Trust Center

Written Information Security Program (WISP)

Our written security program: it goes through the FTC Safeguards Rule’s required elements one by one, uses IRS Publication 4557, and says where we fall short today.

Version: 1.12Effective: 2026-10-26Owner: Howell & Gibbs LLCDownload PDF

This Written Information Security Program ("WISP") describes the administrative, technical, and physical safeguards SendTax maintains to protect customer information and is the program referenced in our Privacy Policy.

This WISP goes through the FTC Safeguards Rule's required elements one by one, uses IRS Publication 4557, and says where we fall short today. It works from:

  • The FTC Standards for Safeguarding Customer Information ("Safeguards Rule"), 16 CFR Part 314 (as amended).
  • IRS Publication 4557 ("Safeguarding Taxpayer Data") and the related Publication 5708 WISP guidance.
  • IRS Publication 1345 ("Handbook for Authorized IRS e-file Providers") to the extent it applies to tax-document workflows.
  • The New York SHIELD Act "reasonable safeguards" standard (N.Y. Gen. Bus. Law § 899-bb).

We publish this WISP on our Trust Center. Our Information Security Policy is a shorter summary of the same program. The internal records this WISP refers to, such as the annual report (§ 11), may be shared with auditors, regulators, enterprise customers under NDA, and the IRS on request.

Items marked Roadmap are not yet in production. The WISP names them explicitly rather than allowing them to read as current controls.


1. Scope and applicability

This WISP covers:

  • Customer information as defined by the Safeguards Rule: any record containing nonpublic personal information about a consumer of SendTax services, in any form. This includes taxpayer identity information, financial account information, tax-return data, and supporting documents.
  • Customer information systems: every SendTax-operated system that processes, stores, or transmits customer information, including application code, databases, document storage, the operator endpoints used to administer them, and the computer we train our document-reading software on.
  • Personnel: SendTax operators, employees, and contractors of Howell & Gibbs LLC with any access to customer information or customer information systems.
  • Service providers: every Tier 1 sub-processor listed on our Sub-processors page.

It does not cover information that is not "customer information" under the Safeguards Rule (e.g., public-website telemetry stripped of identifiers, internal corporate records of Howell & Gibbs LLC unrelated to customer information).


2. Designation of a Qualified Individual

The Safeguards Rule (16 CFR § 314.4(a)) requires designation of a Qualified Individual responsible for overseeing, implementing, and enforcing this program.

By written consent of its members, Howell & Gibbs LLC designates Liam Howell, Co-founder, Howell & Gibbs LLC, as the Qualified Individual responsible for overseeing, implementing, and enforcing this information security program. The members of Howell & Gibbs LLC are the governing body the Safeguards Rule refers to. By the same consent they delegate to the Qualified Individual the authority described below, and they ratify the approval of earlier versions of this WISP and of its companion documents (§ 15).

The Qualified Individual:

  • Has overall responsibility for the design, implementation, and ongoing operation of this WISP.
  • Approves new versions of this WISP and of the companion documents in § 15, for Howell & Gibbs LLC (§ 14).
  • Has authority to make decisions about safeguards, including decisions to suspend services or accept risk during incidents, and approves exceptions to the companion policies (§ 15) where a policy provides for one. An exception the Qualified Individual asks for is approved by the other member of Howell & Gibbs LLC; no one approves their own.
  • Provides a written report to the members of Howell & Gibbs LLC at least annually, summarizing the status of this program, material risks, the results of testing, and any recommended changes (see § 11). The members oversee the program through this report.
  • Delegates day-to-day execution as appropriate but remains accountable.

For absences exceeding two business days, the Qualified Individual designates a delegate in writing.


3. Risk assessment

The Safeguards Rule (16 CFR § 314.4(b)) requires a written risk assessment.

3.1 Scope of the assessment

The risk assessment evaluates reasonably foreseeable internal and external risks to the security, confidentiality, and integrity of customer information that could result in unauthorized disclosure, misuse, alteration, destruction, or other compromise.

3.2 Method

For each identified risk, the assessment records:

  • The nature of the risk (threat).
  • The systems or data affected (assets).
  • The likelihood and impact, each rated 1 to 5, and their product before and after the controls (the method the internal text sets out).
  • The current safeguards that address the risk.
  • Any residual risk and the planned remediation.

3.3 Cadence

The risk assessment is reviewed and updated:

  • At least annually.
  • Whenever a material change occurs in operations, systems, or the threat landscape.
  • Following any Critical-severity incident.

3.4 Current status

A full written risk assessment in the form above has not been approved yet. (Roadmap.) What exists today:

  • Two dated assessments of material changes, for dependents' information (2026-10-06) and for training contributions shared as they are (2026-10-22), each scoring likelihood and impact on a 1–5 scale before and after the controls, with the residual risk named. They are in the internal text of this WISP (§ 3.1).
  • A draft of the first full assessment, written for the security and legal review of 2026-10-26, is with the Qualified Individual for approval. When he approves it, this WISP will cite its date; it is due before the first annual report to the governing body (§ 11), before 2027-05-21.

SendTax holds customer information on fewer than 5,000 consumers, so under 16 CFR 314.6 the Safeguards Rule does not require of it the written-with-criteria assessment of 314.4(b)(1), the penetration testing and vulnerability assessments of 314.4(d)(2), the written incident response plan of 314.4(h) or the annual report of 314.4(i). It still requires that the program rest on a risk assessment, and every other element below. We keep all four of the lifted elements anyway, and say in each section where we fall short of them.


4. Safeguards

The Safeguards Rule (16 CFR § 314.4(c)) requires specific administrative, technical, and physical safeguards. The following subsections take each required element in turn: what the rule asks, what SendTax has in place, and what it doesn't have yet.

4.1 Access controls (314.4(c)(1))

Requirement: Implement and periodically review access controls to authenticate and permit access only to authorized users.

SendTax controls: Three-layer authorization (identity provider → application layer → PostgreSQL Row-Level Security with FORCE ROW LEVEL SECURITY), documented in the Access Control Policy. Background processing and a few server-side steps run under a database role that row-level security doesn't restrict, and on those paths access is checked in code (Access Control Policy § 4). The Qualified Individual reviews internal access at least quarterly (Access Control Policy § 9).

4.2 Inventory of data, systems, and personnel (314.4(c)(2))

Requirement: Identify and manage the data, personnel, devices, systems, and facilities that enable the organization to achieve business purposes.

SendTax controls:

  • Data inventory. Categories and retention windows are documented in the Data Retention and Deletion Policy.
  • Systems inventory. The application architecture is documented in the monorepo CLAUDE.md and reflected in source-controlled fly.toml configurations.
  • Sub-processor inventory. Maintained on the Sub-processors page and as an internal vendor inventory per the Vendor Management Policy.
  • Personnel inventory. Maintained internally; the Qualified Individual keeps the list current with the small operator population.
  • Device inventory. Each operator's primary device is recorded in the personnel inventory along with full-disk encryption status. (Roadmap: a structured device-inventory record for each named operator.) Our training computer, where we train our document-reading software, is recorded there too, with its full-disk encryption (BitLocker): a company-owned computer in the United States, reachable only over our private network (Tailscale). It receives approved training contributions only as exports sealed to its own key, and deletes them 30 days after each export is built (Data Retention and Deletion Policy § 3.4). They are labeled only there, in a labeling tool our staff open from their own computers over that network: it admits only the staff it names, shows only those pages, tells the browser to keep no copy, uses no AI service, reports nothing off that computer, and records each page view and each label change on the training computer. Our Qualified Individual approved the consent's text for a family pilot while our counsel reviews it: at most ten people, each invited by the Qualified Individual, signing through December 21, 2026. Software trained beside pages shared under the pilot stays on the training computer until counsel approves the text they were shared under.

4.3 Encryption (314.4(c)(3))

Requirement: Protect by encryption all customer information held or transmitted by the organization both in transit over external networks and at rest.

SendTax controls: Implemented, with the exceptions below, and documented in the Encryption Policy:

  • In transit: TLS on all customer-facing traffic. Our application and API hosts accept only TLS 1.2 or higher. The Cloudflare edge for send.tax and Cloudflare R2's storage endpoint still accept TLS 1.0 and 1.1 from old clients that ask for them (Encryption Policy § 6.1). (Roadmap: TLS 1.2 or higher at the send.tax edge.) Our services reach the database and job queue over a private network, and call our providers over TLS 1.2 or higher (Encryption Policy § 6.2).
  • At rest: AES-256-GCM envelope encryption with per-document DEKs, KEK custody in Google Cloud KMS, AAD binding to document and filer identifiers. Database and object-storage layers provide additional encryption at rest. A training export is sealed to our training computer's own key (X25519, HKDF-SHA256 and AES-256-GCM), so neither our storage nor whoever downloads it can read it.

Our last check, on 2026-10-01, found no prohibited algorithm used for any security purpose in our source code. Two endpoints we rely on fall short of that list: the send.tax edge and the storage endpoint above accept TLS 1.0 and 1.1 and offer some older TLS 1.2 cipher suites that use HMAC-SHA-1 (Encryption Policy § 5 and § 11).

4.4 Secure development (314.4(c)(4))

Requirement: Adopt secure development practices for in-house developed applications and procedures for evaluating, assessing, or testing the security of externally developed applications.

SendTax controls: Documented in the Change Management Policy. What runs today, and the gaps:

  • Pre-commit hooks for format, type, and quality checks are defined in the repository; the hook installed day to day is advisory and does not block a commit (Change Management Policy § 5.3).
  • Continuous integration runs lint, type-check, unit tests, and schema-drift checks on every pull request that changes the code they check (Change Management Policy § 5.4). Integration tests and dependency audits run in a daily full run that gates the staging deploy. Our source-control plan does not support branch protection, so a failing check does not technically block a merge; merging only on green is an operator practice, not a technical control.
  • A change off the security surface merges once its CI checks pass, with no human review. A change to the security surface (encryption, RLS, authentication, audit logging, deploy workflows, our policies and more; Change Management Policy § 6.3) also gets an automated AI security review (except on pull requests Dependabot opens or that come from a fork) and needs the Qualified Individual's approval before it merges, recorded on the pull request. GitHub enforces neither rule on our plan. SendTax has one engineer, so no second person reviews these changes. (Roadmap: a second reviewer.)
  • Production releases run through one release workflow. It runs the full test suites on the release commit, deploys that commit to staging and runs end-to-end tests against it, and only then builds and deploys the same commit to production. The exceptions are in § 4.7.
  • External dependencies are checked for known vulnerabilities in the daily full CI run, on each day the code has changed, and before each production release, not on every pull request (§ 5).

4.5 Multi-factor authentication (314.4(c)(5))

Requirement: Implement multi-factor authentication for any individual accessing any information system, unless the Qualified Individual has approved in writing the use of reasonably equivalent or more secure access controls.

SendTax controls:

  • For operators (internal access). Our Access Control Policy (§ 3) requires MFA for the identity provider (Clerk), the hosting provider (Fly.io), the source-control provider (GitHub), the key-management provider (Google Cloud), the storage provider (Cloudflare), the secret-management provider (1Password), and every other Tier 1 vendor that supports it, through each provider's own MFA.
  • For customers (filers and tax-professional users). Two-step verification (MFA) for tax-professional firm users (firm role firm administrator or firm member) and for SendTax administrators isn't available yet; it switches on for pilot firms in December. No second factor is enabled in our identity provider (Clerk) today, so customers cannot enroll, and sign-in enforcement is off. When MFA becomes available, authenticator-app MFA with backup codes comes first and enforcement follows. The enforcement code is built: once switched on, it runs at sign-in (the /oidc-login token exchange, evaluated once per sign-in rather than on every authenticated request), with a grace window in which a not-yet-enrolled user is warned and allowed through before sign-in is blocked. Until then, customer sign-in relies on Clerk's password requirements, breach-password detection, and rate limiting. For filers, MFA will be optional. The per-client Preparer/Viewer designations are access levels on a filer link, distinct from the firm role that drives MFA.
  • Where that leaves 314.4(c)(5) today: it is met under neither branch. No second factor is available, and the Qualified Individual's written approval of reasonably equivalent controls, the Rule's alternative, is still an unsigned draft. The 2026-10-26 legal review's advice is to enroll a second factor for the two SendTax administrator accounts first, on the identity provider's paid plan, and to sign the approval for customers only until December.

4.6 Disposal of customer information (314.4(c)(6))

Requirement: Develop, implement, and maintain procedures for the secure disposal of customer information no later than two years after the last date the information is used.

SendTax controls: Documented in the Data Retention and Deletion Policy. The default retention is up to 7 years after the last activity on an account, the same period our public Privacy Policy states. The sweeps that will enforce it (ACCOUNT_RETENTION_DAYS / DOCUMENT_RETENTION_DAYS = 2,557 days) count from the day an account is closed and, for each document on a closed account, from its upload; nothing yet measures inactivity on an account that stays open (Data Retention and Deletion Policy § 3.1). (Roadmap.) The Rule allows retention past two years where the information is necessary for a legitimate business purpose. Business need: a firm needs the documents behind a return for as long as the IRS can still examine it or a taxpayer can still claim a refund, which is up to 7 years (IRS Publication 583, "How long to keep records"). Customers can have their information deleted sooner on request. The automated hard-deletion sweeps for soft-deleted accounts, expired documents, and identity records are implemented and run daily on the worker schedule; they currently operate in dry-run mode (logging would-purge candidates with a queryable audit row each) pending production enablement of their retention flags. Until they are enabled, nothing is deleted automatically: deletion happens on request, by hand. Cryptographic erasure via DEK destruction is available as a mechanism, per the Encryption Policy. Training contributions keep their own clock: deleted at once when the person withdraws them, at the end of their consent's 3-year term, and from our training computer 30 days after each export (Data Retention and Deletion Policy § 3.4).

4.7 Change management (314.4(c)(7))

Requirement: Adopt procedures for change management.

SendTax controls: Documented in the Change Management Policy. Changes reach our main branch through a pull request and CI; the one exception is the automated version bump of our tax-rules engine (Change Management Policy § 5.1). A change off the security surface merges on green CI; a change to the security surface needs the Qualified Individual's approval, recorded on the pull request (Change Management Policy §§ 6.2 and 6.3). GitHub enforces neither rule on our plan, so both are rules we follow and record, not technical gates. Routine production deploys run only through release.yml: the deploy workflows can be started by hand for staging only. Rolling back to an earlier image and the break-glass deploy are the documented exceptions, and each is recorded as a privileged action (Access Control Policy § 7).

4.8 Monitoring and logging (314.4(c)(8))

Requirement: Implement policies, procedures, and controls designed to monitor and log the activity of authorized users and detect unauthorized access or use of, or tampering with, customer information.

SendTax controls: In place, with the gaps named at the end:

  • Application audit log. A hardened audit_logs table records authentication events, role changes, administrative actions, sensitive-data access, and tenant-filer relationship changes. Documented in Access Control Policy § 8.
  • Application telemetry. Sentry captures errors and anomalies; payloads are scrubbed of known sensitive fields before transmission.
  • Authentication telemetry. Clerk records sign-in events, failed attempts, and suspicious patterns.
  • Infrastructure telemetry. Fly.io platform metrics, health checks, and deploy history.
  • Audit-log retention. At least seven years, and at least as long as the records they describe (Data Retention and Deletion Policy § 3.6).
  • What isn't in place. No central log store or security-monitoring system: application logs stay with each provider for as long as it keeps them (Fly.io about seven days, Sentry ninety), and we ship them nowhere. Nobody reviews the audit log routinely (Access Control Policy § 8), and nothing pages a person for a security event; Sentry's alert rule emails us. (Roadmap: ship logs to a store we control, and page on security events.)

5. Regular testing or monitoring (314.4(d))

Requirement: Regularly test or otherwise monitor the effectiveness of safeguards' key controls, systems, and procedures, including those to detect actual and attempted attacks on, or intrusions into, information systems. Where continuous monitoring is not used, annual penetration testing and vulnerability assessments at least every six months are required.

SendTax current posture:

  • Continuous monitoring of application errors and authentication anomalies via Sentry and Clerk.
  • Continuous testing of code-level safeguards via CI (lint, type, unit, integration, and selected E2E suites).
  • Automated dependency scanning for known vulnerabilities in the daily full CI run, on each day the code has changed (not on every pull request).
  • Continuous health monitoring of production services via Fly.io health checks.

Currently missing — Roadmap:

  • Tabletop exercise for representative incident scenarios, at least once a year. None has been run yet; the first is due by 2026-11-21. (Roadmap, also tracked in the Incident Response Policy § 13.)

Currently missing — not scheduled:

  • Third-party penetration test. We haven't had an external penetration test, and none is scheduled. We'll commission one when a client's requirements call for it.
  • Vulnerability assessment by a qualified external party. None has been done, and none is scheduled.

SendTax does not claim that this monitoring is the "continuous monitoring" that 314.4(d)(2) accepts in place of testing. The program does not include the annual penetration testing and six-monthly vulnerability assessments that 314.4(d)(2) describes, and neither is scheduled (above). The Qualified Individual has documented this gap.


6. Personnel security and training (314.4(e))

Requirement: Implement policies and procedures to ensure that personnel are able to enact the program. Provide security awareness training. Use qualified information-security personnel. Verify that key personnel maintain current knowledge of changing information-security threats.

SendTax controls:

  • Hiring. All personnel with access to customer information are employees or named contractors of Howell & Gibbs LLC and acknowledge the obligations of this WISP at onboarding.
  • Onboarding. Each new operator receives an orientation to this WISP and the companion Trust Center documents before being granted production access.
  • Training. The Qualified Individual maintains current knowledge through industry publications, FTC and IRS guidance updates, and security-vendor advisories.
  • Acceptable use. All operators are bound by the Employee Acceptable Use Policy, which applies stricter production-system handling rules than the customer-facing User Acceptable Use Policy.
  • Annual security-awareness training with documented completion records for all personnel is on our roadmap and will be in place before the first annual report to the governing body. (Roadmap.)

7. Service provider oversight (314.4(f))

Requirement: Oversee service providers by (1) taking reasonable steps to select and retain providers capable of maintaining appropriate safeguards; (2) requiring providers by contract to implement and maintain such safeguards; and (3) periodically assessing such providers based on the risk they present and the continued adequacy of their safeguards.

SendTax controls: Documented in the Vendor Management Policy. Tier 1 sub-processors are chosen against documented selection criteria and reviewed at least annually. Vercel and Google Workspace were in use before they were in our vendor inventory; we added them on 2026-10-01, and we review both against our pre-engagement checks by the next annual review of Tier 1 vendors. SendTax has not yet signed a Data Processing Agreement with every Tier 1 sub-processor; until it does, it relies on each vendor's standard terms. The current list of Tier 1 sub-processors, and where their DPA terms stand, is published at Sub-processors.


8. Evaluation and adjustment (314.4(g))

Requirement: Evaluate and adjust the information security program in light of the results of testing and monitoring; any material changes to operations or business arrangements; the results of risk assessments; or any other circumstances that may have a material impact on the program.

SendTax controls: This WISP and each companion policy carries an explicit review cadence (at least annually, and after any material change) and a list of triggers (material change, incident, regulatory change) for out-of-cycle review. The Qualified Individual is responsible for ensuring the program is updated when any trigger fires.

A combined annual review of every Trust Center document and the underlying internal evidence is conducted before each annual report to the governing body. Each review is recorded in the document's change history, even when it changes nothing (§ 14).


9. Incident response (314.4(h))

Requirement: Establish a written incident response plan designed to respond to and recover from any security event materially affecting the confidentiality, integrity, or availability of customer information in the organization's control.

SendTax controls: Documented in the Incident Response Policy, which addresses each required element below. The plan hasn't been tested in an exercise yet: the first tabletop exercise is due by 2026-11-21 (§ 5).

  • Goals. Stated at the start of the IR Policy, before § 1.
  • Internal processes for responding. IR Policy §§ 4–7.
  • Roles, responsibilities, levels of decision-making authority. IR Policy § 5.
  • External and internal communications and information sharing. IR Policy §§ 8–9, including statutory notification obligations (the FTC Safeguards Rule, U.S. state breach notification statutes, IRS tax-data obligations, and another country's law, such as the GDPR, where it applies).
  • Identification of requirements for remediation. IR Policy §§ 7.3–7.4.
  • Documentation and reporting regarding security events and related incident response activities. IR Policy §§ 10 and 11, backed by a real database-level incident-management system.
  • Evaluation and revision of the incident response plan following a security event. IR Policy §§ 11 and 15.

SendTax additionally commits to next-business-day incident reporting to the IRS for events affecting tax-data systems, as our Privacy Policy says. SendTax is not an Authorized IRS e-file Provider, so the commitment is voluntary (§ 12.2).


10. Notification of FTC and customers

Requirement (314.4(j), as amended 2023): Notify the FTC electronically as soon as possible, and no later than 30 days after discovery, of any notification event involving the information of 500 or more consumers.

SendTax control: Acknowledged and committed to in our Privacy Policy. Operational responsibility sits with the Communications Lead during incident response, who works with the Qualified Individual to confirm the threshold has been met and to file the required notification via the FTC's electronic reporting portal. The documented incident-management system retains a durable record of any such notification.

State breach notification statutes are tracked separately; notification is made on the most stringent applicable timeline, as described in Incident Response Policy § 9.


11. Annual report to the governing body (314.4(i))

Requirement: The Qualified Individual reports in writing, regularly and at least annually, to the board of directors or equivalent governing body on:

  • The overall status of the information security program and compliance with the Safeguards Rule.
  • Material matters related to the information security program, addressing issues such as risk assessment; risk management and control decisions; service provider arrangements; results of testing; security events or violations and management's responses; and recommendations for changes in the program.

SendTax control: The Qualified Individual produces a written Annual WISP Report addressed to the members of Howell & Gibbs LLC. The first report will be issued before 2027-05-21 and at least annually thereafter.

Each report covers, at minimum:

  • The status of each safeguard in § 4.
  • The results of testing and monitoring during the period (§ 5), including the results of any penetration test or vulnerability assessment.
  • The status of each item on the WISP roadmap, including any items that have shipped during the period and any new items added.
  • Material changes to operations, systems, or sub-processors.
  • A summary of any Critical or High-severity incidents during the period, with links to the incident records and any postmortems.
  • Each Trust Center document reviewed in the period, with the version reviewed and the review date.
  • Recommendations for changes to this WISP.

The completed annual report and the members' written acknowledgment are retained in the WISP's evidence archive.


12. Tax-data-specific overlay (IRS guidance)

In addition to the Safeguards Rule, we use the IRS's guidance for tax professionals, and say where SendTax falls short of it.

12.1 IRS Publications 4557 and 5708

This WISP uses IRS Publication 4557 and the WISP template in Publication 5708 ("Creating a Written Information Security Plan for Your Tax & Accounting Practice"). We don't keep a section-by-section mapping of Publication 5708 to this WISP.

12.2 IRS Publication 1345 (e-file ecosystem)

SendTax is not an Authorized IRS e-file Provider: it prepares, signs and transmits no return. Publication 1345's six security standards bind Authorized Providers (the first five, Online Providers), not SendTax, and the Privacy Policy no longer lists two of them as roadmap items. We use them as a reference and say where we stand:

Pub 1345 expectationStatus
Public-facing tax sites secured with an Extended Validation (EV) TLS certificateNot applicable to SendTax, and not planned: browsers no longer display EV certificates, and ours are domain-validated certificates managed by Cloudflare (send.tax) and Fly.io (the app hosts).
External vulnerability scans on a weekly cadenceNot applicable; not scheduled (§ 5).
U.S. domain registrationIn place. send.tax is registered to a U.S. entity.
Next-business-day incident reporting to the IRS for events affecting e-file or tax-data systemsA voluntary commitment (Incident Response Policy § 9); operational responsibility sits with the Communications Lead during incident response.
Retention of IRS-required e-file records (e.g., Forms 8878 and 8879) for three yearsIn place. See Data Retention and Deletion Policy § 3.2.

12.3 PTIN and Circular 230

Tax professionals using SendTax are required to hold a valid PTIN and comply with Treasury Circular 230. The PTIN obligation is enforced at the data-model level: the preparers.ptin column is required and stored encrypted, and a unique keyed hash of it (ptin_hmac) stops two preparer records from carrying the same PTIN. Circular 230 compliance is the responsibility of the individual preparer; SendTax facilitates but does not validate it. Both obligations are set out for users in the User Acceptable Use Policy § 5.


13. New York SHIELD Act

The New York SHIELD Act (N.Y. Gen. Bus. Law § 899-bb) requires "reasonable safeguards" administrative, technical, and physical. The safeguards in § 4 of this WISP and the practices in §§ 5–9 are what SendTax relies on for the SHIELD Act's "reasonable safeguards" standard, for any New York resident whose private information SendTax processes. The gaps those sections name apply here too.


14. Document control

FieldValue
Approval authorityQualified Individual, for Howell & Gibbs LLC, under authority delegated by its members
OversightMembers of Howell & Gibbs LLC, through the annual report (§ 11)
Storage locationst-apps/docs/trust-center/written-information-security-program.md in the SendTax monorepo is the master copy. The signed annual report is retained in a separate evidence archive.

From version 1.5, this WISP, like every policy on our Trust Center, has one version number and one set of dates on its page, in its PDF and, where the policy has one, in its internal text. Each version is dated the day the Qualified Individual approved it for Howell & Gibbs LLC, and takes effect that day. Two kinds of change are the exception: a change that reduces a commitment about how customer information is collected, used, shared, retained or protected applies only from the day the version is published here, plus any notice we owe, and the version's row says so. So does a change that adds to what users must or must not do (User Acceptable Use Policy § 10). A version that corrects a statement that was not true says so in its row, and takes effect on approval. An annual review that changes nothing is recorded as a row with the same version.

Change history

VersionDateAuthorApproved byChanges
1.02026-05-30Holly GibbsNot recordedInitial publication. Drafted with Claude.
1.12026-09-24Liam HowellNot recordedBrought in line with how SendTax runs today. § 4.4 and § 4.7: local hooks, CI and review as they actually run (one engineer, advisory hooks, no branch protection). § 4.5: MFA for firm accounts is being switched on, not yet enforced. § 4.6: 7-year retention with its business-need reason (the "two-year" figure matched neither the Privacy Policy nor the code). § 5: dependency scanning runs daily. Drafted with Claude.
1.22026-09-25Liam HowellNot recorded§ 4.5: MFA for firm accounts is not available yet; no second factor is enabled in Clerk, and nothing about it is scheduled. Drafted with Claude.
1.52026-10-07Liam HowellThe members (written consent of 2026-10-06) and Liam Howell (QI)One version number and one change history for this WISP's internal and public texts and its Notion page (internal § 7.4, public § 14); earlier versions were numbered separately (internal 1.0–1.4, public 1.0–1.2, Notion drafts 1.0–1.1). Earlier rows are re-dated to the day each version was merged into main; no approval was recorded for them. The public 1.1 and 1.2 rows had given planned release days, not merge days. Also records changes made without a row: #67 and #69 (2026-05-26), #93 (2026-05-30), #155 (2026-06-08), #233 (2026-06-16), #345 (2026-09-11), #358 (2026-09-17, which named the Qualified Individual). § 2: the members of Howell & Gibbs LLC designate the Qualified Individual by written consent, delegate to him the approval of versions and the safeguard decisions, oversee the program through the annual report, and ratify the approval of earlier versions; an exception the Qualified Individual asks for is approved by the other member. Approval authority: the Qualified Individual, for Howell & Gibbs LLC, under the members' written delegation (the public text said the governing body). Adds the rule for versions and dates (internal § 7.4, public § 14) and a review cadence of at least annually. Corrected, as not in place: a dated risk assessment (internal § 3; none is done yet, and the first is due before 2027-05-21), an annual third-party penetration test and access-control review (internal § 7.2), weekly dependency upgrades and a statement about logging (internal § 4.2), meeting the continuous-monitoring branch of 314.4(d) (public § 5), and a signed DPA with every Tier 1 sub-processor (public § 7). Also corrected: the public text is published, not internal-only (preamble); dependency audits run on each day the code has changed, not every day (public § 5); the certificate for send.tax is managed by Cloudflare, not Fly.io (public § 12.2). Internal § 4.4 adds Google Cloud KMS, Modal and Google Workspace, says Vercel hosts only www, lists Stripe Identity, Stripe payments and cloud-folder delivery as built but not switched on, and makes the Sub-processors page the list of record; internal § 5.3 offboarding covers every vendor in § 4.4; internal § 6 commits to next-business-day IRS reporting, as the public text does, adds the FTC notice for 500 or more consumers, and tracks incidents in the Security Incident system, each closed with a postmortem or a closing note (Incident Response Policy 1.1); internal § 2 lets the other operator act as incident commander when the Qualified Individual hands the role over in writing or can't be reached; internal §§ 2 and 4.1: the workforce is both members, and both members take the annual training, the first before 2027-05-21; none has been done yet (it said the Qualified Individual alone); public § 9 says where the Incident Response Policy states its goals. Public § 4.8: audit logs are kept at least seven years and at least as long as the records they describe (it said three years; Data Retention and Deletion Policy § 3.6), in the audit_logs table (it said audit_log). Access review: at least quarterly (internal § 5.2 and public § 4.1 said annual, with quarterly on the roadmap). Public § 4.4: CI's lint, type and unit checks run on pull requests that change the code they check. Public §§ 4.4 and 4.7 describe change review as the Change Management Policy does: a change to the security surface needs the Qualified Individual's approval, recorded on the pull request; other changes merge on green CI; GitHub enforces neither. Public § 4.5 and internal § 4.3 state operator MFA as the Access Control Policy's requirement (public § 4.5 said implemented in full). Corrected: the send.tax edge and R2's storage endpoint still accept TLS 1.0 and 1.1 and some older cipher suites (internal §§ 3.3 and 4.2, public § 4.3 said TLS 1.2 everywhere and no prohibited algorithm); the RLS-bypass role also serves migrations and a few API steps, which check access in code (internal §§ 1.2 and 4.2 said background processing only); production is staged from the same commit, not the same artifact, and a rollback to an earlier image or a break-glass deploy can reach production outside release.yml (public §§ 4.4 and 4.7); the 7-year sweeps count from an account's closure and a document's upload, and nothing yet measures inactivity (public § 4.6); Vercel and Google Workspace were not in the vendor inventory and have not been reviewed, and were added on 2026-10-01 (internal § 4.4, public § 7, which said every Tier 1 sub-processor is evaluated); no restore drill or tabletop exercise has run yet, and the first of each is due by 2026-11-21 (internal § 4.2, public § 5). Internal § 4.3 allows a local copy of customer information only as the Employee Acceptable Use Policy does; public § 9 lists notice rules as the Incident Response Policy does (it listed GDPR first; we operate only in the United States); public § 12.3: a keyed hash of the PTIN, not the PTIN column, is what must be unique. Approved only after the members sign their written consent (MEMBERS_CONSENT_QI.md). Drafted with Claude.
1.62026-10-07Liam HowellLiam Howell (QI), for Howell & Gibbs LLCInternal § 7.2, public §§ 5 and 11: no third-party penetration test, vulnerability assessment or review of access controls and RLS policies has been done, and none is scheduled (they were on the roadmap: an annual penetration test, six-monthly vulnerability assessments and an access-control review); we will commission a penetration test when a client's requirements call for one. This change applies from the day this version is published on send.tax. Internal § 7.2 names what runs instead: the RLS and PII invariant tests and the dependency audits. Drafted with Claude.
1.72026-10-07Liam HowellLiam Howell (QI), for Howell & Gibbs LLC§ 1.1 adds dependents' information (name, date of birth, relationship, citizenship or residency, and per-year claim status and answers, including disability and death status; no SSN or ITIN). § 3.1 records the risk assessment for that new data category (2026-10-06), including the preparer's narrow delete when it removes a return it uploaded and the daily sweep for graph dependents written during the deploy. The public text is unchanged. Drafted with Claude.
1.82026-10-16Liam HowellLiam Howell (QI), for Howell & Gibbs LLCThe training computer, where approved training contributions are used (#564). Internal § 4.3 (public § 4.2): a company-owned Windows computer in the United States, the one place outside our providers' services that holds customer information for longer than an operator's 7-day working copy: it receives contributions only as exports sealed to its own key, which the Qualified Individual asks for as a privileged action, and deletes them 30 days after each export is built; BitLocker on, reachable only over Tailscale, its key readable by its training account alone; recorded in the device inventory before the first export. This applies only from publication. Internal § 1.1 adds training contributions; § 4.4 adds Tailscale as an internal tool (corrected: in use, it was not listed); § 5.3 offboarding removes a person's devices from Tailscale. Public § 1 names the training computer among the systems covered, § 4.3 the export's seal and § 4.6 the contributions' own deletion clock. Drafted with Claude.
1.92026-10-21Liam HowellLiam Howell (QI), for Howell & Gibbs LLCTraining contributions are labeled on the training computer (#581). Internal § 4.3 (public § 4.2): approved contributions are labeled there and nowhere else, in a labeler a member opens from their own computer's browser over the company's Tailscale network (tailscale serve, to a labeler that listens on 127.0.0.1 only, at the computer's tailnet name, never published to the internet). It admits only the Tailscale logins it names, answers only the routes its labeling pages use, shows only the contributions' pages, sends every response no-store so the viewer's browser is told to keep no copy, calls no AI service and reports no errors off the computer, keeps a page's form and page function and no other value from it, and records each page view and each label change in an access log on the training computer, refusing a request it can't record. The viewer's computer meets the device baseline, and nothing from a page is kept on it. Drafted with Claude.
1.102026-10-22Liam HowellLiam Howell (QI), for Howell & Gibbs LLCSecurity statements, the Qualified Individual's decision of 2026-10-21: the WISP says how it was built and where SendTax falls short, instead of claiming to meet a standard. Public text: the opening says it goes through the FTC Safeguards Rule's required elements one by one, uses IRS Publication 4557 and says where we fall short today (it said it was designed to satisfy the requirements of the Safeguards Rule, Pubs 4557, 5708 and 1345 and the SHIELD Act); § 4 takes each required element in turn (it said it maps each one to the controls SendTax has implemented); § 12 says we use the IRS's guidance for tax professionals (it said SendTax aligns with it), and § 12.2 gives where SendTax stands on each Pub 1345 expectation (it said SendTax aligns with them); § 13 is "New York SHIELD Act" and says its safeguards are what SendTax relies on for the reasonable-safeguards standard, with the gaps §§ 4–9 name (it said they are designed to satisfy it). Public § 4.5, internal §§ 3 and 4.2: customer MFA switches on for pilot firms in December (16 CFR 314.4(c)(5)). Corrected, public text: §§ 4.1, 4.4 and 9 said "Implemented in full"; § 4.1 now names the paths row-level security doesn't restrict, § 4.4 lists its gaps, and § 9 says the plan hasn't been tested in an exercise yet. § 4.4 said external dependencies are scanned automatically; they are checked in the daily full CI run on days the code changed and before each release, not on every pull request. § 12.1 said a mapping of Publication 5708 to this WISP is kept and available on request; none exists. §§ 5 and 9 call the Privacy Policy's Pub. 1345 text a paragraph (they said an alignment statement). Drafted with Claude.
1.112026-10-22Liam HowellLiam Howell (QI), for Howell & Gibbs LLCTraining contributions are shared as they are, under the family pilot (consent v4, #601, 2026-10-22). § 3.1 records the risk assessment for this change (2026-10-22). Internal § 1.1 (dependents): a training contribution's pages must not name a dependent, and one that does is rejected when seen. Internal § 1.1: a contribution's pages are exactly as they are on the contributor's device, with their name, address, taxpayer identification number and account numbers on them (it said the device replaced their details). Internal § 4.3 (public § 4.2): the Qualified Individual approved the consent text for a family pilot while counsel reviews it (consents signed through 2026-12-21, by at most ten people the Qualified Individual invited), and a model trained beside a contribution shared under it stays on the training computer until counsel approves the text that contribution was signed under: each export marks those contributions, a training run's provenance counts them, and ml.tools.upload_model refuses to serve the model. Sharing pages with the person's details on them, where their device used to replace those details first, applies only from publication. Drafted with Claude.
1.122026-10-26Liam HowellLiam Howell (QI), for Howell & Gibbs LLCThe security and legal review of 2026-10-26 (docs/reviews/2026-10-26-security-and-legal-review.md). Internal § 3.1 and public § 3.4: SendTax holds information on fewer than 5,000 consumers, so 16 CFR 314.6 lifts 314.4(b)(1), (d)(2), (h) and (i) (the written-with-criteria assessment, penetration testing and vulnerability assessments, the written incident response plan and the annual report); it keeps all four anyway, and a draft of the first written risk assessment is with the Qualified Individual (docs/policies/drafts/RISK_ASSESSMENT_2026-10-26_DRAFT.md). Corrected, public §§ 3.2 and 3.4: two dated material-change assessments exist (2026-10-06 and 2026-10-22, internal § 3.1), scored on the 1–5 likelihood × impact scale the internal text sets out (the public text said none had been completed and described a four-step scale). Public § 4.5: 16 CFR 314.4(c)(5) is met under neither branch today: no second factor is available, and the written approval of equivalent controls is an unsigned draft. Public § 4.8 "Implemented" becomes what is in place and what isn't: no central log store or SIEM, application logs kept only as long as each provider keeps them (Fly.io about seven days, Sentry ninety), no routine audit-log review, and nothing that pages a person for a security event. Public §§ 5, 9 and 12.2: SendTax is not an Authorized IRS e-file Provider, so Publication 1345's standards for online providers (an EV certificate, weekly scans by an approved vendor) don't bind it and are no longer roadmap items; the next-business-day report to the IRS stays, as a voluntary commitment. Internal § 4.2: Cloudflare R2 offers no object versioning (it said versioning was deliberately not enabled), so the bucket is the one copy of each document; the API now sends HSTS and the other security headers on every response and serves its schema outside production only; staging and production refuse to boot unless Cloud KMS is reached through Workload Identity Federation, which the Fly secret names confirmed on 2026-10-26. Internal § 4.3: no training-computer key is pinned yet, so a production export is refused. Internal § 4.4 adds Turnstile to the Cloudflare row and the tools the review found in use but unlisted: Anthropic's Claude (code and the security review), OpenAI's API (the labeling tools, on public or members' own documents), Notion, Upstash behind Fly's Redis, Expo, Hugging Face and the dormant Typeform webhook; and notes that visitors' browsers fetch our web fonts from Google Fonts. Drafted with Claude.

15. Companion documents

This WISP is the umbrella program; the controls it describes are implemented and operationalized through the following companion policies and standards:

Together with this WISP they constitute the complete information-security program of SendTax.


16. Contact

Qualified Individual(see § 2)
Security disclosures[email protected]
Privacy inquiries[email protected]
General[email protected]
Operating entityHowell & Gibbs LLC