Security

PlanVantage is a financial modeling platform for employee benefit plans, operated by PlanVantage LLC in the United States. This page describes the security measures in place today, written for the people who evaluate vendors: brokers, consultants, and the security reviewers behind them.

Every statement on this page describes the platform as it currently runs. Nothing here is a roadmap item. If a question your review needs is not answered here, email contact@planvantage.ai and you will get a written answer.

PlanVantage works on the contents of a renewal package: rates, enrollment counts, contributions, plan designs, funding arrangements, fixed costs, and monthly aggregate claims experience. Most of that is not protected health information, and it is enough to build a projection.

Our Terms of Service (section 6) do not permit you to submit, store, or process PHI as HIPAA defines it unless you and PlanVantage have executed a business associate agreement. Where a BAA is in place, PHI-bearing data is processed only through infrastructure and subprocessors covered by that BAA. A BAA and a data processing agreement are available on request.

The platform runs on Google Cloud Platform. Customer data is stored in the United States (region us-east1).

• The application runs in containers on Cloud Run.
• Customer data is stored in a PostgreSQL database on Cloud SQL.
• Uploaded files are stored in Google Cloud Storage.
• The database is backed up automatically every night.

PlanVantage operates no physical servers of its own. Physical and environmental security for the infrastructure above is provided by Google Cloud's data centers.

• In transit: all traffic to the site and the application is served over HTTPS (TLS). Every response carries a Strict-Transport-Security header with a one-year max-age, so a browser that has visited once will not attempt a plaintext connection.
• At rest: the database and file storage are encrypted at rest with AES-256 under Google Cloud's default encryption, which covers all stored customer data.
• Secrets: application secrets, including database credentials, signing keys, and API keys, live in Google Secret Manager. They are not stored in source code, container images, or configuration files.

• Passwords are hashed with bcrypt at a work factor of 12. Plaintext passwords are never stored.
• Sessions use short-lived access tokens that expire after 15 minutes, refreshed through rotating refresh tokens. The server stores only cryptographic hashes of session tokens, never the tokens themselves.
• Multi-factor authentication is available on every account: a one-time code delivered by email at sign-in, with single-use recovery codes.
• Single sign-on is available with Microsoft Entra ID, over OpenID Connect using the authorization code flow with PKCE. Entra is the identity provider wired today. An organization can require SSO for its email domain, so its users cannot bypass its identity provider with a password.
• Authorization is enforced server-side on every request. Every record is scoped to the account that owns it, and ownership is checked through the data model on the server, never trusted from the client.
• Sign-in endpoints are rate limited to slow down credential-stuffing and brute-force attempts.

• Security headers are set on both the website and the API: a Content-Security-Policy, X-Frame-Options DENY against clickjacking, nosniff content-type handling, a strict referrer policy, and a permissions policy that disables browser features the product does not use.
• Cross-origin requests to the API are restricted to PlanVantage's own origins.
• Uploaded files are processed in isolation. Computational processing of uploaded files, including any code execution performed during AI-assisted extraction, runs in a separate sandboxed service inside infrastructure PlanVantage controls.
• Audit logging: security-relevant account actions and data changes are recorded in an audit log with the acting user, the action, and the affected resource.
• Monitoring: an external uptime check polls the application every minute from multiple regions, and alerting policies notify the operator automatically when a check fails or when server errors spike. The application also sends its own error alerts.
• Deployment: changes reach production through a gated deployment pipeline with automated checks, tagged releases, and a defined rollback procedure.

The platform uses large language models to extract data from the documents you upload and to power the in-application assistant. Every model call goes through one subprocessor: Google Cloud, using Claude and Gemini models served on Google Vertex AI in US regions. The model developers, including Anthropic, do not receive your data. Google is the processor, under the same Google Cloud terms that cover the rest of the platform.

• No model training on your data. Google's service terms prohibit using customer data to train or tune any model, and we do not use customer data to train models.
• Zero data retention at the model provider. Prompt and response logging is off. Google keeps nothing from a request beyond what is needed to return the output, and we use no model in Google's advanced-monitoring tier, which would log prompts for 30 days. The one exception Google reserves for every customer: a prompt its automated abuse classifiers flag may be logged for up to 90 days to investigate a policy violation, and never used for training.
• Human review before write. AI-assisted extraction proposes values that you confirm before they are applied to your records. You control what is saved.
• Execution stays in our infrastructure. Code execution performed as part of extraction runs in PlanVantage's own sandboxed infrastructure, not in a third-party AI provider's hosted execution environment.

The specific models we use will change over time. The provider, the region, and the retention position above do not change without this page changing first. The Privacy Policy covers AI processing in more detail.

PlanVantage runs on Google Cloud Platform, which maintains SOC 2, ISO 27001, and other third-party security certifications covering both the infrastructure and the AI processing described on this page. Google Cloud is the only subprocessor that handles your documents. This page exists so that what is actually in place is stated plainly rather than implied.

• A business associate agreement and a data processing agreement are available on request.
• Security questionnaires and due-diligence requests are answered directly and in writing: email contact@planvantage.ai.
• We do not sell personal information, we do not use advertising cookies, and we do not engage in cross-context behavioral advertising. See the Privacy Policy.

To report a suspected vulnerability, email contact@planvantage.ai with enough detail to reproduce it. Reports are read by the people who operate the platform.

PlanVantage LLC
3500 Lenox Rd NE, STE 1500-2490
Atlanta, Georgia 30326
Email: contact@planvantage.ai