# The Real Talk

*Security & privacy without the gloss · last updated 18 June 2026*

Most "enterprise security" copy is written to pass a checklist, not to protect you.
So here's the honest version — what actually moves the needle, what's mostly
theater (the stuff we do included), and exactly where we can and can't see your
data. If a claim here ever stops being true, it's a bug.

## What actually protects your data

These are the things that change the outcome if something goes wrong. Each one is
live, not a roadmap:

- **Encryption at rest with a key unique to your organisation — on by default.**
  Your meeting content and audio are encrypted in the database with your org's own
  key. This is application-level, not just "the disk is encrypted."
- **Our off-site backups hold ciphertext, not your meetings.** The master key that
  unlocks the data lives only in the live cluster's secret store — it is never
  written into a backup. A stolen backup is mathematically useless without it.
- **Crypto-shred erasure.** We can destroy your org's key — instantly making all of
  your encrypted data unreadable on live systems, and unrecoverable from backups
  once it's purged from the separate key store. "Deleted" means deleted.
- **EU hosting, EU jurisdiction** (Germany), with every sub-processor named on the
  [Data flows](/data-flows.md) page — including which ones can and can't use your
  data.
- **No meeting bot** joining your calls, **no training** on your meetings
  (contractual, not just a promise), **bring-your-own-keys**, and a **self-hosted**
  mode where your audio never reaches us at all.

## Where we draw the honest line

Here's the part most vendors won't print: **to turn your audio into a transcript
and a summary, a computer has to read it in the clear.** During that processing your
data is decrypted in memory — on our servers, and (unless you self-host or bring
your own keys) briefly at the AI provider that does the transcription/summarisation.
That means:

- We are **not "zero-knowledge" / not end-to-end encrypted** while running cloud AI.
  **Nobody doing cloud meeting-AI is**, no matter what the homepage says. Encryption
  protects your data *at rest* and *in backups* — not from a server that's actively
  processing it.
- If you need "the vendor genuinely cannot read it," the only real answers are
  running the AI on **your** hardware (self-host) or inside hardware-attested secure
  enclaves. A passphrase on top of a cloud service that still processes your data is
  not that.

## What's mostly theater — ours included

If a vendor (us included) leans on these, ask what they actually *do*:

- **"AES-256 encryption!"** — Fine, but the only questions that matter are *who
  holds the key* and *who can read it during processing*. The cipher is never the
  weak link.
- **Compliance badges (SOC 2, ISO, …)** — Evidence of *process*, not of whether the
  vendor can read your data. Useful, not the same as security.
- **"Customer-managed keys"** — Often the vendor still holds a copy and can decrypt
  anytime. Real control is the ability to *revoke* (and the vendor losing access).
- **"Zero-knowledge" on a cloud AI product** — Can't be true while a model reads
  your words to summarise them. We won't claim it.
- **"Bank-grade / military-grade"** — Marketing words with no technical meaning.
  Ignore them.

## The honest roadmap

Things we think are real, that we haven't built yet — and we'll tell you when we
have:

- **Confidential computing** (hardware-attested secure enclaves): the genuine way to
  process your data so even we can't read it in memory. Maturing fast; not something
  we run today.
- **Customer-revocable keys** for organisations that need to pull the plug on vendor
  access for incident response or compliance — pitched as exactly that, revocation,
  not "we can't see it."

## Questions to ask any vendor (including us)

- Can you read my data? If so, **when** — at rest, during processing, in backups?
- Where does my audio actually go, and which third parties touch it? (Ours:
  [Data flows](/data-flows.md).)
- If I leave, can you **prove** deletion — including backups?
- Is it *really* end-to-end encrypted during processing? (If it's cloud AI, the
  honest answer is no.)
- What happens to my data if your storage bucket leaks tomorrow?

If those answers are vague, that's the tell. Ours are on the [Privacy](/privacy.md)
and [Data flows](/data-flows.md) pages, and we'll answer anything else straight.

---

*Markdown edition for AI assistants — canonical page: [https://lexicanon.com/real-talk](https://lexicanon.com/real-talk) · Lexicanon.*