How we protect your data

Security, privacy, and compliance for sensitive legal data

The short version

Where it is stored
EU servers (Supabase), encrypted at rest with AES-256. Nothing is stored outside the EU.
Who processes it
Claude accessed through Amazon Bedrock in AWS’s London region in the United Kingdom (AWS London region), after redaction. Crisis alert emails go through SendGrid (United States) carrying a reference number and severity only.
How long it is kept
Your firm's configured retention period, 365 days by default, then deleted. Amazon Web Services (Bedrock): zero data retention; nothing written to durable storage.
What is redacted
On by default: phone numbers, NHS and NI numbers, card and bank details, email addresses, postcodes, dates of birth and street addresses are stripped before analysis. Names stay so you can match repeat callers and handlers. With redaction on, the original transcript is deleted once analysis completes.
Model training
Never. Your transcripts are not used to train any AI model.

HarkIQ processes call transcripts that contain some of the most sensitive information a law firm holds: client health details, vulnerability indicators, legal proceedings, and financial circumstances. We built HarkIQ knowing that this data demands the highest level of care. Here is exactly how we handle it.

Looking for the full briefing for your DPO? Download the DPO Briefing Document or the DPIA Template.

Your data, your control

You are the data controller. We are the data processor.

Your firm decides what data to upload and how the analysis is used. HarkIQ processes transcripts solely on your instructions, to provide the analysis service. We never make independent decisions about your data.

Your data is never shared with other firms.

HarkIQ is multi-tenant with strict data isolation. Every firm’s data is segregated at the database level using PostgreSQL row-level security. No firm can ever see, access, or be affected by another firm’s transcripts or analysis. This is enforced at the infrastructure level, not just the application level.

Your data is never sold, licensed, or monetised.

We do not sell your data. We do not licence it to third parties. We do not use it for marketing purposes. Your transcripts and analysis exist solely to provide you with the HarkIQ service.

You can export or delete your data at any time.

Every analysis can be exported as a PDF report or CSV file directly from your dashboard. Data is kept for the retention period your firm configures (365 days by default), then deleted. Everything can be exported on request, including after you cancel.

How AI analysis works

HarkIQ uses the Anthropic Claude API, a closed AI model.

When you upload a transcript, it is sent to the Anthropic Claude API for analysis. This is important to understand:

  • PII redacted by default: phone numbers, NHS and NI numbers, card and bank details, email addresses, postcodes, dates of birth and street addresses are removed from every transcript before analysis. Names are kept so your firm can match repeat callers and handlers.
  • Closed model: your transcripts are not used to train the AI. Claude is accessed through Amazon Bedrock, account pinned to zero data retention: no request or response written to durable storage by AWS or shared with Anthropic; not used for training.
  • Zero data retention: the Bedrock account is pinned to zero data retention, so no request or response is written to durable storage by AWS or shared with Anthropic.
  • No human review: requests are processed by the model inside AWS. Neither AWS nor Anthropic staff can read them, because nothing is retained.
  • Encrypted in transit: all data sent to the API is encrypted using TLS.

What the AI does and does not do

HarkIQ’s AI analyses each transcript against eight quality dimensions and returns a structured score, vulnerability flags, and recommendations. It does not:

  • Record, intercept, or listen to live calls. HarkIQ only processes transcripts your firm has already created.
  • Make decisions. Scores and flags are management intelligence tools. Your professionals apply the judgement.
  • Keep transcripts longer than needed. Anthropic holds inputs for a limited period for safety monitoring and then deletes them; it never uses them for training.

Personal identifiers are stripped before analysis

HarkIQ automatically redacts personal identifiers and financial details from every transcript before it is sent to the AI for analysis. This is on by default for every firm. Names are the deliberate exception: they stay in so your firm can match repeat callers and handlers.

Handler
Good morning, new enquiries, you’re speaking to Priya. How can I help?
Caller
Hello, it’s Margaret Doyle. My GP said to call. My NHS number is [NHS NUMBER REDACTED] and my date of birth is [DOB REDACTED].
Handler
Thank you, Margaret. What’s the best number to reach you on, and your address?
Caller
[PHONE REDACTED]. It’s [ADDRESS REDACTED], Leeds, [POSTCODE REDACTED]. I’ve been struggling since the fall, I can’t manage the stairs.
Handler
I’m sorry to hear that. Let me take some details and get the right person to call you back today.
Handler
Good morning, new enquiries, you’re speaking to Priya. How can I help?
Caller
Hello, it’s Margaret Doyle. My GP said to call. My NHS number is 485 777 3456 and my date of birth is 14 March 1962.
Handler
Thank you, Margaret. What’s the best number to reach you on, and your address?
Caller
07700 900123. It’s 22 Alder Road, Leeds, LS6 4QP. I’ve been struggling since the fall, I can’t manage the stairs.
Handler
I’m sorry to hear that. Let me take some details and get the right person to call you back today.
As uploadedWhat the AI sees
Drag the handle, or focus it and use the arrow keys. Synthetic transcript: every name, number and address is invented. Names stay in so repeat callers and handlers can be matched.

What is removed before AI analysis:

  • Phone numbers
  • Email addresses
  • Addresses and postcodes
  • NHS numbers
  • NI numbers
  • Card numbers and expiry dates
  • CVV/security codes
  • Bank sort codes and account numbers
  • IBANs
  • Dates of birth

What the AI sees:

The conversation itself: what was said, how it was said, whether the caller was handled well, and the names used in it.

What happens to the original:

When redaction is on (the default), the original transcript is deleted immediately after analysis and only the redacted copy is stored. Only the redacted copy is available in your dashboard.

Why this matters:

Call quality scoring is based on how a conversation was handled: the empathy, the listening, the information gathering, the compliance. None of that requires a caller’s phone number, address or bank details. By stripping those before they reach the AI, we reduce the sensitivity of the data being processed and simplify your firm’s data protection position.

Security architecture

HarkIQ’s security is designed for the sensitivity of legal sector data.

Encryption at rest

All stored data is encrypted using AES-256 via Supabase managed encryption.

Encryption in transit

All communications between your browser, our servers, and the AI API use TLS.

Row-level security

Database-level isolation ensures each firm’s data is completely separated. This is enforced by PostgreSQL, not application code.

Role-based access

Four permission levels (Admin, Manager, Analyst, Viewer) control who sees what within your firm. Handler performance data is restricted to management roles.

Authentication

Email and password login with optional multi-factor authentication. Session tokens with configurable expiry.

Audit trail

Every data access event, analysis, and user action is logged with timestamp and user identity for compliance reporting.

PII redaction

On by default. Phone numbers, email addresses, postcodes, NHS and NI numbers, card numbers, bank details, IBANs, and dates of birth are automatically stripped from transcripts before AI analysis. Names are kept for repeat-caller matching.

Rate limiting and input validation

All API endpoints are rate-limited and validated using structured schemas to prevent abuse.

Where your data is stored and processed

All stored data (transcripts, analyses, and firm accounts) is held on EU servers. AI analysis stays in AWS’s London region (Claude accessed through Amazon Bedrock in AWS’s London region, account pinned to zero data retention: no request or response written to durable storage by AWS or shared with Anthropic; not used for training). One thing leaves the UK and EU: crisis alert emails sent through SendGrid (United States), which carry only a call reference number and a severity level.

Transparency about where data flows is important. A full list of services, their data processing locations, and transfer mechanisms is set out in the International data transfers section below.

Compliance and governance

UK GDPR

HarkIQ is designed for UK GDPR compliance. We operate as a data processor under Article 28. A Data Processing Agreement is available for all customers and is provided during onboarding. Our DPA covers processing scope, security measures, breach notification, sub-processor management, and data deletion.

Special category data

We recognise that call transcripts may contain special category data under UK GDPR, including health information and details of legal proceedings. Our security architecture, data isolation, closed AI model, and retention policies are specifically designed to handle this classification of data appropriately.

Data Protection Impact Assessment

We provide a DPIA template to help your firm’s DPO assess the data protection implications of using HarkIQ. This covers the data processed, lawful basis guidance, risk assessment, and the technical mitigations HarkIQ provides. Download our pre-populated DPIA template to get started. It includes HarkIQ-specific information covering data flows, sub-processors, risk assessments, and recommended measures, with fillable fields for your firm’s DPO to complete.

Breach notification

In the event of a personal data breach, we will notify affected customers within 72 hours, in line with UK GDPR requirements. Notification includes the nature of the breach, data subjects affected, likely consequences, and measures taken.

Our approach to SRA compliance

HarkIQ has been designed in line with the SRA’s warning notice on the misuse of AI, published 17 August 2026, and its earlier compliance tips on the use of AI and technology. We recommend that every firm considering HarkIQ reads both before making a decision: SRA warning notice: misuse of AI (17 August 2026) and SRA compliance tips for solicitors on AI and technology.

The sections below set out how HarkIQ addresses each of the SRA’s requirements.

COLP involvement

What the SRA expects from your firm

The SRA is explicit: the Compliance Officer for Legal Practice (COLP) should be responsible for regulatory compliance when new technology is introduced, with board-level oversight of both the purchasing decision and ongoing use.

Before going live with HarkIQ, we recommend your firm does the following:

  • Involve your COLP in approving the decision to use HarkIQ. Download the DPO Briefing Document and DPIA Template to support their review, and contact us at david@harkiq.com for our Data Processing Agreement.
  • Ensure board oversight is in place, both for the initial decision and for monitoring ongoing use.
  • Complete a risk and impact assessment. Our DPIA template gives you a ready-made starting point tailored to HarkIQ’s specific processing activities.
  • Create a brief internal policy covering how HarkIQ outputs should and should not be used; in particular, that AI scores support professional judgement and are not used as the sole basis for any consequential decision about a client or a member of staff.
  • Provide awareness training to anyone who will use HarkIQ, covering what the scores mean, how vulnerability flags should be acted on, and the limits of AI analysis.

Automated decision-making

Our position under UK GDPR

UK GDPR Article 22 gives individuals the right not to be subject to decisions based solely on automated processing where those decisions produce significant legal or similarly significant effects.

HarkIQ does not constitute automated decision-making within the meaning of Article 22. HarkIQ produces scores, flags, and recommendations. It does not make decisions. Every output requires a human professional to review it and apply their own judgement before any action is taken. HarkIQ has no authority to take any action affecting a client or a member of staff.

Your internal policy should confirm that HarkIQ scores are not used as the sole basis for any consequential decision, including disciplinary action, performance management, or safeguarding referrals. The score informs. The professional decides.

If HarkIQ produces an incorrect result

AI analysis is not infallible. A transcript may be mis-scored because of transcript quality, unusual call content, or the inherent limitations of AI pattern recognition.

HarkIQ scores are management indicators, not verified assessments. If a score appears inconsistent with your professional experience of a call or handler, treat your professional judgement as the primary source. Do not act on a score that does not make sense to you without first reviewing the underlying transcript.

If you believe HarkIQ has produced a materially incorrect analysis, contact us. We will investigate, and where the issue is with the model or prompt we will correct it. We maintain version control of all AI prompts and can re-analyse transcripts if required.

Your firm retains full professional responsibility for any decision made using HarkIQ outputs. HarkIQ is a tool. The solicitor remains accountable.

International data transfers

How we handle US-hosted services

HarkIQ’s database is hosted entirely within the European Union. All transcripts, analysis results, and firm account data are stored on EU servers and never leave the EU for storage purposes. AI analysis stays in AWS’s London region (Claude accessed through Amazon Bedrock in AWS’s London region, account pinned to zero data retention: no request or response written to durable storage by AWS or shared with Anthropic; not used for training). One thing leaves the UK and EU: crisis alert emails sent through SendGrid (United States), which carry only a call reference number and a severity level.

The sub-processor table below lists several US-based services. Seeing “United States” in that table does not mean your data is unprotected. UK GDPR does not prohibit international transfers; it requires appropriate safeguards. The position for each service that handles personal data:

ServiceWhat it doesLocationReceives transcript data?Transfer mechanism
SupabaseDatabase and authenticationEuropean UnionYes: stores all data on EU servers. AES-256 encryption at rest.No international transfer. All data stored on EU servers.
Claude accessed through Amazon Bedrock in AWS’s London regionAI transcript analysisUnited Kingdom (AWS London region)Yes: PII-redacted transcripts. Zero data retention; nothing written to durable storage. Not used for training.No international transfer for AI analysis: processing stays in AWS’s London region under the AWS Customer Agreement and its GDPR Data Processing Addendum.
SendGrid (Twilio)Crisis alert and account emailsUnited StatesNo: a call reference number and severity level onlyTwilio standard terms incorporating Standard Contractual Clauses with the UK Addendum; under review
VercelWeb application hostingUS with global edge nodesNoNot applicable
StripePayment processingUnited StatesNo: billing details onlyNot applicable
HubSpotMarketing emailsUnited StatesNo: email address and firm name onlyNot applicable

Anthropic’s full Data Processing Addendum, which includes the UK Addendum covering transfers from the UK to the US, is published at anthropic.com/legal/data-processing-addendum.

Our Data Processing Agreement sets out the transfer mechanism for each sub-processor in full. Enterprise customers can request copies of all sub-processor DPAs by contacting david@harkiq.com.

HarkIQ is a management intelligence tool.

It provides AI-generated scores, vulnerability flags, and recommendations to help law firms understand call quality patterns, identify coaching opportunities, and evidence compliance improvement.

HarkIQ should not be used as:

  • The sole basis for disciplinary action against individual handlers
  • A clinical or safeguarding determination about individual callers
  • Regulatory reporting or SRA self-reporting without independent verification
  • A substitute for professional judgement on vulnerability or risk

Vulnerability detection is an assistive flag aligned with the FCA’s four-driver model. It highlights calls that may warrant further attention. Your professionals make the decisions.

Documents for your Data Protection Officer

Two documents written for Data Protection Officers and compliance leads at law firms considering HarkIQ.

DPO Briefing Document

Full briefing covering data flows, Law Society GDPR guidance mapping, sub-processors, international transfers, and security architecture. 18 sections.

DPIA Template

Fillable PDF form pre-populated with HarkIQ’s information. Your DPO completes the firm-specific sections, records the approval decision, and retains it as part of your data protection records. 10 sections, 41 interactive fields.

Our Terms of Service, Privacy Policy, and Data Processing Agreement are also available on request. Email david@harkiq.com.

Questions?

If you have questions about how HarkIQ handles your data, or if your DPO needs additional information before approving HarkIQ for use, contact us at david@harkiq.com. We are happy to provide our Data Processing Agreement or discuss specific security requirements.

Download the DPO Briefing Document and DPIA Template above, or email david@harkiq.com for our Terms of Service, Privacy Policy, and Data Processing Agreement.