Commitments to the hospital.
Each commitment below is formalized in the contract and in the Data Processing Agreement (DPA).
- 01
The data belongs to the hospital
Accrua acts as a processor under the LGPD. The hospital defines the purpose; we process the data only to deliver the contracted service. No use for our own purposes, external benchmarking or resale.
- 02
Minimum necessary
We only ingest the fields the revenue cycle requires. Clinical data that doesn't affect billing or denials doesn't enter the platform.
- 03
No training on your data
Identifiable hospital and patient data never trains, fine-tunes or evaluates AI models, ours or our suppliers'. The platform improves with anonymized, aggregated patterns that don't identify the hospital, unit or patient, with express authorization in the contract.
- 04
Transparency about who has access
Every human or automated access to hospital data is logged, attributed to an identity and auditable. The hospital can request the report at any time.
- 05
Incident notification
If an incident affects your data, we notify you within 24 hours of confirmation, with the information available, and follow up as the investigation progresses. The controller's legal deadline to notify the ANPD is 3 business days.
All controls described on this page can be verified by external audit.
Classification, processing and disposal of hospital data.
What data Accrua receives, for what purpose, how long it is kept and who has access to it.
What enters the platform
We work with three families of data, each with its own rules:
| Family | Examples | Classification at Accrua | Legal basis (LGPD) |
|---|---|---|---|
| Payer contracts | Negotiated tables, billing rules, deadlines, denial clauses | Commercially confidential | Performance of contract (art. 7, V) |
| Billing and denials | TISS forms, batches, remittances, denial reasons, appeals | Confidential, may contain sensitive personal data | Regular exercise of rights and performance of contract (art. 7 and art. 11, II, d) |
| Patient data | Name, insurance card, procedures, ICD, dates of care | Sensitive personal data (art. 11) | Health protection, under the conditions of art. 11, II, f, on the hospital's instruction |
An administrative field becomes health data the moment it is linked to a patient and reveals a procedure, diagnosis or episode of care. That is why we treat the entire billing flow as sensitive by default, even when an isolated field would seem harmless.
The contract formalizes this classification into three confidentiality levels: the hospital's commercial and financial data, patients' personal and clinical data, and Accrua's intellectual property. The first two always belong to the hospital.
Minimization and pseudonymization
- We map field by field what each revenue cycle process needs. What isn't needed to bill, appeal or reconcile is not ingested.
- We separate patient identifiers (name, CPF, insurance card) from clinical-financial content whenever the process allows. In denial and contract analysis flows, AI models receive pseudonymized data; in flows that require identification, such as checking a form, the data goes to the model with zero retention and comes back without being stored.
- Pseudonymizing isn't anonymizing: we keep treating this data as personal and sensitive, with the same controls.
Segregation per hospital
Each hospital has its own logical data space: databases, indexes and backups separated per client, and a dedicated encryption key in AWS KMS. No query, model or report crosses data from two hospitals. Market benchmarks, when they exist, use only aggregated and anonymized data, with express contractual authorization.
Lifecycle: retention and disposal
| Stage | Rule |
|---|---|
| During the contract | Data stays available for the operational period defined with the hospital, aligned with payer denial appeal and audit deadlines. |
| Termination | Full return in an open, structured format, as part of the Reverse Transition Plan, with no data retained for any reason, within 30 days after termination. |
| Deletion | Deletion from production, replicas and backups according to the hospital's instruction, within 90 days after return, with formal proof, unless there is a legal retention obligation. |
| Audit logs | Kept for 12 months after the contract ends, to meet legal obligations and hospital audits, then deleted. |
Roles and responsibilities
The hospital is the controller; Accrua is the processor. Our Data Protection Officer (DPO), Rodrigo Goytacaz, is the point of contact for the hospital's DPO, data subjects and the ANPD, at dpo@accrua.ai. Any data subject request (access, correction, deletion) received by Accrua is forwarded to the hospital within 48 hours, and we support the response within the legal deadline.
Infrastructure on AWS, with Accrua's own controls.
Accrua runs entirely on Amazon Web Services. Under the shared responsibility model, AWS is responsible for the security of the physical, network and virtualization infrastructure; Accrua is responsible for the application, the data and access. Below are the controls on each side.
What AWS guarantees
| Guarantee | What it is | Why it matters to the hospital |
|---|---|---|
US East (N. Virginia) region, us-east-1 | AWS's most mature region, with the largest number of availability zones and the broadest availability of AI models on Amazon Bedrock. The data is physically located in the United States. | This constitutes an international transfer under the LGPD (art. 33). We support it with the ANPD standard clauses (CD/ANPD Resolution No. 19/2024) and the AWS DPA, and we state this in the contract. |
| Infrastructure certifications | ISO/IEC 27001, 27017 (cloud), 27018 (personal data in the cloud) and 27701 (privacy); SOC 1, 2 and 3 reports; HIPAA-eligible services (US health standard). | Reports are available in AWS Artifact and we can share them during due diligence. They cover the physical, network and hypervisor layers. |
| Physical security | Biometric access control, 24x7 surveillance, power and cooling redundancy, audited by third parties. | No Accrua or hospital employee has physical access to servers. |
| Network isolation | Virtual private networks (VPC) isolated by default. | Accrua's traffic doesn't share a network with other AWS customers. |
What Accrua does on top of that
| Control | AWS service | How we use it |
|---|---|---|
| Encryption at rest | AWS KMS + native database, storage and backup encryption | All data at rest is encrypted with AES-256. Keys are managed in KMS, with a dedicated key per hospital and automatic rotation. No engineer sees keys in plain text. |
| Encryption in transit | TLS 1.2+ on all connections | Integrations with the hospital (ERP, HIS, payer portals) and internal traffic between services are encrypted. No exceptions for "internal" connections. |
| Identity and access | AWS IAM, corporate SSO, mandatory MFA | Principle of least privilege. Production access is temporary, approved and logged. There are no shared credentials or permanent access keys. |
| Private network | VPC, private subnets, security groups, private endpoints | Databases and AI services have no public IP. The internet only reaches the application through an entry point protected by a WAF. |
| Threat detection | Amazon GuardDuty, AWS Security Hub | Continuous monitoring of anomalous behavior (credentials used from unusual locations, scans, exfiltration). Findings centralized and prioritized. |
| Audit trail | AWS CloudTrail + application logs | Every API call and every data access is logged in immutable storage, with retention defined in the contract. |
| Backup and recovery | AWS Backup with Vault Lock | Automatic, encrypted backups with point-in-time recovery and a copy locked against deletion, including by administrators. Full restore tested every six months. |
| Secrets and configuration | AWS Secrets Manager | Integration passwords, tokens and certificates are never kept in code or in exposed environment variables. |
We follow the security pillar of the AWS Well-Architected Framework, with an annual architecture review conducted with an AWS partner.
Access control, secure development and vulnerability management.
The most frequent incidents in healthcare stem from poorly managed access, reused credentials and outdated systems. The controls below address these vectors as a priority.
Who accesses what
Mandatory SSO and MFA
For all hospital users on the platform. We integrate with the corporate directory (Microsoft Entra ID and Google Workspace, via SAML/OIDC), so that when an employee leaves the hospital, their access to Accrua is revoked instantly.
Role-based access control (RBAC)
Billing staff, auditors, contract managers and executives see only what their role requires. A denial analyst can't access the contract's financial clauses; a contract manager doesn't see patient data beyond what's necessary.
Accrua team access
No employee has permanent access to production data. Support and engineering get temporary access, with justification, approval from a second person and full logging (who, when, what). The hospital can request a report of these accesses at any time.
Employees
Background checks at hiring, confidentiality agreements and annual LGPD and security training. Access to hospital systems is individual and named, and any departure or leave is communicated to the hospital within 1 business day for revocation.
Audit trail
Every read, change, export and automated action on hospital data generates an immutable record with identity, date, source and affected object. The hospital can access these logs through the platform and export them as CSV or JSON for analysis or for its SIEM.
How we build software
| Practice | How we do it |
|---|---|
| Code review | No change reaches production without review by a second engineer and automated tests. |
| Dependency analysis | Continuous checks for libraries with known vulnerabilities (SCA) and accidentally committed secrets. |
| Static and dynamic analysis | Static analysis (CodeQL), dependency scanning (Dependabot) and secret scanning on every code change; dynamic analysis (OWASP ZAP) against the staging environment on every release. |
| Separate environments | Development, staging and production are isolated. Real hospital data never leaves production; staging uses synthetic data. |
| Infrastructure as code | All infrastructure is versioned and reviewed as code, eliminating untracked manual changes. |
Penetration testing and vulnerability management
- Annual external pentest by an independent firm, covering the application, APIs and infrastructure, with an executive summary available to the hospital under NDA.
- Continuous vulnerability scanning of hosts, containers and images.
- Remediation timelines: critical vulnerabilities within 7 days, high within 30, medium within 90.
- Responsible disclosure: security researchers and clients can report flaws or suspicious behavior via security@accrua.ai and receive a response within 2 business days. Good-faith reporters will not face legal action.
Perimeter protection
WAF with rules against OWASP Top 10 attacks, per-client rate limiting, DDoS protection (AWS Shield) and blocking of traffic from known malicious sources.
How AI models handle data.
AI models process data without storing it. Identifiable data stays within Accrua's controlled environment, isn't used for training and isn't retained by the model provider. The handling of anonymized data is described below.
Where the models run
Accrua consumes language models exclusively through Amazon Bedrock, within our own AWS account. We don't use model providers' APIs directly. This means that:
- Traffic between the application and the model doesn't go over the public internet; it stays within AWS's private network.
- AWS states that content sent to Bedrock isn't used to improve models or shared with model providers.
- Prompts and responses aren't stored by the provider after processing. Where Bedrock offers zero-retention mode for the chosen model, it's enabled; where AWS keeps short retention only for abuse monitoring, this is stated in our technical documentation and the data sent is pseudonymized.
- Models run in the same region as our infrastructure (us-east-1), with no cross-region routing. No data is sent to another provider or region without appearing on the subprocessor list.
What trains and what doesn't
The rule applies in three layers:
- 01
Identifiable data never trains
Forms, contracts, names, insurance cards, ICD codes and any data that identifies a hospital, unit or patient are not used to train, fine-tune or evaluate models, ours or third parties'. This is in the contract, in the model provider configuration (enterprise tier, no training, zero retention) and in our development process.
- 02
Anonymized, aggregated patterns improve the platform
With the hospital's express contractual authorization, we extract statistical patterns that identify no one: for example, what type of mismatch between form and contract usually leads to a recoverable denial. This is what makes the platform better for all hospitals without exposing any of them. Prices, tables and negotiated terms are never disclosed to third parties.
- 03
Anonymization first, not afterwards
Personal data only enters this process after being anonymized, and the result never allows re-identification.
Human oversight
Accrua suggests, classifies, fills in and flags discrepancies. Decisions with financial or patient impact (submitting a denial appeal, changing a charge, accepting a payer dispute) go through human approval, with a record of who approved. The level of autonomy is configurable per hospital and per action type, and the default is conservative. This design follows Brazil's Federal Council of Medicine Resolution No. 2,454/2026 on human oversight and traceability of AI in healthcare, and is formalized in the contract.
Traceability
Every AI suggestion keeps its origin: which document, which contract clause, which form field, which model version and which prompt generated it. The hospital can reconstruct why the platform flagged a denial as recoverable, months later, for a payer or internal audit.
Protection against AI-specific attacks
We follow the OWASP Top 10 for LLM applications. The most relevant risks for the revenue cycle and what we do:
| Risk | What it is | Our protection |
|---|---|---|
| Prompt injection | A document (a denial PDF, a payer email) contains hidden instructions to manipulate the model. | Third-party content is treated as data, never as instructions. Outputs are validated by rules before any action. |
| Cross-client leakage | A model or search index returns data from another hospital. | Indexes and embeddings separated per hospital; no query crosses client boundaries. |
| Sensitive data exposure in responses | The model repeats a CPF, diagnosis or unnecessary data. | Pseudonymization before the model and output filters that remove unrequested identifiers. |
| Excessive agency | The AI executes an action with an irreversible effect without approval. | Write actions require human approval; agent credentials have least privilege. |
| Unbounded consumption | Abusive use that degrades the service or generates cost. | Limits per client, per user and per task. |
AI governance as a discipline
We maintain an inventory of models in use, risk assessment per use case and quality testing before each model change, following the structure of the NIST AI Risk Management Framework (govern, map, measure, manage). ISO/IEC 42001 certification (AI management system) is on the roadmap after ISO 27001.
Brazilian healthcare regulation and international references.
Running a hospital's revenue cycle means operating within the scope of the LGPD, medical confidentiality and the TISS standard at the same time. Below is how each regulation translates into practice at Accrua.
| Regulation | What it requires | How Accrua complies |
|---|---|---|
| LGPD (Law 13,709/2018) | Legal basis for sensitive health data (art. 11), technical and administrative security measures (art. 46), controller and processor roles (art. 39), international transfer (art. 33). | We act as a processor, with a DPA signed with each hospital. The infrastructure is on AWS in the United States (us-east-1), which constitutes an international transfer: we support the transfer with the ANPD standard clauses (CD/ANPD Resolution No. 19/2024) and the AWS DPA, state this in the contract and list each subprocessor and its location. |
| ANPD, Resolution No. 15/2024 | Notification of security incidents with relevant risk to the ANPD and data subjects within 3 business days, by the controller. | Contractual commitment to notify the hospital within 24 hours of confirmation, with a structured report to support the hospital's notification to the ANPD. |
| Medical confidentiality (Code of Medical Ethics, CFM Res. No. 2,217/2018) | Medical records in the institution's custody; clinical information only circulates for care or legal purposes. | Clinical data supporting billing is treated as highly sensitive, with role-restricted access, pseudonymization and logging of every access. |
| CFM Resolution No. 2,454/2026 (AI in medicine) | Governance, human oversight, auditing and transparency in the use of AI that affects medical practice. | Decisions with clinical or financial effect go through human approval; every suggestion is traceable to its origin and model version. |
| TISS standard (ANS) | Mandatory standard for electronic data exchange in private healthcare, including a security and privacy component. | Forms and batches follow the current TISS version published by the ANS, and each new version is adopted within the deadline set by the agency; integrations comply with the confidentiality requirements of the security component. |
Certifications and attestations
Current status of each certification and expected completion.
| Milestone | Status | Expected |
|---|---|---|
| SOC 2 Type II report | Certification in progress | March 2027 |
| ISO/IEC 27001 | Certification in progress | March 2027 |
| ISO/IEC 27701 (privacy) | Planned, after 27001 | No date set |
| ISO/IEC 42001 (AI management) | On the roadmap | After ISO 27001 |
| Inherited from AWS | Inherited from the infrastructure | ISO 27001/27017/27018/27701, SOC 1/2/3, HIPAA eligible. Available via AWS Artifact |
Note on sequencing: for healthcare SaaS vendors, the market usually asks first for SOC 2 Type II (proof that controls operated over a period) and ISO 27001 (management system). 27701 and 42001 come next. HITRUST is only relevant if there are clients in the United States.
Documents available for due diligence
Under NDA, the hospital receives: information security policy, model DPA, subprocessor list, incident response plan, executive summary of the latest pentest and the hospital's own security questionnaire completed (or SIG Lite / CAIQ, if preferred). AWS infrastructure reports come with the package.
Availability, recovery and incident response.
Defined targets for detection, containment, notification and recovery.
Availability and recovery
| Metric | Commitment |
|---|---|
| Platform availability | 95.5% monthly, guaranteed by contract, excluding scheduled maintenance announced 72 hours in advance. Our operational target is 99.5%, and the actual history is public on the status page. |
| RPO (maximum data loss) | 1 hour: point-in-time database recovery and hourly snapshots. |
| RTO (maximum recovery time) | 8 hours for full restoration in another availability zone. |
| Recovery testing | Full restore tested every six months, with results reported to the Governance Committee. |
| Status page | accrua.com.br/status, with real-time availability and a history of incidents and maintenance. |
An Accrua outage doesn't stop the hospital's revenue cycle: the platform doesn't replace the hospital's billing system, and exportable data allows operational continuity even in the event of prolonged unavailability.
Incident response
We maintain a written response plan, with named owners and one simulation exercise per year. The stages and deadlines we commit to:
- 01continuous
Detection
Alerts from GuardDuty, Security Hub, application logs and the responsible disclosure channel.
- 02within 4 hours of detection
Triage and containment
Isolate the affected system, revoke compromised credentials, preserve evidence.
- 03within 24 hours of confirmation
Notice to the hospital
Known facts, data and data subjects potentially involved, actions taken and next steps, updated as the investigation progresses.
- 04within 3 business days
Support for regulatory notification
We provide the hospital with the technical report needed to notify the ANPD (Resolution No. 15/2024), and data subjects when required.
- 05
Eradication and recovery
Root cause fix, restoration from clean backups, validation.
- 06within 15 days
Post-mortem
Root cause report and preventive actions, shared with affected hospitals.
Ongoing transparency
- Quarterly report to the hospital's Governance Committee: Accrua team access to its data, incidents, restore test results and relevant changes to architecture and subprocessors.
- 15 days' advance notice before adding or replacing any subprocessor that handles hospital data, with the right to object.
- Right to audit: the hospital can audit our controls, directly or through a third party, under the terms of the contract.
- Changes to this page are logged with a date; nothing is removed silently.
Suppliers that process hospital data.
Each subprocessor below went through a security and privacy assessment before being onboarded, signed clauses equivalent to those in our DPA and is reassessed annually.
| Subprocessor | Purpose | Data location | Guarantees |
|---|---|---|---|
| Amazon Web Services | Hosting, storage, database, backups, AI inference (Bedrock) | United States (N. Virginia, us-east-1) | ISO 27001/27017/27018/27701, SOC 1/2/3, AWS DPA |
| No model provider contracted directly | All AI models are consumed through Amazon Bedrock, within Accrua's AWS account | United States (us-east-1), covered by the row above | Data not used to train models or shared with providers (Bedrock policy) |
Criteria for joining the list
Independent security certification or report, LGPD-compatible DPA, known data location, support for on-demand deletion and an incident notification channel. Suppliers that don't meet one of these criteria don't receive hospital data.
Changes
Any addition or replacement is communicated to hospitals 15 days in advance, with the right to a reasoned objection, handled at the Governance Committee. The list, with purpose, data categories and processing location, is accessible to the hospital within the platform.
Frequently asked questions
How does an early-stage company ensure the security of patient data?
Through verifiable architecture and controls. Accrua runs on audited AWS infrastructure and maintains its own controls, testable by external audit. We recommend that the hospital's security team apply the same questionnaire used for any vendor.
Does my data leave Brazil?
Yes. Our infrastructure runs on AWS in the United States (us-east-1 region), with the same encryption, isolation and audit controls described on this page. The transfer is supported by the ANPD standard clauses and the AWS DPA, and is stated in the contract. No data is processed outside AWS without appearing on the subprocessor list.
Will your AI learn from my hospital's data and benefit my competitors?
Identifiable data, no: no data that identifies your hospital, units or patients trains any model, ours or third parties'. What the platform learns are anonymized, aggregated patterns authorized in the contract: for example, which billing rules tend to generate denials. Your prices, tables and terms negotiated with payers never leave your environment.
Who at Accrua can see my hospital's data?
By default, no one. Support and engineering access is temporary, justified, approved by a second person and logged. You can require prior approval and you receive a report of these accesses.
What happens if Accrua suffers an attack?
You are notified within 24 hours of confirmation that your data was affected, with facts, scope and plan; this deadline is in the contract. We deliver the technical report for your notification to the ANPD within 3 business days. Backups locked against deletion allow operations to be restored even in a ransomware scenario.
What if I want to leave?
Full return in an open, structured format within 30 days, as part of the Reverse Transition Plan, at no additional cost and with no data retained. Within 90 days after that, we delete your data from production, replicas and backups and issue formal proof of disposal.
Do you have SOC 2 or ISO 27001?
We are in the certification process, with the SOC 2 Type II report and ISO 27001 certification expected for March 2027. Meanwhile, we provide under NDA our policies, the summary of the latest pentest and the AWS infrastructure reports, and we update this page with each completed milestone.
Does this replace the hospital's responsibility under the LGPD?
No. The hospital remains the controller. Accrua fulfills its role as processor, documents each step and provides the hospital with the evidence needed to demonstrate compliance to the ANPD, patients and payers.
Documentation and contacts.
Due diligence package
Policies, model DPA, pentest summary, subprocessor list and completed questionnaires, under NDA.
AWS reports
SOC reports and ISO certificates for the infrastructure, via AWS Artifact.
Security team
For the hospital's CIO, CTO or CISO to talk directly with the people who design and run our controls.
Security channel
For security researchers and the hospital team to report vulnerabilities, suspicious access or phishing in Accrua's name. Response within 2 business days; we don't take legal action against good-faith reporters.
This page was updated on October 13, 2026. Relevant changes are logged here and never removed silently.