Draft · pending counsel review

Legal

How Daryl protects your data

Daryl holds calendars, health records, bills and family details, so security is part of the product. This page describes how we protect it, in terms that are true today.

Effective October 29, 2026Privacy Policy →

Summary

  • Built to SOC 2 and OWASP ASVS standards. SOC 2 Type II audit not started.
  • Your content is encrypted field by field and isolated to your account in the database itself.
  • Our people don’t read what you keep in Daryl, and the operator portal can’t show it.
  • Report a vulnerability to security@getdaryl.com. We reply within 3 business days.

This summary is not part of the policy. The full text below applies.

Our program

Daryl’s security program is written to the SOC 2 Trust Services Criteria (security, availability, processing integrity and confidentiality) and the OWASP Application Security Verification Standard. It is a self-assessment: a SOC 2 Type II audit has not started, and we will say so here until an independent firm is engaged. The program has a named owner, written policies reviewed at least once a year and after any incident, a risk register for the AI and its tools, and a record of accepted risks with the reason and a date to revisit each one.

Access to your data

  • Your sign-in. A single-use emailed link and a second factor, either an authenticator code or a passkey, on every sign-in. Sessions expire on their own, can be ended everywhere at once, and sensitive actions ask for your second factor again.
  • Your devices. Each phone, watch, computer or extension gets its own access, limited to what that kind of device needs. You can see and revoke each one, access expires and renews on its own, and Daryl tells you when a device is used from a new network or country.
  • Isolation. Every account’s data is walled off in the database itself, not only in the app, so a fault in one part of the code can’t hand one person’s data to another. Household sharing, caregivers and bots work through the same rules.
  • Our operators. The operator portal shows account details, plans, counts and spend, never your content. Access to production systems is limited to named people, and our policy requires multi-factor authentication on every system and an access review every quarter. The only content an operator can open is a safety report someone chose to send, and every opening is recorded.

Encryption

  • Every connection to Daryl is encrypted, and browsers are told to use only encrypted connections.
  • On top of the encryption our providers apply to their disks, the personal content you keep in Daryl is encrypted field by field with authenticated encryption before it is stored, and the database refuses to store those fields unencrypted.
  • A household’s shared lists, events and notes, and any model key it adds, are encrypted with a key that belongs to that household. Deleting the household destroys the key.
  • Encryption keys are kept apart from the data, limited to production, and rotated on a schedule and at once if exposure is suspected.

The AI and its actions

Daryl treats the AI model as untrusted and everything it reads from outside your account as data, never as instructions. Text from web pages, email, invitations and bots is marked as outside text and checked for hidden instructions; after Daryl reads it, deletions, notifications, scheduling and home controls are held for you to review, and the most sensitive actions run only from your own live request. In conversation Daryl has no tool that moves money or sends messages in your name; payments during an errand, and changes to locks, alarms and cameras, run only in your own live request. Secrets are never placed in what the model sees, and answers and notifications are filtered so keys, card numbers and similar values are never shown. Spending and activity limits stop runaway use. These rules are covered by an automated suite of attacks that is part of our release checks.

Errands on websites run on your own devices or in an isolated browser whose session is discarded after each task, limited to the sites you allow, and never type into password, card or one-time-code fields.

Building and changing Daryl

  • Every change goes through a pull request with automated checks, including static analysis, secret scanning and security tests.
  • Changes to sign-in, encryption, data isolation and the database get extra testing and a written way to roll back.
  • Dependencies are monitored. Security updates are applied within 7 days, and critical ones within 48 hours.
  • Database changes are recorded with who applied them and when, and are never edited after they ship.

Monitoring and records

Security events raise alerts to the people on call, and logs never contain passwords, tokens or the content of your records. Every change to your data and every sign-in is written to an activity log that can only be added to. Each day’s log is sealed with a cryptographic fingerprint chained to the day before, so any later alteration would show. You can see your own activity in the app.

Availability and backups

The basics (timers, alarms, lists and reminders) work without any AI model, so they keep working during an AI outage. The database keeps point-in-time recovery, and our backup plan adds nightly encrypted copies, kept 35 days, that can be decrypted only with keys held outside our systems. Our targets are to lose no more than 15 minutes of data and to restore service within 4 hours after a major failure; these are targets, not guarantees. Our plan includes regular restore tests. Every incident, including small ones, is posted on our status page.

Our providers

Before we use a provider that will hold your data, we check that it has an independent security audit, multi-factor access and encryption at rest, and we review those audits every year. Providers process data only to provide their service to us, under contracts that require them to protect it. The kinds of providers we use, and what each receives, are in the Privacy Policy.

Incidents

We have a written plan to detect, contain, fix and learn from security incidents, including the switches to stop Daryl acting for one person, one household or everyone at once. If an incident affects your personal information, we tell you and the authorities as the law requires, and without unreasonable delay, with what happened, what it means for you and what we are doing.

Reporting a vulnerability

Email security@getdaryl.com with steps to reproduce. We acknowledge reports within 3 business days and give a status update within 10. Please don’t open a public issue, and give us a reasonable time to fix the problem before you disclose it. The scope is in our security.txt. Denial-of-service, social engineering, findings that need an already compromised device, and scanner output without a working proof of concept are out of scope.

Test only against your own account, stop as soon as you reach data that isn’t yours, and don’t keep or share it. If you follow these rules in good faith, we won’t pursue legal action against you for your research.

What we don’t claim

Daryl is not SOC 2 audited yet and is not a HIPAA covered entity. No system is perfectly secure, and we won’t describe ours in terms we can’t back. If a statement on this page stops being true, we change the page.