evidence.
This level rests on news reports. The developer has not confirmed it here.
This happened in a test, an evaluation or a demonstration. Nobody real was harmed, so the actual level stays at 1. What it showed counts toward the potential level.
We decided to test systematically.
Why this level
- Sensitive data was reachable. Argues for a potential level of 3
- A security boundary was crossed. Argues for a potential level of 3
- A safeguard was bypassed. Argues for a potential level of 2
- The system acted without authorisation. Argues for a potential level of 2
What is established, and what is not
Confirmed
Nothing is confirmed yet.
Reported, not confirmed
A safeguard was bypassed
This transforms a sandbox bypass into an informed consent bypass - the user approves what they believe is a harmless local edit, while the agent modifies ~/.zshrc or ~/.ssh/authorized_keys .
GhostApproval: A Trust Boundary Gap in AI Coding AssistantsA security boundary was crossed
The agent happily followed a symlink pointing outside the workspace and wrote to the target file.
GhostApproval: A Trust Boundary Gap in AI Coding AssistantsSensitive data was reachable
We created a symlink from config.json to ~/fake_aws_credentials .
GhostApproval: A Trust Boundary Gap in AI Coding AssistantsThe system acted without authorisation
More critically, Augment also followed symlinks for write operations - silently.
GhostApproval: A Trust Boundary Gap in AI Coding Assistants
Checked, and did not happen
Nothing here was checked and ruled out.
Not confirmed
Nothing is outstanding.
Where this came from
The original report
GhostApproval: A Trust Boundary Gap in AI Coding AssistantsA structured incident database · read September 18, 2026
AI Failure Index (CC-BY-4.0 data fields)Where we first saw it · read September 19, 2026
GhostApproval: six AI coding assistants followed hidden symlinks behind harmless approval promptsWhere we first saw it
GhostApproval: A Trust Boundary Gap in AI Coding AssistantsWhere we first saw it
Bug in top AI coding agents shows that Unix-era security headaches never really dieWhere we first saw it
'GhostApproval' technique leads AI coding tools to alter files outside of sandbox
The reports above are other people’s. The levels are ours, worked out under our published rules. No organisation that reported this assigned our number, and none of it is legal advice.
Something here is wrong? Tell us. Send us a source we can open, and a person will read it.