Security & Privacy FAQs

Your data is encrypted before it ever leaves your environment, using keys you hold. The AI model receives your question and the structure of your data (like column names and data types), never the underlying values. Computation happens on encrypted data. Results return encrypted and are decrypted with your key. Neither Nera nor the model provider ever has access to your data in a readable form, at any point.

TLS 1.2 or higher in transit, AES-256 at rest, plus your own encryption applied client-side before data leaves your environment, using keys you hold.

No, deliberately. Those techniques protect data by removing or substituting the underlying values, which also removes a model’s ability to compute on them meaningfully, and re-identification risk in anonymized datasets is well documented. Nera keeps your values intact and computable while keeping them unreadable to the model, a stronger and more useful guarantee.

This is exactly what Nera is built for. Whether it’s client financials under an engagement letter, patient data under HIPAA, or any other confidentiality obligation, the underlying values are never exposed to the model or to Nera. The confidentiality commitment isn’t a promise, it’s a property of the architecture.

Yes. API access is authenticated with per-customer credentials, encrypted in transit (TLS 1.2+), and scoped to least-privilege permissions. Customer-supplied model API keys are stored encrypted.

Two categories. Customer content: the encrypted datasets you upload or connect, which Nera cannot read in plaintext. Operational data: account identifiers, user emails, authentication metadata, query history (what was asked, which model was used, when), and billing information (processed by Stripe). We don’t require or request PHI, PII, or cardholder data in readable form to deliver the service.

No. Your data is never used to train or improve any model, ours or a third party’s. This is both a contractual commitment and an architectural fact: since the model never receives your underlying values, there’s nothing for it to train on.

Nera works with Anthropic Claude and Google Gemini today, both accessed via commercial APIs, with support to switch models mid-conversation. You can also supply your own model API keys. All supported model providers are US-based.

Yes. You can switch between supported models mid-conversation, and you can supply your own model API keys, keeping that relationship under your own direct control.

Google Cloud Platform and Microsoft Azure, in United States regions only. There’s no replication to other regions by default.

You can delete data at any time during your subscription, and your data is deleted on request and at contract termination, following the cloud provider’s documented erasure process.

Yes. A DPA is available to all customers, and covers how we process personal data on your behalf as a processor under GDPR and applicable law. Read it in full on our DPA page.

To be precise about it: there is no certification recognized by the US Department of Health and Human Services for HIPAA, and no equivalent official certification for GDPR, so no vendor can hold one. What we can tell you is what we do. Nera acts as a data processor under GDPR, with a DPA available to all customers, and we support HIPAA compliance by signing a BAA with covered entities and business associates and maintaining the safeguards required of us as a business associate. SOC 2 Type II is in progress; in the interim, the cloud infrastructure Nera runs on (GCP, Azure) holds SOC 2 Type II and ISO 27001 certification at the infrastructure layer.

No. Signing a BAA or DPA means Nera has committed to meeting its own obligations as a business associate and processor. Your organization remains responsible for its own compliance program, for configuring Nera appropriately, for training your people, and for determining what data you may lawfully submit. Compliance under both HIPAA and GDPR is a shared responsibility, and the agreement covers our half of it.

The Nera ChatApp, the Excel add-in, and API/SDK access are covered by our BAA and DPA. A signed BAA must be in place before any protected health information is submitted, and a DPA before personal data subject to GDPR is submitted. Support channels, email correspondence, and trial or sandbox environments are not covered, so please don’t send regulated data through them. Third-party tools you connect to Nera are governed by your agreement with that provider rather than ours.

We provide Nera’s standard BAA, which lets us apply consistent terms and controls across every customer handling PHI. Email info@nera.systems to start, and we’ll route it to the right people on our side. The same applies to the DPA for GDPR.

No. Nera has experienced no information security breach and no non-minor security incident since founding in November 2023.

Yes. Under NDA, we share system architecture and data-flow documentation, our policy set, and our subprocessor list. We’re also glad to get your security team on a call directly with ours, we find that’s often faster and more useful than a document exchange alone.

MFA is enforced across all Nera infrastructure today. SSO via your identity provider is supported for enterprise configurations, with OAuth currently supported and more identity providers on the roadmap.

Production access is limited to named engineering staff on a least-privilege basis, requires multi-factor authentication, and is logged. Environments (production, staging, development) are separated with distinct credentials and network policies.

Yes, documented at a working level and being formalized further, covering severity classification, roles, and customer notification thresholds. Nera has experienced no security breach and no non-minor security incident since founding in November 2023. Customers are notified of any incident affecting their data or service under the notification terms of their agreement; where law imposes its own notification timeline (such as GDPR’s 72-hour requirement), that legal obligation applies independently.

Production and staging branches are protected; changes require pull request review, with production changes requiring approval from someone other than the author. Security is considered at design time, particularly for the encryption and key-handling paths that are core to the product’s guarantee, and code is reviewed before deployment. Dependencies are updated as part of the regular build and deploy process.

An independent penetration test covering Nera’s web application and cloud infrastructure is planned as part of our compliance roadmap, alongside SOC 2 Type II. Dependency and container image vulnerability scanning is already performed in the build pipeline today.

Nera runs on Google Cloud Platform with Google Kubernetes Engine’s native scaling and self-healing, combined with the geo-redundancy both GCP and Azure provide at the infrastructure layer. A fully documented business continuity and disaster recovery plan, including defined recovery time and recovery point objectives, is in progress as part of our broader compliance roadmap.

Nera doesn’t train or fine-tune models, so governance focuses on the providers themselves: evaluating new model versions before adoption, and notifying customers of material changes. All output is advisory, reviewed by a human analyst, never used for automated decisions about individuals. Numerical results are computed directly from your data rather than generated from model memory, so they’re reproducible and traceable back to source, addressing the most consequential failure mode in AI-assisted analytics: fabricated figures.

No. Nera’s supported model providers are Anthropic, Google, and OpenAI (planned), all US-based. Customers may also supply their own model API keys, keeping that relationship under their own direct control.

Yes. Third-party services are reviewed before adoption, with emphasis on data handling terms, security posture, and certification status. The vendor set is deliberately small and limited to what’s listed in the subprocessor list on our Trust Center.

All Nera personnel go through security and privacy awareness training at onboarding and annually. Production access requires multi-factor authentication and is limited to named engineering staff on a least-privilege basis, logged and reviewed.