File size: 2,207 Bytes
ade765f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 | # CORTX Security Model
## Threat model
CORTX gives a language model the ability to rewrite its own operating rules.
The core risk: **the model (or a user manipulating the model via prompts)
escalates self-editing beyond policy text into something dangerous** β its
weights, its code, its identity, or credentials.
## Defense layers (all enforced in code, not prompt text)
1. **Whitelist** β only `constitution`, `tool_rules`, `metacognitive_rules`,
`refusal_rules`, `playbook` are editable. Anything else is rejected.
2. **Locked sections** β `security_rules`, `identity`, `weights`, `code`,
`admin_token`, `system_prompt` can never be edited through the channel.
Attempts are rejected *and* logged.
3. **Size caps** β 8,000 chars/section, 8 edits/turn.
4. **Rate limit** β 40 edits per rolling hour. Violations blocked and logged.
5. **Hash-chained audit log** β every edit and every rejected attempt is
appended as `hash = sha256(prev_hash + entry_json)`. Anyone can run
`audit.verify()` to prove the log is untampered.
6. **Approval mode** β admin can require human sign-off before edits go live.
7. **Rollback** β config is version-pinned; admin restores any prior version.
8. **Isolation** β the edit channel reaches only policy-text files. It has no
filesystem path capability, no code execution, no network access beyond the
config repository. Weights, code, and tokens live in a different trust zone
and are physically unreachable from the edit path.
## What the security model does NOT guarantee
- **Edit wisdom.** The boundary guarantees edits are *valid and bounded* β not
that they are *good*. A small model can write worse rules than it started
with. Review the audit log in production; that is why it exists.
- **Prompt-injection resistance of the base model.** The boundary is enforced
in code, so no prompt can cross it β but the model's *content* remains its
own responsibility.
## Verifying the audit log
```python
ok, n = editor.audit.verify()
assert ok, "audit log tampered at entry %s" % n
```
## Reporting
Security issues: open an issue on the repo or contact ChatPBC Labs.
|