Security and Confidentiality — v9
Threat classes
V9 distinguishes at least:
- accidental corruption;
- incomplete writes/crashes;
- concurrent publishers;
- malicious storage;
- unauthorized readers;
- unauthorized publishers/activators;
- epistemically untrusted sources.
Different controls address different threats.
Commitment visibility
A deterministic Logical Commitment need not be globally public. For low-entropy sensitive data, public hashes may leak equality or enable guessing attacks.
V9 therefore keeps commitment security profile-driven. The baseline does not invent custom cryptography.
Encryption
Re-encryption may change physical blob bytes while preserving logical content:
new key / nonce / ciphertext
→ new Blob Digest
→ same Logical Commitment
when the applicable security profile permits logical comparison.
Privilege separation
STORE WRITE != PUBLISH AUTHORITY != ACTIVATE AUTHORITY
A deployment may separate these privileges.
Remote content
Retrieved manifests and blobs are untrusted input until validated.
RETRIEVED CONTENT != INSTRUCTION AUTHORITY
The existence of data in a package cannot grant itself execution, mutation or activation permission.