Security
The trust review, decided in advance.
We don’t leave the security review until the last meeting of the deal. The trust layer is wrapped around the whole product — built so your security team can say yes the first time, and your CISO doesn’t kill it in the last meeting. Six commitments, in plain language.
- EU-hosted
- No model training
- ISO 27001certified
- ISO 42001certified
- SOC 2certified
- SSO + SCIM
Six commitments
What we promise. Not what we hope.
EU data residency by default
EU-hosted infrastructure. Your data never leaves European borders unless you ask it to. US residency available on request.
No model training on your data
Ever. Contractually. Audited. Your meetings, decisions, and plans are used to provide the service — and nothing else.
Permissioned by design
A project’s context is visible to a project’s people. AIs see what their user can see — permissions are a property of context, not a feature of chat.
SSO, SCIM, MFA
Okta, Azure AD, Google Workspace. SCIM provisioning. SSO/SAML enforced at Enterprise. MFA available on every tier.
Sub-processors, listed
A clear, current list of every third party that touches your data, and what they do with it. Updated when it changes — not when you ask.
Incident response, in writing
Notification SLAs, escalation paths, and the post-incident report you’ll get. Decided in advance, not in the hour.
Data flow
Where your meeting data lives, end to end.
Four stages, all inside the EU by default. Encrypted at rest and in transit. Retention windows you control.
-
01
Join
Notes joins the call as a participant. The host sees it in the participant list. The host can remove it at any time.
Your meeting platform · Zoom / Teams / Meet
-
02
Capture
Audio is transcribed in real time. Nine signal types extracted. Stored encrypted at rest (AES-256) and in transit (TLS 1.3).
EU servers
-
03
Surface
Decisions, actions, and signals routed to the relevant Living Plan. Permissioned to your workspace policy. Visible to the people who should see them.
Your workspace · permission-scoped
-
04
Retain
The transcript is treated as disposable. The decisions and commitments extracted from it are what the workspace keeps, so the context outlives the recording.
EU servers
The context cube
Every share is a point in three dimensions: access rights, sensitivity, decay.
Whether the reader is a person or an AI agent over MCP, In Parallel gives them just enough context and never too much. Eight settings place a share inside the cube; the platform enforces where it sits. A file has exactly one resolution — all of it. A share does not.
Access rights
Who may reach what
How deep — status, storylines, observations, or the meetings and documents beneath — about which projects, for whom, named or by role, and to do what with it: brief, read, forward, train, act. Nothing beyond the grant, including by following a link.
Sensitivity
What is confidential, by source and by policy
Classification comes from two places: inherited upward from the source material, and declared by company policy — HR threads, deal terms, client names, whatever legal says never leaves. Compiling can raise the floor; it can never lower it. Sources travel only as far as the grant allows.
Decay
How long it stays true
Two clocks. A page rebuilds as new material lands and carries the date it was last confirmed true. Every outside grant runs out unless someone renews it. Nothing goes stale quietly, and nothing stays open by accident.
Who sees what
Access follows context, not the chat window.
The access-rights axis, spelled out. A workspace’s context is visible to the workspace’s people — and a meeting to the people who were in it. Nothing is broadcast across workspace lines. The same rules bind every AI agent.
Resource
Workspace member
Meeting participant
Everyone else
-
A workspace’s plans, decisions & signals
——A workspace’s plans, decisions & signals
- Workspace member Visible
- Meeting participant Hidden
- Everyone else Hidden
-
Meeting records & transcripts in a workspace
——Meeting records & transcripts in a workspace
- Workspace member Visible
- Meeting participant Hidden
- Everyone else Hidden
-
A meeting you personally attended
—A meeting you personally attended
- Workspace member Visible
- Meeting participant Visible
- Everyone else Hidden
-
Any workspace you don’t belong to
———Any workspace you don’t belong to
- Workspace member Hidden
- Meeting participant Hidden
- Everyone else Hidden
An AI agent sees exactly what the person it acts for can see — the same rows, the same columns, never more. How AI access works ↓
AI access model
Four things that cannot happen.
An AI agent acting on In Parallel context holds the same grant as the person it acts for — the same point in the cube, re-checked on every call. These four rules are enforced by the platform, written as tests before the features were built, not promised in a policy.
Nothing is shared wider than its sources
Classification is inherited upward as material is compiled — from the source material, and from what company policy marks confidential. A page built from restricted material cannot be shared at a lower classification. Compiling can raise the floor; it can never lower it.
Access expires unless someone renews it
Every outside grant carries an end date by default. Keeping one open indefinitely is an explicit, logged, administrator-level choice — not what happens when nobody thinks about it.
Revoking access is immediate — for AI too
No cached view outlives its grant, and an AI agent re-checks the grant on every call. Withdrawing access is a state change, not a request that propagates when it gets around to it.
Your data never trains a model
Not ours, and not the model your team connects. An agent reads the context its grant allows and nothing else; what it reads is not used to train anything.
Incident response
Targets, written down before they’re needed.
Detection, notification, containment, and the post-incident report you’ll receive. The full incident response policy is on the trust portal.
-
Detection
< 15 min
24/7 monitoring on availability, integrity, and access anomalies.
-
Initial notification
< 4 hours
Direct email to your nominated security contact. No PR-managed silence.
-
Containment
< 24 hours
Targeted access revocation, key rotation, or service pause as scoped.
-
Post-incident report
< 5 business days
Root cause, scope of impact, remediation, prevention. PDF.
Want a sample post-incident report? Ask — we’ll send one redacted to a previous (resolved) event.
Certifications & standards
Audited, attested, current.
SOC 2 Type II
Certified. Report available on request.
ISO 27001
Information security management. Certified.
ISO 42001
AI management system. Certified.
GDPR-ready
DPA out of the box. DPIA documentation on request.
Sub-processors
The third parties that touch your data.
A representative selection. Updated when it changes, not when you ask. The full, current list lives on the trust portal.
-
Cloudflare
CDN, DDoS protection, bot management
EU edge
-
Hosting provider
Compute and storage (EU region)
EU only
-
Transcription / LLM provider
Speech-to-text and signal extraction (zero training on your data)
EU · Frankfurt
-
Email delivery
Transactional email (DPA in place)
EU
-
Observability
Logs and metrics for service uptime
EU
Talk to security.
We’ll send the SOC 2 letter, the DPA, the sub-processor list, and walk your team through any concern they have.