Security & Trust

An AI with the keys to your IT. Guarded accordingly.

Resolv holds your staff conversations, your runbooks, and credentials into your systems. Here is exactly how we keep them safe, who can touch them, and what we would hand your auditor if you asked.

SOC

SOC 2

designed to controls

ISO

ISO 27001

alignment in progress

PD

UAE PDPL

Federal Decree-Law No. 45

IA

NESA / UAE IA

government roadmap

01 · Pillars

Six principles, no exceptions.

Resolv’s security posture comes back to six ideas. Every product decision passes through them; every engineer onboards on them.

Pillar 01

Tenant isolation, enforced twice

Company A cannot see Company B — enforced in the application and again at the database. Not a setting. Architecture.

  • Every query scoped to your company
  • Row-Level Security forced at the database
  • Cross-tenant isolation tested in CI on every merge
  • Privileged code paths explicitly re-scoped
  • Per-company config, credentials and spend caps

Pillar 02

Encrypt everything

Data is encrypted in transit and at rest. The credentials that reach into your systems get their own layer.

  • TLS 1.2+ end-to-end
  • Encryption at rest across the platform
  • AES-256-GCM envelope encryption for integration credentials
  • Secrets never in code, never in logs
  • Constant-time comparison for tokens and secrets

Pillar 03

Least privilege, always

Access follows the role, not the person. Privileged access is gated, logged and reviewable.

  • Role-based access: admin, agent, operator
  • Auth checked first on every route and action
  • SSO for enterprise deployments
  • Operator console gated behind a separate secret
  • Every privileged action lands in the audit trail

Pillar 04

The AI is under control

The part most vendors hand-wave. Resolv treats its own AI as a principal to be governed, not trusted.

  • Policy engine decides what agents may do unaided
  • Sensitive actions queue for one-click human approval
  • Reasoning loops hard-capped; low confidence escalates
  • Structured, typed outputs — never free-form execution
  • Prompt-injection defence and PII redaction in logs

Pillar 05

Data stays where you need it

Residency is a deployment decision you make, not a promise we make. Up to and including your own perimeter.

  • UAE / regional data-residency options
  • Regional model routing for regulated industries
  • On-premise sovereign deployment for government
  • Your knowledge base is never shared across tenants
  • Customer-initiated export and deletion

Pillar 06

Auditable by you, not us

Every state change writes a permanent record. The log is append-only — the database refuses edits and deletes.

  • Append-only audit trail, enforced at the database
  • Who, what, when, before and after — plus request ID
  • Every AI action attributed and reviewable
  • Exportable for your auditors
  • Written post-mortem after any incident

02 · Architecture

How a request is handled.

From a staff message to an action in your systems, six layers stand between an attacker and your environment.

Defence in depth

Each layer is independently logged and monitored. Compromise one and the next still stands.

EDGEWAF · rate limiting per company · signed webhooks (HMAC, freshness, idempotency)
IDENTITYJWT with tenant claims · role-based sessions · SSO on enterprise
SERVICEAuth first on every handler · input validation at every boundary · no mass assignment
AI RUNTIMEPolicy engine · human approval gates · capped loops · per-company spend caps
DATAForced Row-Level Security · tenant-scoped vector search · field-level credential encryption
AUDITAppend-only trail · request IDs end to end · every privileged action recorded
Staff message · Teams / email / webTLS 1.2+
Edge · WAF + rate limitsper-company
API · auth before anythingtenant-scoped JWT
AI runtime · policy-gatedapprovals for sensitive actions
Postgres · RLS enforcedyour tenant only
Audit log · append-onlypermanent record

03 · Certifications

Where we stand.

We are building toward formal certification and will not claim what has not been audited. Here is the honest status of each, and what we can share under NDA today.

SOC

SOC 2 Type II

We build to the Trust-services criteria across Security, Availability and Confidentiality, working toward a Type II audit.

In progressAsk us
ISO

ISO/IEC 27001:2022

Our information security management system is being built to align with ISO 27001, scoped to all Resolv production systems.

Alignment in progressAsk us
PDPL

UAE PDPL

We design to Federal Decree-Law No. 45 of 2021, including residency and processing controls, ahead of an independent gap analysis.

Design targetAsk us
NESA

NESA / UAE IA

Government deployments are designed against the UAE Information Assurance standard; alignment work is scoped per engagement.

Government roadmapAsk us

04 · Subprocessors

Who else touches your data.

The vendors we rely on in the managed cloud edition, each bound by a data-processing agreement. We update this page before adding anything new.

SubprocessorPurposeData handled
AWS · EKS + CloudFrontAmazon Web ServicesCompute, hosting and edge deliveryApplication traffic · UAE (me-central-1)
AWS · Aurora PostgreSQLAmazon Web ServicesDatabase and vector searchTenant data · UAE (me-central-1)
AWS · S3, KMS & CognitoAmazon Web ServicesStorage, encryption keys and identityDocuments, keys and logins · UAE region
AnthropicAnthropic PBCAI inference (primary model)Conversation content · no training
OpenAIOpenAI LLCAI inference + text embeddingsConversation and document text · no training
GoogleGoogle LLCAI inference (Gemini)Conversation content · no training
StripeStripe PaymentsSubscription billingBilling details only
AWS · SESAmazon Web ServicesTransactional emailEmail addresses only · UAE region
SentryFunctional Software Inc.Error monitoringScrubbed metadata only

Government and regulated deployments can run the Ops Platform Edition entirely inside your perimeter — with no external subprocessors at all.

05 · Vulnerability program

Tell us if you find something.

Responsible disclosure + third-party pen test

We run a responsible-disclosure programme and commission independent penetration testing. Researchers acting in good faith are covered by our safe-harbour policy.

Report any vulnerability to security@resolv.so.

Critical target

48 hrs

patched and deployed

High target

7 days

patched and deployed

Triage target

1 day

first response to a report

Pen test

Independent

summary shared under NDA

Our commitment — transparency by default

Acknowledge a report< 1 day
Fix criticals< 48 hrs
Regulator notification (PDPL)< 72 hrs
Post-mortem after incidentsAlways
Safe-harbour for researchersYes
Pen-test summary under NDAOn request

Talk to us about security

Get the architecture walkthrough and our security documentation in a 30-minute call.

Live in days · Nothing to install · Every action audited