Donna handles phone calls, which means she handles some of the most personal things people say. We take the security of the service, and of the people on both ends of every call, seriously. This page explains how we protect Donna and how to tell us if you find a problem.
Reporting a vulnerability
If you believe you've found a security vulnerability in Donna, please email [email protected]. Please include:
- what you found and where: the URL, API endpoint, app screen or app version;
- step-by-step instructions to reproduce it, with any proof-of-concept code, requests or screenshots;
- what an attacker could do with it, as you understand it; and
- how you'd like to be credited, if at all.
One clear report is more useful than many automated ones. You can write in English. If you'd rather not be identified, you can report anonymously.
Please don't report vulnerabilities through app store reviews, social media or public issue trackers.
What to expect from us
- Acknowledgement within 3 business days.
- An initial assessment within 10 business days, telling you whether we can reproduce the issue and how serious we think it is.
- Updates at least every two weeks while we work on a fix, and a note when it ships.
- Credit, if you want it, once the issue is fixed.
We aim to fix critical issues within 7 days and high-severity issues within 30 days. We'll tell you if a fix will take longer, and why.
We don't run a paid bug bounty at the moment. We're grateful for every good report, and we're happy to credit researchers publicly.
Safe harbor
We won't pursue or support legal action against you for security research that follows this policy. If you act in good faith and within these rules, we consider your research authorized under the Computer Fraud and Abuse Act and similar laws, and under our terms of service, and we won't bring a claim against you for circumventing technical measures in the course of that research. If someone else brings legal action against you for research that followed this policy, we'll make it known that your work was authorized.
Good-faith research means you:
- only access, modify or keep data that belongs to you or to test accounts you control;
- stop as soon as you reach anyone else's data, such as call records, recordings, transcripts, phone numbers or account details, keep none of it, and tell us what you saw;
- avoid harming the service or its users, including through degraded performance, spam or data loss;
- give us reasonable time to fix a problem before you disclose it (see Disclosure); and
- don't use a vulnerability beyond what's needed to show it exists.
Scope
In scope:
- the Donna apps for iPhone and Android;
- the Donna service APIs the apps talk to, including sign-in;
- donna.lab798.com, including the waitlist.
Out of scope:
- placing calls to phone numbers you don't own or control, dialing premium-rate, international or emergency numbers, or anything else that could cost someone money or tie up a phone line. Donna places real calls on the real phone network, and calling premium-rate numbers through Donna will lead to your account being terminated;
- denial-of-service or load testing, and any automated scanning heavy enough to affect the service;
- social engineering, phishing or physical attacks against Lab798 staff, offices or users;
- attacks on other Donna users or their calls, or calling someone's Donna number to test it without their consent;
- vulnerabilities in third-party services we use, such as Apple, Google, OpenAI, Telnyx, Cloudflare or Amazon Web Services. Please report those to the vendor. If the issue is in how we use or configure them, it's in scope;
- reports that only list missing best-practice headers, software versions, or theoretical issues without a way to exploit them;
- self-XSS, clickjacking on pages without sensitive actions, and CSRF on logged-out forms;
- issues that require a rooted or jailbroken device, or physical access to an unlocked device.
Prompt injection and AI behavior. Donna talks to strangers, so we're especially interested in ways a caller or a called business can make Donna reveal a user's information, act beyond what the user allowed, agree to things outside a call plan, or pretend not to be an AI. Please report these as vulnerabilities, using only your own Donna account and numbers you control. Ordinary AI mistakes, such as a wrong summary or category, aren't security issues, but we'd still like to hear about them at [email protected].
Disclosure
We ask you to keep a vulnerability confidential until it's fixed, or for 90 days after your report, whichever comes first. If we need more time, we'll explain why and agree a date with you. We're happy to coordinate publication and review your write-up before it goes out.
How we protect Donna
This is a summary of how Donna is designed. It's not a complete list, and some parts will change as the service grows.
Sign-in and access
- You sign in with Apple or Google. We don't have passwords to leak.
- The beta is invite-only: only admitted accounts can sign in.
- The apps use short-lived access tokens (an hour or less), signed with keys we rotate. Longer-lived refresh tokens are stored only as one-way hashes, replaced every time they're used, and if one is ever reused, every token in its chain is revoked.
- Every API request is checked against the account that owns the call, and permissions are separate for reading calls, controlling calls, and placing calls.
Calls and data
- All connections between the apps, the website and the service use TLS.
- Call recordings are private. The app gets a signed link that expires after 15 minutes, and the link never passes through logs.
- Call records are deleted automatically after 30 days, and you can delete any call at any time.
- Your contacts never leave your phone. When you allow location access for placed calls, only your city and country are sent, never your precise location.
- Push notifications carry identifiers only. The app fetches caller details from us directly, so Apple and Google never see them.
- Our logs are designed to record identifiers and states, not what anyone said.
- Our AI provider, OpenAI, processes calls under zero data retention and doesn't train on them.
Calls Donna places
- Donna shows you a plan and dials only after you confirm it. She won't agree to anything beyond what the plan, or a reply from you, allows.
- Donna announces that she is an AI and that the call is recorded at the start of every call she places.
- Placing calls requires a separate permission on your account, and is subject to daily limits and destination restrictions that protect against misuse and toll fraud.
Infrastructure
- The service runs on Amazon Web Services in the United States, with data encrypted at rest and access limited to the few people who operate it.
- The website sends a strict Content Security Policy and no third-party trackers.
Contact
Security reports: [email protected]. Everything else: [email protected].
Our machine-readable contact file is at /.well-known/security.txt.