# BAAs and DPAs when ciphertext has no host keys

This is **guidance for an evidence pack**, not legal advice.

## The operator statement

Data at rest lives in the **encrypted Peergos cryptree**. Restore uses
**distributed, hash-verified, encrypted IPFS blocks**. Hosts and backup
media see **opaque blobs** (and random CHAMP labels), not filenames, not
PHI, not social graph — if keys never leave the client / customer HSM.

That is **true of the Peergos design** (see `PEERGOS-VERIFICATION.md`).
It **changes the BAA/DPA conversation**. It does **not** auto-delete the
need for contracts.

## HIPAA (US) — Business Associate

A Business Associate is a person who **creates, receives, maintains, or
transmits ePHI** for a covered entity.

**Technical argument you can make (and should document):**

1. ePHI is encrypted **before** it leaves the endpoint (Peergos client /
   DataCube writer).
2. The storage node, IPFS, disk, and backup hold **ciphertext + hashes**.
3. Private keys **do not** reside on the storage host (customer HSM /
   client). Hash verification detects tampering.
4. Therefore the host **cannot maintain ePHI in intelligible form**.

HHS has long distinguished **conduits** and parties that never have
access to ePHI. Many security programs still **sign a BAA anyway** when
a vendor touches infrastructure, because:

- misconfiguration could later expose keys or plaintext logs;
- support staff might be asked to handle a decrypted export;
- OCR looks at **your** risk analysis, not only crypto papers.

**Practical pack:** risk analysis memo citing cryptree + IPFS hash
checks + key custody (`HSM-CUSTODY` in system-git-sync); BAA **or**
written determination that the vendor is **not** a BA because they
cannot decrypt; list every party with disk/backup (ns1, IPFS cluster,
offsite replica).

## GDPR (EU/UK) — processor vs technical measure

Encrypted data can still be **personal data** if it is reasonably
attributable (usernames, IPs, invoice identity). GDPR **Art. 32**
explicitly lists encryption as a security measure.

**Technical argument:**

- Peergos is designed so the **server is an adversary** for content and
  most metadata.
- EU-based **independent pentests** (Cure53 DE, ROS NL) evaluated that
  design.
- Processing location for **self-host** is *your* ns1 / customer region,
  not peergos.net’s Germany hosting, unless you use peergos.net.

You may still need a **DPA (Art. 28)** with anyone who **processes**
personal data (even usernames, logs, billing). Ciphertext-only storage
**narrows** the DPA’s technical annex (they never see PHI bodies) but
rarely deletes the contract if they run the VM.

## NATS operators

NATS is an **untrusted broker**. Content is HPKE; destinations and
subjects are in the clear. A NATS operator:

- **cannot** read message bodies if they lack endpoint private keys;
- **can** see routing metadata (who talks to which inbox, sizes, timing);
- **can** drop/delay/copy **ciphertext**.

Treat NATS like a **transit provider**: DPA/BAA depends on whether
routing metadata is personal data in your jurisdiction. Document
passthrough + HPKE in the data-flow diagram.

## What “making yourself audit-ready” means here

You are **not** buying Peergos’s Cure53/ROS reports as *your* SOC 2.
You are **reusing** those public evaluations as **vendor/component
assurance** (like using an audited TLS library), then adding:

- your scope, RoPA, key-custody, inspect k-of-n, admin-history;
- evidence *your* instance ran those controls;
- contracts that match who can actually see plaintext.