Policies
The catalog of rules Surface checks every AI tool against, and how to read it.
console.superalign.ai/policies
A policy is one rule that describes a risky situation for a kind of tool. Surface ships with about 200. SuperAlign writes and updates them, and scanners pick up new versions on their own. You can read them here but not switch them on or off.
Examples:
- An MCP server has a plain-text credential in its configuration.
- A browser extension asks for access to every website.
- An agent has a hook that runs shell commands.
Reading the page
| Column | Meaning |
|---|---|
| Policy | The rule’s name and a one-line description |
| Non-Compliant Instances | How many tool copies currently break this rule |
Click a rule for its full description, score, severity, the tool types it applies to, and the laptops currently breaking it.
Policy types
Rules are grouped by what they are about.
| Policy type | Categories |
|---|---|
| Capability | Code execution, privilege escalation, destruction, persistence, permissions |
| Exposure | Injection, exfiltration, secrets, network exposure, poisoning |
| Governance | Misconfiguration, supply chain, hardening, oversight evasion, sanctioning |
Each rule also carries:
- A severity: critical, high, medium, low, or info.
- A score from 1 to 9.5, used in the risk calculation.
- A risk group, so rules about the same problem count once.
- Where relevant, a CWE reference (a public list of common software weaknesses).
Rules for conversations
A smaller set of rules looks at recorded AI conversations instead of installed tools, for example a password pasted into a prompt. They only tag the session so you can find it on the Sessions page. They never stop a message from being sent.