Version 1.0 · Adopted 6 November 2026 · Next review 6 November 2027
This document is the information security policy of Coffee & Fun LLC, trading as Helperbird. It is written for the people who have to review us: district privacy teams, enterprise security reviewers, and anyone completing a vendor assessment on our behalf. It is organized by the eighteen NIST SP 800-53 Rev 5 control families so that it can be read alongside an assessment form without translation.
It is published rather than kept internal, because a policy a customer cannot read is a policy a customer cannot rely on.
What this document does not claim. Helperbird holds no SOC 2 Type II report, no ISO/IEC 27001 certificate and no FedRAMP authorization today. Where a control is not met, this document says so in the same plain language it uses for the controls that are. A section that describes an intention rather than an operating practice says which it is.
What this public version leaves out. It states each control at the level of policy. It does not name internal tools, configuration, monitoring thresholds or architecture, and it does not give the technical detail of open remediation items, because that detail helps an attacker more than it helps a reviewer. If you are completing a security review of Helperbird, you can request it from [email protected], under a non-disclosure agreement where appropriate.
How to read this
Helperbird is a micro-business running a deliberately small system. That shapes everything below, in both directions.
It makes some controls unnecessary. There is no customer database, no server we run, no operating system we patch, no physical or removable media, no office that hosts any part of the service, and no end-user account directory. Controls that protect those things have nothing here to protect.
It makes some controls harder. A small team cannot staff the separate security, change-approval and operations functions a larger organization would. Where a control is met in a simpler form, this document says so rather than describing a structure that does not exist.
Scope
This policy applies to the Helperbird service in its entirety: the browser extension (Chrome, Edge, Firefox, Safari), the Google Workspace add-on, the Helperbird API, the cloud infrastructure that runs it, the source repositories, and every third-party service that processes data on Helperbird's behalf. It applies to every person with access to any of those, whether employee, contractor or founder.
Roles and responsibilities
| Role | Responsibility |
|---|---|
| Information Security Officer | Owns this policy, approves exceptions, leads incident response |
| Data Custodian | Named custodian under customer data agreements; owns the subprocessor register |
| System Administrator | Production access, infrastructure change, credential management |
| Approving Authority for change | Final approval for any change reaching production |
Helperbird is a small team, so one person may hold more than one of these roles. Who holds each role is confirmed to customers in their agreement, or on request.
Management commitment
The owner of the business adopts this policy and is accountable for it. Security work is funded and scheduled as part of ordinary product work, not deferred to a separate budget that does not exist.
Coordination
Helperbird has no separate organizational entities to coordinate between. Coordination that matters is external: with subprocessors through their agreements and status channels, with browser vendors through their review processes, and with customers through the contact addresses below.
Compliance
Compliance with this policy is assessed at the annual review, and whenever a customer assessment, an incident, or a material architecture change gives cause. Non-compliance is recorded and remediated, and reviewed by the Information Security Officer.
Contact
Security and compliance: [email protected]. Vulnerability reports: [email protected].
1. Access Control (AC-1)
Policy. Access to Helperbird production infrastructure, source repositories and third-party service consoles is limited to named individuals with a demonstrated need, authenticated with multi-factor authentication, and granted at the least privilege sufficient for the task. Shared credentials are prohibited without exception.
1.2 Procedures.
- Every administrative account is individual. No account is shared between people, and no generic or role account exists for interactive use.
- Multi-factor authentication is required on every account with reach into source code, infrastructure, payments, or any third-party service that processes data for Helperbird.
- Cloud permissions are granted through roles scoped to least privilege, and the permissions the service runs with are kept separate from the permissions used to deploy it. Role definitions are held in version control.
- Credentials are generated by and held in a password manager. They are never committed to a repository, never shared over chat or email, and never reused.
- Access is granted by the Information Security Officer and reviewed whenever a person joins, changes role or leaves, and at the annual policy review.
- Administration is performed only from devices with full-disk encryption, over encrypted connections, with multi-factor authentication.
What end users get. Nothing. Helperbird issues no user accounts, holds no user directory and stores no end-user credentials. There is no account for an attacker to take over.
2. Awareness and Training (AT-1)
Policy. Anyone with access to Helperbird production systems, source code or customer data completes security awareness training before that access is granted and annually thereafter, covering phishing, credential handling, secure development and incident reporting.
2.2 Procedures. No training procedure is yet in operation. This policy establishes the requirement; the first training cycle has not been delivered or evidenced. We record that as a gap rather than describing an aspiration as a practice.
What is already in place and does not depend on training: multi-factor authentication on every account, secure-coding rules checked automatically on every build rather than left to anyone's memory, and review of every change before it ships.
3. Audit and Accountability (AU-1)
Policy. Administrative actions and service requests are logged, retained for a defined period, protected from casual modification, and reviewed when there is cause. Logs never contain customer-submitted content.
3.2 Procedures.
- Administrative actions in the cloud account are logged and attributable to an individual.
- Each request to the Helperbird API is logged. Request logs contain metadata only. Text, images and audio submitted to a feature are never written to a log.
- Logs are kept outside the application's own write path and retained according to a defined schedule.
- Logs are reviewed during incident response and whenever a High-severity vulnerability is identified (see §13).
4. Security Assessment and Authorization (CA-1)
Policy. The security posture of the service is assessed on every release and reviewed in full annually. Findings are tracked to closure.
4.2 Procedures.
- Every release goes through dependency scanning, static code analysis, and dynamic scanning of the public API.
- External encryption and security-header configuration is checked with independent public testing services.
- A CASA Tier 2 self-assessment is maintained for Google's review requirements.
- Every release is independently reviewed before publication by Google (Chrome Web Store), Microsoft (Edge Add-ons), Mozilla (AMO) and Apple (App Store).
- This policy and the control set behind it are reviewed annually and after any material architecture change.
Known gap. None of the above is an independent third-party attestation. Helperbird holds no SOC 2 Type II, ISO/IEC 27001, HITRUST or FedRAMP report. A SOC 2 Type II audit is in preparation.
5. Configuration Management (CM-1)
Policy. The configuration of the system is defined entirely as code, held in version control, and changed only through a reviewed and approved path.
5.2 Procedures.
- The baseline is the repository. Every cloud resource, permission and route is defined as code, and every dependency is pinned to an exact version and integrity hash.
- Deployment runs only from a tagged commit through the project's deployment tooling. A change made outside that path shows up as drift from the defined configuration.
- Every change is reviewed before merge. Commits follow a structured convention so security-relevant changes are identifiable in history.
- A security impact assessment is required for any change that touches authentication, encryption, the subprocessor set, data flow or access permissions. The assessment is recorded with the change.
- Rollback is a redeploy of the previous tag.
Not applicable. CIS Benchmarks, USGCB and SCAP checklists have no applicable content for the managed cloud services Helperbird uses. There is no operating system, host or container image that Helperbird hardens.
6. Contingency Planning (CP-1)
Policy. The service can be restored to a known-good state after an outage, breach or provider failure, and the plan for doing so is documented and tested annually.
6.2 Procedures.
- Recovery. The whole system is rebuilt by redeploying a tagged release, typically in minutes. Because the service is stateless, there is no customer data to restore and no recovery-point objective to miss.
- Resilience. The API runs on managed services spread across several facilities within one United States region, so it survives the loss of a single facility but not of the whole region. On-device Helperbird features keep working if the API is unavailable.
- Dependency outage. If an AI or speech provider is unavailable, the affected features degrade gracefully and the rest of Helperbird continues. No feature failure cascades into a general outage.
- Testing. An annual tabletop exercise covers a regional outage, a subprocessor outage and a credential compromise.
Known gap. The plan is newly adopted and the first tabletop has not yet been run.
7. Identification and Authentication (IA-1)
Policy. Every person with administrative reach is uniquely identified and authenticated with multi-factor authentication. End users are deliberately not identified at all.
7.2 Procedures.
- Administrative users authenticate as individual identities with multi-factor authentication, and their actions are attributable to them.
- Credentials are long, unique and generated by a password manager.
A deliberate design decision, stated plainly. Helperbird does not identify its end users. There is no sign-up, no user record and no password. This means Helperbird cannot satisfy a per-user authentication control, and cannot provide per-user audit trails to a customer. We accept that trade: a tool used by students should not require us to hold a list of them.
8. Incident Response (IR-1)
Policy. Security incidents are detected, contained, eradicated, recovered from and reported against defined timeframes, and affected customers are notified within the period their agreement requires.
8.2 Procedures.
- Detection. Automated monitoring alerts on abnormal service behavior and on application errors.
- Triage. The Information Security Officer assesses scope, data involvement and severity, and records the assessment with timestamps.
- Containment and eradication. Options include revoking credentials, rotating secrets, disabling an affected endpoint, and redeploying a known-good tagged release.
- Customer notification. Within 24 hours of confirming an incident where a customer's agreement requires it, and otherwise within 72 hours under the standard Data Privacy Agreement. Where the two differ, the shorter period in the customer's own agreement controls.
- Reporting. A status report within 72 hours of a customer request, and a postmortem within two business days of resolution, covering nature, data involved, who was affected, mitigation and corrective action.
- Testing. An annual tabletop exercise.
Known gap. The plan is newly adopted and has not yet been exercised.
9. Maintenance (MA-1)
Policy. Where Helperbird operated maintainable hardware, hosts or operating systems, maintenance would be scheduled, authorized, logged and performed only by authorized personnel.
9.2 Procedures. No procedure is in operation, because nothing exists to maintain. Helperbird owns no servers, appliances or network equipment. The compute platform, host operating systems and hypervisor are maintained by AWS, and Helperbird has neither access to nor patching responsibility for that layer. This section exists so that the gap is a stated conclusion rather than an omission; it would need to be written properly before Helperbird ever ran infrastructure of its own.
10. Media Protection (MP-1)
Policy. Customer data is sanitized to a standard no weaker than NIST SP 800-88 Rev 1 before disposal, and a certificate of destruction is provided to a customer on request.
10.2 Procedures.
- No physical or removable media is used at any point in the service. There are no disks, tapes, drives or printed records holding customer data.
- Logical sanitization on request: on a customer's instruction or on termination, account-level metadata is deleted and a written confirmation issued. Because Helperbird retains no user content, there is in practice almost nothing to delete.
- Where data sits on AWS-managed storage, cryptographic erase through AWS key destruction is the sanitization method, consistent with NIST SP 800-88 Rev 1 Clear and Purge guidance for cloud storage.
- A certificate of destruction is issued within 30 days of a request.
No routine media-handling procedure operates, because there is no routine media to handle.
11. Physical and Environmental Protection (PE-1)
Policy. Physical and environmental protection of the systems processing customer data is inherited in full from Amazon Web Services. Helperbird operates no facility that hosts any part of the service.
11.2 Procedures. No Helperbird procedure operates. Physical access restriction, access logging, intrusion detection, surveillance, power redundancy and environmental controls are AWS responsibilities, evidenced by the AWS SOC 2 Type II report. No Helperbird person can obtain physical access to the infrastructure.
NIST does not permit a policy to be inherited, which is why this section exists as Helperbird's own statement. The underlying controls genuinely are inherited, and we do not present AWS's report as our own attestation.
12. Personnel Security (PS-1)
Policy. Background screening appropriate to the access being granted is completed before any person is given access to Helperbird production systems, source code or customer data. Access is revoked on the day a person's engagement ends.
12.2 Procedures. No screening procedure is in operation. This policy makes screening a precondition for granting production access to anyone new. It has not been applied retrospectively to existing access, and there is no screening program to describe today.
Access revocation on departure is covered by §1.
13. Risk Assessment (RA-1)
Policy. Vulnerabilities are identified continuously, triaged by severity, and remediated within defined windows: High within 30 days, Medium within 60 days.
13.2 Procedures.
- Continuous advisory monitoring across the full dependency tree.
- Static code analysis on every release.
- Dynamic scanning of the public API.
- Advisories are remediated deliberately, one dependency at a time, never by a blanket automatic upgrade.
- Where an advisory cannot be cleared by upgrade, the options in order are: replace the package, pin a safe version across the whole tree, or record an accepted risk with the reasoning. A permanently unfixable advisory is never left standing, because a real finding can hide behind it.
- When a High-severity vulnerability is identified, logs are reviewed for evidence of exploitation before the finding is closed.
Known gap. Item 6 is newly adopted and has no operating history behind it yet.
14. System and Services Acquisition (SA-1)
Policy. No third party processes data on Helperbird's behalf until it has been assessed and placed under a written agreement. No dependency enters the codebase without review.
14.2 Procedures.
- A prospective subprocessor must hold a current SOC 2 Type II or equivalent independent attestation and sign a Data Processing Agreement, and a Business Associate Agreement where the context requires one, before it processes anything.
- The subprocessor register is published in the Data Privacy Agreement and on the security page. Customers receive 30 days' notice of a material change where reasonably practicable, and may object on reasonable data-protection grounds.
- New dependencies are reviewed for maintenance status, license and transitive weight before adoption. Packages that leave the registry or stop being maintained are replaced.
- Runtime platforms are kept on supported versions and updated ahead of end of support.
15. System and Communications Protection (SC-1)
Policy. Data is encrypted in transit and at rest, customer processing is isolated, and the boundary of the service is defended.
15.2 Procedures.
- In transit: TLS 1.2 and TLS 1.3 only. Older protocols are disabled, and only modern FIPS-approved cipher suites are offered.
- At rest: AES-256 through AWS-managed encryption, with keys held in FIPS 140-2 Level 3 validated hardware security modules.
- Isolation: each request runs in its own isolated execution environment, with no shared state between requests.
- Boundary: managed edge services front the API, and standard security headers are applied to every response.
- No content at rest: the API provisions no database. Content submitted to a feature passes through and is discarded on completion.
- Residency: all processing and all logging occur within the continental United States.
The algorithms in use are FIPS-approved. Helperbird does not claim FIPS 140 validated status for the cryptographic modules themselves.
16. System and Information Integrity (SI-1)
Policy. Flaws are identified and remediated promptly, and the integrity of the deployed system is maintained through immutable, reviewed deployment.
16.2 Procedures.
- Static analysis and dependency scanning on every release (see §13).
- Deployment is immutable: the running system is built from a tagged commit and replaced wholesale, never patched in place.
- Every feature that processes untrusted page, PDF or selection text is protected against prompt injection on the server. Any new text-processing feature must be too.
- Output from AI models is validated before it is used. What a model reports about its own output is checked, not trusted.
- Manifest V3 is used for the browser extension, which prevents remotely hosted code from running and enforces a strict content security policy.
17. Planning (PL-1)
Policy. Security is considered at design time, and the architecture is documented well enough that a reviewer can understand the data flow without reading the source.
17.2 Procedures.
- Data flow, subprocessors and security measures are documented publicly across the security page, the Data Privacy Agreement and this policy. More detailed architecture documentation is shared with reviewers on request.
- On-device processing is the default. A feature sends data off the device only where it cannot work otherwise, and only content the user has explicitly selected.
- New features that would introduce a network call, a new subprocessor or a new data category require a documented security impact assessment before implementation.
Known gap. There is no System Security Plan in the NIST SP 800-18 sense, and no Plan of Action and Milestones. The material that an SSP would draw on exists, but it has not been assembled into control-by-control implementation statements. Producing one is tied to the SOC 2 Type II preparation.
18. Supply Chain Risk Management (SR-1)
Policy. Both supply chains — the software dependency tree and the vendor set — are actively managed, and weaknesses in either are identified and addressed.
18.2 Procedures.
- Software. Exact-version lockfiles with integrity hashes, continuous advisory monitoring, and review before any new dependency is added.
- Build. Deployment artifacts are built from a tagged commit. Secrets are never committed to source control.
- Vendors. Independent attestation and a signed agreement are preconditions for any subprocessor (see §14). The register is published and customers are notified of material change.
- Concentration risk, stated openly. The service depends on a small number of major providers, all listed in the subprocessor register. If one fails, the features that rely on it degrade gracefully. A regional failure at the hosting provider takes the API offline (see §6).
Change history
| Version | Date | Change |
|---|---|---|
| 1.0 | 6 November 2026 | First published version. Documents existing practice across all eighteen NIST SP 800-53 Rev 5 control families at policy level, and states the gaps where practice does not yet exist. |
