Platform Governance
How DropOps controls what AI can do on your systems.
Human-in-the-Loop
All state-changing operations require your explicit approval before execution.
Auto-Approved (Read-Only)
- File reads
- Directory listings
- System status checks
- Log viewing
- Cloud Operator:
sts:GetCallerIdentity,iam:SimulatePrincipalPolicy
Requires Approval
- File writes, modifications, deletions
- Package installations
- Service start/stop/restart
- Configuration changes
- Any destructive operation
AI Risk Analysis
Every command is analyzed before execution. The AI classifies risk and explains what each command does.
HIGH risk commands include semantic approval prompts explaining the danger in plain language. Analysis failures default to HIGH risk.
File Operation Safety
Automatic blocking for dangerous file operations:
- System File Protection - Blocks modification of
/etc/,/usr/,/sys/,/boot/ - Git Status Checks - Blocks destructive operations when repository has uncommitted changes
- Backup Requirements - Ensures rollback capability for deletions
Unsafe operations are rejected before reaching user approval.
Cloud Operator: Zero Standing Privileges
The Cloud Operator launches with zero AWS access. Permissions are granted just-in-time through conversation.
Cloud Operator Role (Minimal Start)
Intent-Based Permissions: The AI asks in plain English:
- "Should I have permission to see other EC2 instances?" →
ec2:Describe* - "Should I be able to read from your S3 buckets?" →
s3:GetObject,s3:ListBucket - "Do you want me to manage Terraform state?" → S3 + DynamoDB for tfstate
You answer Yes or No. The AI translates intent to IAM policy. Revoke any permission anytime through conversation.
Safety Controls
Implemented
- Emergency stop (immediate halt)
- Manual approval gates for all changes
- AI risk analysis per command
- System file protection
- Permission boundary (Cloud Operator)
In Development
- Operation rate limits
- Rollback capabilities
- Change windows
- Configurable approval thresholds
Audit Trail
Every action is logged with complete context:
- What was requested (user instruction)
- What the AI proposed (command + reasoning)
- Approval decision (approved/denied + timestamp)
- Execution result (stdout, stderr, exit code)
- Permission changes (Cloud Operator IAM modifications)
You can delete conversations from your dashboard. Immutable compliance logs are maintained separately following industry-standard data retention practices.