Skip to content
Security & trust

Your data is safe. Full stop.

The controls, certifications and legal documents your security team needs, in one place.

Certifications

Audited, certified, compliant

Not just logos on a page. Here’s what each one actually requires.

SOC 2 Type 2

Independent third-party audit of security controls. Covers access controls, change management, risk mitigation, and system monitoring.

Independently audited over an observation period, not a point in time.

ISO 27001

International standard for information security management systems. Requires documented policies, risk assessments, and continuous improvement.

Covers the full lifecycle of how data is handled, stored, and protected.

GDPR

EU regulation for personal data protection. Covers the lawful basis for processing, the duty under Article 14 to inform people whose data was obtained indirectly, and the rights in Articles 15 to 21.

Applies to all customer data regardless of where Evergrowth operates.
Data principles

What happens to your data inside Evergrowth

Specific answers, not vague promises.

Encrypted in transit and at rest

All data is protected using encryption both during transmission and while stored. Access is restricted through unique user IDs and role-based permissions.

Never used to train AI models

Your CRM data, research outputs, and agent activity are never used to train or fine-tune any language model. Not ours, not our providers'.

Workspace isolation

Each customer workspace is logically isolated. Your data is never accessible to other customers or shared across workspaces.

Minimal personal data processing

AI agents only process what they need: full name, job title, email, phone, and LinkedIn profile. Processing is limited to contacts matching your approved ICP and buyer personas. That is data minimisation under Article 5(1)(c), applied at collection.

You control retention

When you cancel, your data is returned or deleted at your choice. If no request is made, all personal data is permanently deleted within 30 calendar days.

Logging and monitoring

All access and processing activities are logged and monitored. Evergrowth maintains secure development practices, vulnerability management, and documented incident response procedures.

A documented lawful basis for every contact

Legitimate interest under Article 6(1)(f), which Recital 47 contemplates for direct marketing. The reason a person qualified is stored against their record, so the basis is specific to them rather than a blanket policy.

Rights requests, answered

Individuals keep their rights under Articles 15 to 21, including the right to object under Article 21. We assist you with any request, and the reason each contact qualified is already on the record.

Architecture safeguards

How the agent infrastructure is built

The plumbing behind the agents matters as much as the policies on top.

Review access and infrastructure controls

Authenticated entry only

Every request enters through a signed JWT (frontend) or API key (service-to-service). The token carries the requesting user’s organisation UUID, which scopes all downstream access. Users sign in with email and password or Google Workspace, and can add 2-factor authentication using a time-based code from an authenticator app such as Microsoft Authenticator or Google Authenticator, or a single-use code sent by email.

Cloud-hosted with managed encryption

Workloads run on AWS ECS with KMS-encrypted secrets and cluster-level audit logging. Persistent stores use managed Postgres with KMS encryption at rest and Multi-AZ availability for primary databases.

Mostly stateless agent service

The agent service holds no customer business records of its own. It orchestrates LLM calls and forwards validated outputs to authenticated downstream systems where the customer chooses to persist them.

Separation of instructions and data

System prompts are loaded from a controlled registry. Customer-supplied content and third-party web content enter the model only through delimited user-message blocks, never concatenated into the system prompt.

AI & your data

How AI agents handle your information

You’re trusting AI agents with your CRM data and prospect research. Here’s exactly how that works.

Review the AI safeguards

Enterprise model endpoints only

Agents reason via enterprise endpoints from Microsoft, AWS, OpenAI and Google, all under contractual terms that prohibit training on your data. Default routing is AWS Bedrock in the EU, which does not store prompts or completions. No consumer endpoints, and no providers in non-aligned jurisdictions.

Versioned prompts, server-side

System prompts are authored, peer-reviewed, and versioned in a controlled registry. New versions only reach customers after staging review. No silent prompt or model swaps.

Typed outputs, never free-form

Every agent response is bound to a typed schema. Outputs that fail validation are rejected, so an injected directive cannot reach a downstream system as free-form text.

Narrow tool sets

Each agent type has a hand-picked tool set: web search, scraping, internal lookup. No code execution, no filesystem, no shell, no arbitrary network. Writes to customer systems go through separate authenticated paths.

Workspace-scoped retrieval

Vector retrieval and embeddings are scoped per workspace and never shared across tenants. A query inside one customer environment cannot surface another customer’s data.

Full traceability

Every agent run produces a structured trace: prompt version, tool calls, inputs, outputs, latency, and token usage. Errors are captured in a monitored pipeline so reasoning is reconstructible after the fact.

AI risk frameworks

Mapped to the controls security teams ask about

Reviewers want to know how Evergrowth maps to OWASP and the EU AI Act. These four cards cover the most common questions; the full mapping is available under NDA.

Review the risk and control mapping

Prompt injection

System instructions are server-side and versioned. Third-party content, including web pages, CRM records and search results, is treated strictly as data inside delimited blocks. Outputs are bound to typed schemas. Maps to OWASP LLM01 + LLM05.

Sensitive data disclosure

Enterprise endpoints only, under contractual no-training terms. Each task receives the minimum context required to complete it. Maps to OWASP LLM02.

Excessive agency

Narrow tool sets per agent. No code execution, filesystem, or shell access. Writes to customer systems go through separate authenticated paths. Maps to OWASP LLM06.

Supply chain & poisoning

Foundation models are consumed via reviewed enterprise providers. Evergrowth does not train or fine-tune them, and never trains on customer data. Maps to OWASP LLM03 + LLM04.

Evergrowth’s agents perform B2B research, qualification, and copy generation. They do not make decisions in the high-risk domains listed in Annex III of the EU AI Act (employment, credit scoring, law enforcement, education access). A human reviews any action that affects external parties, and outputs are produced with provenance and reasoning visible to the user, supporting Article 52 transparency obligations.

Global compliance

Built for teams everywhere

Evergrowth complies with data protection regulations across the EU, US, UK, and Canada.

GDPR EU / UK

Full compliance with Regulation (EU) 2016/679. You are the controller; Evergrowth is the processor under Article 28, acting on your documented instructions through a formal DPA. Standard Contractual Clauses (SCCs) used for any transfers outside the EEA/UK.

CCPA / US State Privacy Laws US

Compliant with CCPA/CPRA, VCDPA, CPA, CTDPA, and UCPA. Evergrowth acts as a service provider or processor. Your data is never sold or shared as defined under these laws.

PIPEDA & Provincial Laws Canada

Compliant with PIPEDA and Quebec’s Law 25. Breach notification, access rights, and appropriate safeguards all covered.

When regulations overlap

Where multiple frameworks apply simultaneously, the more protective standard for data subjects governs. Always.

Legal

Everything your legal team needs

The contract itself, the data processing terms and the data flow overview, all published in full.

Master Services AgreementEU entities
Master Services AgreementUS entities
Data Processing AgreementAnnexed to every customer contract
Data Flow OverviewWhat we process and who receives it
Eva in Microsoft TeamsSecurity overview for IT reviewers
Privacy PolicyHow we collect and use data
Sub-processors ListAvailable on request
FAQ

Security questions we hear most

Data, privacy and contracts10 questions
Yes. The SOC 2 Type 2 report and the ISO 27001 certificate are available to satisfy audit obligations. Request them through your account manager or during a security review.
Only what’s needed: full name, job title, email address, phone number, and LinkedIn profile. Processing is limited to contacts matching your approved ICP and buyer personas, in line with GDPR data minimisation principles.
Legitimate interest under Article 6(1)(f) of the GDPR, which Recital 47 contemplates for direct marketing. 2 gates run before any external source is called: the company has to match the profile you approved, and the person a persona you defined. The reason each contact passed is recorded at that moment, so the basis is specific to the individual.
Article 14 applies where personal data came from somewhere other than the person, and it asks for 9 pieces of information, not 3 or 4. Most are facts about your own organisation: who you are, your DPO, your retention criteria, the rights, the right to complain, and the timing. Three are facts about the individual record and cannot be reconstructed later: the legitimate interest relied on, the categories of data held, and the source it came from. Evergrowth records those 3 against every contact at the moment it qualifies. Our GDPR guide for B2B outbound maps every item to its article and gives a notice template for the rest, and the data flow page shows the full table.
The Play Agent that writes your outreach can carry it, so the notice travels with the first message rather than depending on a rep to remember. It is not on by default, because 6 of the 9 Article 14 items are facts about your organisation that we do not hold. Add those to the Play Agent instructions once, with the wording from our guide, and every first-touch message carries it from then on.
Individuals keep their full rights under Articles 15 to 21, including the right to object under Article 21. You are the controller, so the request reaches you, and our DPA commits us to assist you in giving effect to it. Because every contact carries the reason it qualified and the source it came from, the answer does not have to be reconstructed.
A database gives you the record. It does not give you the reason. Article 14(2)(b) of the GDPR obliges you to tell a person which legitimate interest you relied on, for that individual and not as a general policy. Here it is generated and stored at the moment the contact qualifies. From a purchased list or a browser extension it has to be reconstructed by hand afterwards, if it can be at all.
You choose: data is returned or deleted. If no request is made, all personal data is permanently deleted within 30 calendar days of termination.
Only when necessary to provide the service, and always using Standard Contractual Clauses (SCCs) or equivalent safeguards under GDPR Chapter V.
Yes. A GDPR-compliant Data Processing Agreement is included as a standard annex in every customer contract.
AI providers and safeguards4 questions
No provider uses your data to train or improve their models. Our default routing is AWS Bedrock, which does not store prompts or completions. Where another provider handles a request, data may be retained briefly for abuse monitoring under that provider's published terms, and is never used for training.
System prompts are versioned and loaded server-side. Third-party content (web pages, CRM records, search results) is delimited as data, never concatenated into instructions. Outputs are bound to typed schemas, so an injected directive cannot reach a downstream system as free-form text.
Microsoft Azure AI Foundry, AWS Bedrock (Anthropic), the OpenAI Enterprise API, and Google AI Studio. All operate under contractual terms that prohibit training on your data. No consumer endpoints, and no providers in non-aligned jurisdictions.
No. Our agents perform B2B research, qualification, and copy generation. They do not make decisions in the Annex III high-risk domains (employment, credit scoring, law enforcement, education access). A human reviews any action that affects external parties.
Technical walkthrough1 question
Yes. Our data flow overview sets out what we process, who receives it and where it runs. Beyond that, we will walk your team through the architecture under NDA, alongside the SOC 2 Type 2 report and the ISO 27001 certificate.

Need a security review before you buy?

We’ll walk your security team through our architecture, certifications, and data handling. No sales pitch.

Schedule a security review