Security
Security and data privacy: your clients’ records stay with your firm
Brennick holds medical records and privileged work product. This page says how that data is protected, who can see it, what reaches an AI model and what happens when you leave.
Your firm owns its data
Your firm owns its data. Brennick processes it only to provide the service, as the Data Processing Addendum describes, and never sells or shares it.
Encryption in transit and at rest
- Every connection to Brennick uses HTTPS with TLS 1.2 or 1.3. Plain HTTP only redirects to HTTPS.
- Brennick checks the database’s certificate on every connection to it.
- Stored files are encrypted at rest with a key Brennick manages. The file store refuses any request that is not encrypted in transit.
- Uploads go straight from the browser to the file store over HTTPS, on a short-lived link that fixes the file’s type and size.
- Sign-in tokens for services you connect, such as Clio and Google Calendar, are encrypted in the application before they reach the database, with a key kept outside it.
One firm never sees another
Every table that holds firm data records the firm it belongs to. Row-level security in the database decides what each request may read, and the application connects through a restricted database role that cannot bypass it.
Before each query, the server sets the signed-in firm for that transaction only. A query that forgets to filter by firm still returns only that firm’s rows.
No one writes to a table directly. Every change goes through server code or a database function that checks who is asking.
- Firm A staffsees only Firm A
- Firm B staffsees only Firm B
- Firm A rowsfirm id A
- Firm B rowsfirm id B
Who can see what
Your clients
A client signs in to the portal and sees only their own case status, messages, uploads and to-dos. Clients never see a timeline, a score, a damages range or any other AI output, and the portal’s queries cannot read those tables.
Your staff
Everyone at your firm can see your firm’s cases, and no one outside it can. Approving AI output, accepting or declining a case and exporting a demand take a licensed attorney, enforced in the database as well as on screen.
Brennick staff
Access to your data is limited to Brennick personnel who need it to operate, secure or support the service, and they are bound by confidentiality. When Brennick staff open a firm’s AI output for quality review, the access is written to that firm’s own audit log. Cases designated under 42 CFR Part 2 are withheld from that review entirely.
Email and text messages to clients carry no case details. They say there is an update or a document is needed, and point to the portal.
Every file is treated as untrusted
Before anything reads an uploaded PDF, Brennick rebuilds it without its metadata, scripts, annotations and attachments. Photos are rebuilt from their pixels, which drops location and camera data. Your original is kept as uploaded, unchanged.
Word files, email, spreadsheets and other formats are turned into a reading copy by a separate converter that has no route to the internet. Programs and scripts are refused, even when renamed.
Record text is fenced off from the instructions the AI model follows, and the model is told that anything inside a record is evidence, never an instruction.
What reaches an AI model
Brennick sends case text to its AI model provider under confidentiality terms. The provider may keep the text it receives for up to 30 days to operate and secure its service, then deletes it. Content it flags for a policy review may be kept longer under its own policy.
Brennick does not train models on your data, and its AI provider is barred by contract from training on it.
Only the timeline step reads the whole file. Later steps work from the timeline and the pages it cites, and records request letters are drafted from the patient’s name, the provider, the dates of service and your letterhead only. No model reads an intake submission.
Any staff member can turn AI off for a single case, for example when a client asks. While it is off, no page of that case goes to a model or to the scanned-page reader. Only a licensed attorney can turn it back on.
How well the model reads is measured on files with known answers and graded by code, not by another AI: see how accuracy is measured.
Substance use records under 42 CFR Part 2
Brennick looks for Part 2 records with fixed patterns first. A model can add a flag but never remove one, and a flagged case carries a banner a licensed attorney must acknowledge.
Before an attorney clears the case, no document in it reaches an AI model or the scanned-page reader. To clear it, the attorney records the client’s written consent and confirms that the recipients it names cover Brennick.
Bot protection and rate limits
Your public intake page and the contact form on this site check each visitor for bots, looking at the browser, not the contents of the form. Intake, sign-in and the other public routes are rate limited per address, and an IP address is turned into a one-way hash before it reaches the rate limiter.
An audit log nobody can edit
Every AI draft, every human review, every export and every change of status goes into an append-only audit log. The database refuses to change or delete an entry, for Brennick’s own accounts too. The only exceptions are the deletions described below, which leave a record without content.
When an attorney approves AI output, the model’s original output is kept beside the approved version, so your firm can show what a person changed.
Export and deletion
A firm admin can export all of your firm’s data at any time from Settings: the database records, every original file and the audit log.
After cancellation, your data is kept for 90 days so you can export it, and your admins get reminders. Then Brennick permanently deletes it, and deleted data leaves backups within 7 days.
A legal hold pauses deletion. A firm admin can also delete a single case at any time, unless a legal hold is in place.
A few records are kept after deletion with no case content, such as billing records. Section 5 of the Terms of Service lists each one and how long it is kept.
Sub-processors
The list of companies that process data so Brennick can run, with each one’s purpose and location, is available on request at privacy@brennick.ai.
Brennick gives your firm’s administrators at least 30 days’ notice by email before adding or replacing a sub-processor that handles your clients’ data, and your firm may object.
HIPAA
Brennick is not a HIPAA business associate and does not sign business associate agreements (BAAs). See section 5A of the Terms of Service.
A firm admin makes the firm’s HIPAA attestation when the firm signs up and accepts the Terms. Records your clients authorize their providers to release to you for their own claims are not affected.
Reporting a security issue
If you find a vulnerability, write to security@brennick.ai. Security questionnaires and vendor reviews go to the same address.
FAQs
Who at Brennick can see it?
Only Brennick personnel who need it to operate, secure or support the service, bound by confidentiality. When Brennick staff open your firm’s AI output for quality review, the access is written to your firm’s audit log, and 42 CFR Part 2 cases are withheld from that review.
Is our data used to train AI?
No. Brennick does not train models on your data, and its AI provider is barred by contract from training on it.
How long is it kept?
For the life of your subscription. After cancellation it is kept for 90 days so you can export it, then permanently deleted; deleted data leaves backups within 7 days.
How do we delete it?
A firm admin can delete a single case at any time, unless a legal hold is in place. Canceling the subscription starts the deletion of everything, after the export window above.
What reaches the AI provider?
Case text, under confidentiality terms; the provider may keep it for up to 30 days to operate and secure its service. Only the timeline step reads the whole file, no model reads an intake submission, and any staff member can turn AI off for a case.
Is there single sign-on?
Staff sign in with an email address and password, or with a Google account. SAML single sign-on is not offered today.
How are 42 CFR Part 2 records handled?
Fixed patterns flag them first, and a model can add a flag but never remove one. Nothing in a flagged case reaches an AI model until a licensed attorney records the client’s written consent, and a demand letter cannot be exported until that consent names its recipient.
How do we report a vulnerability?
Write to security@brennick.ai. Security questionnaires and vendor reviews go to the same address.
Bring your security review
Send your questionnaire to security@brennick.ai, and read the Data Processing Addendum your firm accepts.