Trust Center
Information Security Policy
The technical and operational controls SendTax uses to protect tax and identity information.
1. Purpose
This policy describes how SendTax protects the information it handles on behalf of tax filers and tax professionals. It sits under our Written Information Security Program (WISP), which describes the program as a whole, and is the umbrella for the topical policies linked in Section 9 (other than the WISP), which detail specific controls.
SendTax processes tax documents — IRS forms, financial statements, identity documents, and related taxpayer information. We treat this data with the standard of care expected by the FTC Safeguards Rule and IRS Publication 4557, and we have built our systems on that assumption from day one.
2. Scope
This policy applies to:
- All SendTax employees, contractors, and co-founders
- All systems, services, and infrastructure operated by SendTax
- All customer data, including documents uploaded by filers and accessed by tax professionals
- All third-party services (sub-processors) that handle SendTax data on our behalf
3. Roles and responsibilities
SendTax is a small company, operated by Howell & Gibbs LLC. Security responsibility is held by its two co-founders, who are the company's members, with one named owner accountable for the program overall.
- Qualified Individual (Liam Howell, Co-founder, Howell & Gibbs LLC): Designated under the FTC Safeguards Rule (WISP § 2). Owns this policy and the broader security program, and approves new versions of this policy for Howell & Gibbs LLC. Approves exceptions. Coordinates incident response. Reviews access at least quarterly.
- Co-founder (Holly Gibbs): With Liam Howell, as the company's members, oversees the program through the Qualified Individual's annual report (WISP § 11); approves an exception the Qualified Individual asks for, and acknowledges his privileged actions. Vendor reviews are the Qualified Individual's (Vendor Management Policy § 2).
- All personnel: Responsible for following every policy linked in Section 9, reporting suspected incidents to [email protected] immediately, and no later than one business day after noticing them, as the Incident Response Policy describes, and protecting credentials. Annual security-awareness training isn't in place yet (WISP § 6).
As SendTax grows, we will formalize a security committee and may appoint a dedicated security officer.
4. Information classification
SendTax handles four broad categories of information. Each is protected proportionally to its sensitivity.
| Class | Examples | Default handling |
|---|---|---|
| Restricted | Tax documents, SSNs, financial account info, identity documents | Encrypted at rest and in transit; RLS-enforced isolation; downloads and staff access logged |
| Confidential | User account data, firm/preparer records, internal product data | Encrypted at rest and in transit; access restricted to authorized roles |
| Internal | Operational documents, internal communications, source code | Access restricted to personnel; not for public distribution |
| Public | Marketing material, this policy, our Trust Center | No restrictions |
5. Security commitments
5.1 Access control
Access to SendTax systems follows the principle of least privilege. User authentication is brokered through our identity provider (Clerk, SOC 2 Type II), which supports multi-factor authentication. Two-step verification (MFA) isn't available yet; it switches on for pilot firms in December. It will cover tax-professional firm users (firm administrators and firm members) and SendTax administrators. Today no second factor is enabled, and nothing is enforced. Once it is on, it is enforced at sign-in (during the token exchange, not re-checked on every request), with a grace period to enroll before access is blocked. Customer data is isolated at the database layer using PostgreSQL Row-Level Security (RLS) — not application-layer filtering — so that a query run for one firm or filer returns only the data that account may see even when the endpoint's own query forgets to filter. It can't catch a request attributed to the wrong firm or filer (Access Control Policy § 4). Some system operations that must work across accounts, such as background maintenance, run under a separate database role that RLS does not restrict. Cross-tenant access (e.g., a tax professional accessing a filer's documents) is brokered through explicit, status-tracked link records, never through implicit ownership.
See: Access Control Policy.
5.2 Encryption
Customer documents are encrypted by SendTax as soon as they arrive, using AES-256-GCM with a separate key for each document; each document key is wrapped (envelope encryption) by a key encryption key (KEK) held in Google Cloud KMS. Until that step runs, the storage provider's own encryption at rest protects the file. Additional Authenticated Data (AAD) binds each ciphertext to its document and filer identifiers, preventing ciphertext substitution. Data in transit is protected by TLS 1.2 or higher on our application and API hosts and on the identity provider host that receives sign-in credentials, with TLS 1.3 negotiated where the client supports it. Our main site, send.tax, which serves our sign-in and sign-up pages, still accepts TLS 1.0 and 1.1 from clients that ask for them. HTTP Strict Transport Security is set in every Next.js app with max-age=63072000; includeSubDomains; preload (see apps/*/next.config.js) and is served on send.tax and on the app subdomains; send.tax is not yet on the browsers' HSTS preload list. Our API host sends the same HSTS header, and the other security headers, on every response of its own.
See: Encryption Policy.
5.3 Infrastructure security
SendTax runs on managed infrastructure providers. Each lists the third-party security attestations below on its own trust page:
- Fly.io (compute, managed Postgres, Redis) — SOC 2 Type II
- Cloudflare R2 (object storage) — SOC 2 Type II, ISO 27001, ISO 27701, PCI DSS Level 1
- Google Cloud KMS (key management) — SOC 1/2/3, ISO 27001/27017/27018/27701
- Clerk (identity and authentication) — SOC 2 Type II, CCPA-compliant
- Modal (ML inference) — SOC 2 Type II
- Sentry (error monitoring) — SOC 2 Type II, ISO 27001
We do not operate our own physical data centers and rely on these providers' physical, network, and platform security controls. The full list of sub-processors, including those handling transactional email and other supporting functions, is maintained at Sub-processors.
5.4 Secure development
Source code is managed in version control with access restricted to authorized contributors. Production deploys go through CI, a staging deploy and a single release workflow, as the Change Management Policy describes. A deploy outside that workflow is reserved for incident response and for rolling back a bad deploy, and is recorded as a privileged action. A documented break-glass procedure is for one case: during a Critical or High-severity incident, if the Qualified Individual can't be reached (or has delegated the deploy) and the normal release path isn't viable, the other operator may deploy a production fix. It covers deploys only, each use is recorded as a privileged action before the deploy, and its credentials have not yet been tested, so until they are, a fix may have to wait for the Qualified Individual (Business Continuity and Disaster Recovery Plan § 6.10). SendTax has one engineer, so most changes are merged by their author without a second human reviewer. Production secrets are kept in our hosting and compute providers' encrypted secret stores (Fly.io, Vercel and Modal) for runtime use, in GitHub Actions' encrypted secrets for CI and deploys, and in 1Password for operator access. Each service that handles customer data has separate staging and production credentials, and the provisioning scripts in st-backend/infra/lib/common.sh pick the credentials for the environment they set up. Secrets must not be committed to source control; a secret scanner checks every pull request and push to main (below). Dependabot (.github/dependabot.yml) opens monthly update pull requests for Python, JavaScript, GitHub Actions, and Docker base images. A dependency audit (pip-audit, pnpm audit) runs in the full CI run that starts the daily staging deploy and every production release; a critical advisory in a runtime dependency stops that deploy until it is fixed or accepted in writing with an expiry date. A secret scanner, Gitleaks (v8.30.1, pinned and checksum-verified), runs in CI on every pull request and every push to main (.github/workflows/secret-scan.yml), over the commits and the tree, and fails the check on a finding; the pre-commit hook in .pre-commit-config.yaml is optional, and GitHub's own secret scanning and push protection are not turned on for our repository (a paid feature for private repositories). Type checking and linting run in CI on every pull request that changes the code they check. Our source-control plan does not offer branch protection, so a failing check does not technically block a merge; merging only on green is an operator practice, not a technical control.
5.5 Vendor and sub-processor management
Every third-party service that processes customer data is reviewed for security posture before we adopt it and listed on our public Sub-processors page. Vercel and Google Workspace were in use before they were reviewed; we added them to our vendor inventory on 2026-10-01 and review both by the next annual review of Tier 1 vendors (WISP § 7). We prefer vendors with current SOC 2 or equivalent attestations.
See: Vendor Management Policy and our Sub-processors list.
5.6 Personnel security
All personnel agree to confidentiality obligations and acceptable use terms before being granted access to customer data. New personnel complete onboarding that covers data handling, credential hygiene, and acceptable use expectations. Devices used to access customer data meet the baseline in the Employee Acceptable Use Policy: full-disk encryption (FileVault on a Mac, BitLocker on Windows), a screen lock after no more than 10 minutes idle, automatic updates, and the built-in malware protection on. Today those are the members' own Mac computers (Mac Mini, MacBook Pro, MacBook Air); MDM management is not yet in place and remains on the roadmap, so each person sets and checks these settings on their own device and records a monthly check on our compliance dashboard. Personal accounts and personal devices not configured to these standards are not used for accessing customer data. Our training computer, which holds the training contributions people chose to share while we use them, is the one exception to Apple hardware: a company-owned Windows computer with BitLocker full-disk encryption, reachable only over our private network (WISP § 4.2). Access is provisioned based on role, reviewed at least quarterly, and revoked promptly upon role change or departure as part of a documented offboarding checklist.
See: Employee Acceptable Use Policy.
5.7 Incident response
SendTax maintains a defined process for detecting, responding to, and notifying affected parties of security incidents. We commit to notifying impacted customers without undue delay, in accordance with applicable law and our customer agreements.
See: Incident Response Policy.
5.8 Business continuity
Customer data in our database (accounts, profiles, document records, and the data extracted from documents) is backed up by our managed database provider, with point-in-time recovery. Document files are kept in Cloudflare R2, which stores them durably. R2 object versioning is not turned on and we keep no separate copy, so a document file deleted by the application or an operator can't be restored. Formal recovery time and recovery point objectives are under development and will be published when finalized.
See: Business Continuity and Disaster Recovery Plan.
5.9 Data retention and disposal
SendTax retains customer data only as long as needed to provide the service or as required by law, and disposes of it securely when no longer needed.
See: Data Retention and Deletion Policy.
6. Laws and guidance we work from
Our WISP goes through the FTC Safeguards Rule's required elements one by one, uses IRS Publication 4557, and says where we fall short today. The security program works from:
- FTC Safeguards Rule (16 CFR Part 314) — applicable to us as a handler of financial information
- IRS Publication 4557 — the safeguarding standard for taxpayer data, which our tax professional customers are also required to follow
- IRC § 7216 and its regulations — limits on using and disclosing tax return information; our Privacy Policy explains how we apply them
- New York SHIELD Act (N.Y. Gen. Bus. Law § 899-bb) — the "reasonable safeguards" standard (WISP § 13)
- State data protection laws applicable to our customers and users
No outside firm has audited SendTax's security or tested our systems, so there's no audit report to share. None is scheduled.
7. Exceptions
Any deviation from this policy requires written approval from the Qualified Individual, or from the other member of Howell & Gibbs LLC when the Qualified Individual is the one asking; no one approves their own. It is recorded with a defined expiration date. Exceptions are reviewed at least annually.
8. Enforcement
Violations of this policy may result in revocation of access, termination of employment or contract, and, where applicable, civil or criminal referral.
9. Related policies
- Access Control Policy
- Encryption Policy
- Incident Response Policy
- Data Retention and Deletion Policy
- Vendor Management Policy
- Sub-processors
- User Acceptable Use Policy
- Employee Acceptable Use Policy
- Business Continuity and Disaster Recovery Plan
- Change Management Policy
- SendTax Written Information Security Program (WISP)
10. Contact
| Security disclosures | [email protected] |
| Privacy inquiries | [email protected] |
| Operating entity | Howell & Gibbs LLC (operator of SendTax) |
11. Change history
| Version | Date | Author | Approved by | Changes |
|---|---|---|---|---|
| 1.1 | 2026-05-30 | Holly Gibbs | Not recorded | Initial publication (#101). Drafted in Notion as 1.0 (Holly Gibbs) and 1.1 (Liam Howell: verification text in section 5). Drafted with Claude. |
| 1.2 | 2026-09-24 | Liam Howell | Not recorded | § 5.1 corrected (#386): MFA for firm users and SendTax administrators is being switched on and is not enabled or enforced yet (the text had said MFA was available to all users and enforced at sign-in for these users). Drafted with Claude. |
| 1.3 | 2026-09-25 | Liam Howell | Not recorded | § 5.1 corrected (#416): MFA for firm users and SendTax administrators is not available yet: no second factor is enabled, and nothing is enforced (1.2 said it was being switched on). Drafted with Claude. |
| 1.4 | 2026-10-07 | Liam Howell | Liam Howell (QI), for Howell & Gibbs LLC | One version number and one change history for this policy's internal and public texts and its Notion page (WISP § 7.4; public WISP § 14); earlier versions were numbered separately (internal 1.0–1.2, public 1.1–1.3, Notion drafts 1.0–1.1). Earlier rows are re-dated to the day each version reached main, the only approval step before WISP § 7.4; they had carried draft or planned release dates. Also records a change made without a row: public § 5.1 (#155, 2026-06-08) changed the MFA statement to say MFA was enforced at sign-in; public 1.2 corrected it. Review cadence: at least annually (the public copy said semi-annual). Access reviews stay at least quarterly (public § 3 and § 5.6), as the Access Control Policy now also says. Public text: § 1, this policy sits under the WISP; § 3 and § 7, the Qualified Individual, not a "Security Lead", owns the program and approves new versions and exceptions, and the other member approves an exception the Qualified Individual asks for; Holly Gibbs oversees the program with Liam Howell as the company's members, in place of a share of "program decisions"; staff report suspected incidents immediately, and no later than one business day after noticing them, as the Incident Response Policy describes (the text said promptly). § 5.1 corrected: some system operations run under a database role that RLS does not restrict (the text said no query could return another tenant's data). § 5.2 corrected: documents are encrypted as soon as they arrive (the text said all documents were encrypted at rest), and our main site, send.tax, still accepts TLS 1.0 and 1.1 (the text said TLS 1.2 minimum); the HSTS note is brought up to date. § 5.4 corrected: most changes are merged by their author without a second reviewer (the text said changes are reviewed before production); production deploys go through the release workflow, a deploy outside it is reserved for incident response and for rolling back a bad deploy, and the break-glass path lets only the other operator deploy, only during a Critical or High-severity incident when the Qualified Individual can't be reached, covers deploys only, and its credentials have not yet been tested (the text said either operator could deploy from their own machine when CI was unavailable); GitHub secret scanning with push protection and Dependabot security alerts and updates were described as on and are off, and Gitleaks runs nowhere automatically, so secret scanning does not run automatically; a failing CI check does not block a merge on our plan (the text said it was a required check); the text now says where secrets are kept and how dependencies are checked. § 5.5 corrected: Vercel and Google Workspace were in use before they were reviewed (the text said every service is reviewed before adoption). § 5.6 corrected: with no device management, each person sets their own screen lock (the text said it was enforced). § 5.8 corrected: document files are not backed up (the text said customer documents were). § 6 adds IRC § 7216 and the New York SHIELD Act. § 9 adds Change Management and Sub-processors and corrects two titles. Internal text: § 1 takes the umbrella statement from the old header; § 3 corrects where plaintext may exist and where the Trust Center lives, and names the database role RLS does not restrict; § 4 corrects where the excluded segments are shown (the Trust Center does not list them); § 5 names the Qualified Individual and the members, and incident reporting matches the public text: immediately, and still no later than one business day, as INCIDENT_RESPONSE.md describes (it said within one business day). Drafted with Claude. |
| 1.5 | 2026-10-07 | Liam Howell | Liam Howell (QI), for Howell & Gibbs LLC | Clarified, public § 6: we follow the SOC 2 Trust Services Criteria in how we run our controls; we haven't had a SOC 2 audit, and have no plans for one until a client's requirements call for it (it said we would pursue attestation when customer demand and company stage warranted it). Internal § 4: the excluded segments are reviewed if SendTax achieves new attestations (it said when). Drafted with Claude. |
| 1.6 | 2026-10-16 | Liam Howell | Liam Howell (QI), for Howell & Gibbs LLC | Training contributions (#552–#556, #564). Internal § 4 item 4: plaintext is also permitted for a labeler in the cloud labeler (ACCESS_CONTROL.md § 5), and on the training computer, for the approved contributions it opened from a sealed export, at most 30 days after each export is built (ACCEPTABLE_USE.md § 3; WISP § 4.3); this applies only from publication. Public § 5.6: the training computer is the one device holding customer data that isn't Apple hardware: Windows, with BitLocker full-disk encryption, reachable only over our private network. Drafted with Claude. |
| 1.7 | 2026-10-21 | Liam Howell | Liam Howell (QI), for Howell & Gibbs LLC | Training contributions are labeled on the training computer (#581). Internal § 4 item 4: a training contribution's plaintext is permitted on the computer of a member labeling it in the training computer's labeler, which tells the browser to keep no copy, in place of a labeler in the cloud labeler, which is removed. Drafted with Claude. |
| 1.8 | 2026-10-22 | Liam Howell | Liam Howell (QI), for Howell & Gibbs LLC | Security statements, the Qualified Individual's decision of 2026-10-21: when SendTax describes itself it doesn't name SOC 2, its Trust Services Criteria or their numbers, and it says how it built its program and where it falls short instead of claiming to meet a standard. Public § 6 is now "Laws and guidance we work from". It no longer lists the SOC 2 Trust Services Criteria (it said we follow them in how we run our controls), and it says that the WISP goes through the FTC Safeguards Rule's required elements one by one, uses IRS Publication 4557 and says where we fall short today, and that no outside firm has audited our security or tested our systems and none is scheduled. Public § 5.1: two-step verification switches on for pilot firms in December (it said only that MFA is not available yet). Corrected, public text: § 3, annual security-awareness training isn't in place yet (it said personnel complete required security training); § 4, for Restricted data, downloads and staff access are logged (it said access is logged); § 5.3, each provider lists its attestations on its own trust page (it said each maintains current attestations); § 5.4, secrets must not be committed to source control, and nothing checks for them automatically today (it said secrets are never committed). Drafted with Claude. |
| 1.9 | 2026-10-26 | Liam Howell | Liam Howell (QI), for Howell & Gibbs LLC | The 2026-10-26 security and legal review. Corrected, public text: § 3 said the other member shares responsibility for vendor due diligence; the Qualified Individual does each vendor review (Vendor Management Policy § 2), and she oversees the program as a member, approves an exception he asks for and acknowledges his privileged actions. § 5.1 said row-level security returns only permitted rows even if the application code behind a query is wrong; it catches an endpoint's query that forgets to filter, not a request attributed to the wrong firm or filer, as the Access Control Policy already said. § 5.6 said devices are Apple hardware and Apple-issued; the baseline is the Employee Acceptable Use Policy's, met today by the members' own Mac computers and the Windows training computer, with a monthly self-check. § 5.4: a secret scanner (gitleaks, pinned) now runs on every pull request and push to main; GitHub's own secret scanning is still off. § 5.2: the API host sends HSTS and the other security headers on every response; the identity provider's host was probed on 2026-10-26 and refuses TLS 1.0 and 1.1, as the text said. Drafted with Claude. |