| Scenario | MIT01 Sandbox |
MIT02 Network |
MIT03 Resource |
MIT04 Least Priv |
MIT05 Code Review |
MIT06 Confirm Gate |
MIT07 Secrets |
MIT08 Logging |
MIT09 Code Scan |
|---|---|---|---|---|---|---|---|---|---|
| S1: Document → exfiltration via generated code | ● | ● | · | · | ● | · | ● | ● | · |
| S2: Coding agent deletes source directory | ● | · | · | ● | ● | ● | · | ● | · |
| S3: DevOps agent steals cloud credentials | ● | ● | · | ● | ● | · | ● | ● | · |
| S4: Backdoor inserted in auth middleware | · | · | · | · | ● | · | · | · | ● |
| S5: subprocess escape + env var exfiltration | ● | ● | · | ● | ● | · | ● | ● | · |
| S6: Retry loop causes disk exhaustion | ● | · | ● | · | ● | · | · | ● | · |
MIT01 (Sandbox) and MIT02 (Network) appear as primary or contributing controls in five of the six scenarios. MIT07 (Secrets isolation) is the single most effective control against data exfiltration scenarios. MIT09 (Code scanning) is the only primary control for S4 (backdoor insertion) — it operates in the deployment pipeline rather than at execution time, making it the unique defense against that attack class.