Security policy

This Security Policy describes the administrative, technical and physical safeguards Phyto Nex LLP applies to the Phyto Nex AI service and to the Customer Data entrusted to us. It is effective from January 15, 2026 and was last updated on September 1, 2026. It forms the basis of the security measures summarized in Annex II of our Data Processing Addendum.

Our information security program is owned by Arvind Subramaniam, Chief Information Security Officer, who reports to the Chief Executive Officer and presents a risk report to the board of directors every quarter. A Security and Privacy Steering Committee (the CISO, Head of Privacy, VP Engineering and Head of Customer Operations) meets monthly to review risks, incidents, exceptions and policy changes.

• Written policies covering information security, acceptable use, access control, cryptography, secure development, incident response, business continuity and vendor management, reviewed at least annually.
• Mandatory security and privacy training for all staff at onboarding and annually, with targeted secure coding training for engineers and quarterly phishing simulations.
• Background checks for all employees and contractors with access to production systems, to the extent permitted by local law.
• Confidentiality agreements signed by all personnel before access is granted.

Phyto Nex AI is independently audited against recognized frameworks. Reports are available to customers and qualified prospects under a non-disclosure agreement through our trust center.

• At rest: all Customer Data, including receipt images, extracted line items, databases, queues and backups, is encrypted with AES-256.
• In transit: all connections to the Phyto Nex AI application, API and connectors use TLS 1.2 or higher with modern cipher suites. TLS 1.0, TLS 1.1 and weak ciphers are disabled. HSTS is enforced on all Phyto Nex AI domains.
• Internal traffic: service-to-service traffic within our virtual private cloud is encrypted with mutual TLS.
• Secrets: connector OAuth tokens and API credentials are encrypted at the field level with a separate data key before being stored.

Encryption keys are generated and stored in AWS Key Management Service (KMS) in the region where the customer's data resides, backed by FIPS 140-2 validated hardware security modules. Key material never leaves KMS in plaintext. Keys are rotated automatically every 12 months, and access to key usage is restricted to specific service roles and logged in AWS CloudTrail.

Enterprise customers receive dedicated per-tenant customer master keys, so that their data can be cryptographically isolated from other tenants. Enterprise customers may also request that a key be disabled, which renders their stored data unreadable and serves as an additional deletion assurance at contract end.

Customer access

Customers can enforce single sign-on through SAML 2.0 or OpenID Connect (Okta, Microsoft Entra ID, Google Workspace and others), require multi-factor authentication, and provision users with SCIM. Role-based permissions separate administrators, AP reviewers, approvers and read-only auditors, and every reviewer action is recorded in an immutable audit log.

Phyto Nex AI staff access

• All staff authenticate through single sign-on with phishing-resistant multi-factor authentication (FIDO2 security keys).
• Production access follows least privilege, is granted only to engineers on the on-call rotation, and requires just-in-time elevation with a ticket reference and automatic expiry after four hours.
• Support staff cannot view Customer Data unless the customer grants time-limited support access from within the application.
• Access rights are reviewed quarterly by system owners, and access is revoked within 24 hours of a role change and immediately on departure.
• Company devices are centrally managed, with full disk encryption, endpoint detection and response, screen lock and automatic patching enforced.

Phyto Nex AI is designed to observe, not act. Every connector requests the narrowest scope that allows us to read expense reports, receipts and transactions. None of our connectors can approve, reject, edit or delete expense reports, post journal entries, move funds or change card limits.

Primary account numbers are truncated to the last four digits at the ingestion boundary, before data is written to any queue, database, log or backup. A detection rule scans OCR output for any remaining 13 to 19 digit sequences that pass a Luhn check and masks them as well. Full card numbers, card verification codes and PINs are never stored, so the Phyto Nex AI platform remains outside PCI DSS cardholder data scope.

Application and infrastructure logs are scrubbed of email addresses, names, amounts and receipt text before they leave our VPC for our monitoring providers. Error reports are filtered with allow-lists, so only approved fields are transmitted.

• No training: Customer Data is never used to train or fine-tune foundation models, and our model provider is contractually prohibited from doing so.
• Zero data retention: language model requests are sent to Anthropic, PBC under zero data retention terms, so prompts and completions are not stored after processing.
• Minimum necessary context: prompts contain only the specific line items, policy clause and tone settings required for a draft.
• Prompt injection defences: receipt and reply text is treated as untrusted data, isolated from system instructions, and model outputs are validated against a schema before display.
• Human in the loop: drafts are never delivered to employees without explicit approval by a customer's authorized reviewer.
• Evaluation: every model or prompt change is tested against a held-out evaluation set for accuracy, tone and leakage before release.

Every change to production code is peer reviewed, passes automated tests and is scanned before merge. Our pipeline runs static analysis, software composition analysis for vulnerable dependencies, secrets detection and infrastructure-as-code checks. Production infrastructure is defined in Terraform and changes are deployed only through the CI/CD pipeline; direct console changes are blocked and alerted.

Development, staging and production environments are separated in distinct AWS accounts. Production Customer Data is never copied to lower environments; testing uses synthetic receipts and expense reports generated for that purpose.

We scan containers, hosts and dependencies continuously and remediate vulnerabilities within these target windows from discovery: critical within 72 hours, high within 14 days, medium within 45 days and low within 90 days.

Independent penetration tests are performed twice a year by Cobalt Ridge Security, LLC, covering the web application, public API, connectors and AWS configuration. Findings are tracked to closure and retested. An executive summary of the most recent test is available to customers under NDA.

Phyto Nex AI maintains a documented incident response plan that is tested through tabletop exercises at least twice a year. Security events are monitored around the clock with automated alerting, and an on-call security engineer triages every alert.

1. Detection and triage: the on-call engineer classifies severity within 30 minutes of an alert.
2. Containment: affected systems or credentials are isolated, and forensic evidence is preserved.
3. Notification: if a security incident affects Customer Data, we notify the customer's designated security contacts without undue delay and in any event within 72 hours of confirmation, with updates as facts become known. Our DPA sets a shorter 48-hour window for personal data breaches.
4. Eradication and recovery: root causes are removed and services restored from known-good states.
5. Post-incident review: a written review with corrective actions is completed within 10 business days and shared with affected customers.

To report a suspected incident affecting Phyto Nex AI, email security@phytonexai.com or call +91 80 4123 7815.

The Phyto Nex AI platform runs across multiple availability zones within each region. Databases are continuously replicated, with point-in-time recovery, and encrypted backups are copied to a separate AWS account within the same geography so that data residency commitments are preserved.

• Recovery point objective (RPO): 1 hour.
• Recovery time objective (RTO): 4 hours.
• Disaster recovery restoration tests are performed at least twice a year, with results reviewed by the Steering Committee.
• Staff in our Coimbatore office and remote teams can operate the service independently of any single physical location.

Before we engage a vendor that will access Customer Data, we review its security posture (SOC 2 or ISO 27001 reports, questionnaires and contract terms), sign a data processing agreement with appropriate transfer safeguards, and limit the data shared to what the vendor needs. Vendors are reassessed annually. Our current subprocessors are listed at /legal/subprocessors.

We welcome reports from security researchers. Please send findings to security@phytonexai.com with enough detail to reproduce the issue. We acknowledge reports within 2 business days, provide a triage decision within 7 business days and keep you informed until resolution. Valid, previously unknown findings are eligible for a reward from US$150 to US$5,000 depending on severity and quality.

In scope

• app.phytonexai.com and api.phytonexai.com
• www.phytonexai.com
• Phyto Nex AI's Slack and Microsoft Teams apps

Out of scope

• Denial of service, volumetric or resource exhaustion testing.
• Social engineering, phishing or physical attacks against staff or offices.
• Findings in third-party platforms such as SAP Concur, Ramp or Slack themselves (please report those to the vendor).
• Missing security headers or cookie flags without a demonstrated exploit, self-XSS, clickjacking on pages without sensitive actions, and reports from automated scanners without validation.
• Accessing, modifying or exfiltrating data belonging to other customers. Use only accounts you created.

Safe harbor

If you make a good-faith effort to comply with this policy, we will consider your research authorized, will not pursue or support legal action against you, and will work with you to understand and resolve the issue. Please give us a reasonable time to remediate before public disclosure, generally 90 days.

Postal correspondence on security matters may be sent to Phyto Nex LLP, Attn: Chief Information Security Officer, 312, Thiruvalluvar Street, Vellakinar Pirivu, Gnanambika Mills, Coimbatore North, Coimbatore, Tamil Nadu 641029, India.