AI compliance built into your agent platform

Banks, insurers and governments answer to the EU AI Act, GDPR, DORA and their central bank. Blits.ai gives every agent the controls those rules ask for, and the records you show your regulator.

Who is responsible for what?

No platform makes an organization compliant on its own. The rules put the duty on the organization that puts the agent in front of people. This is how the work divides.

Blits.ai, as your supplier

  • Certified for ISO 27001, SOC 2 Type II, PCI DSS Level 1 and ISO/IEC 42001
  • Processes personal data for you under a data processing agreement
  • Runs your tenant in the region you choose, or on your own servers
  • Tells you before a model or a sub processor changes
  • Publishes an SLA and a status page

The platform, in your hands

  • Approvals, a stop button, guardrails and an AI notice per agent
  • A decision log for every conversation, and an audit log of every change
  • An AI register of all your agents, ready to export
  • Test suites, bias checks and monitors that run on a schedule
  • Masking, consent, retention periods and forget me requests

Your organization

  • Classifies each use case and sets its risk class
  • Writes the DPIA, and the fundamental rights impact assessment where required
  • Gets board approval, and regulator approval where the rule asks for it
  • Files the register and reports incidents to the regulator
  • Decides when an agent goes live, and staffs the reviews

Which controls does the platform give you?

Every control below is part of the platform. Under each one you see the rules that ask for it, with the article or section, so your compliance team can check it in the source.

Tell people, and keep a person in control

AI notice

Every assistant tells people they are talking to AI, in their own language. A voice agent says it when the call starts. You set the text per language.

Asked for by

Approval per tool

Each tool runs automatically, waits for approval, or is switched off. Tools that move money or change data wait for approval by default. The approver sees the exact call that will run, and can approve in the conversation or from the Approvals inbox, optionally only above an amount.

Asked for by

Review by a person

When an agent tells a customer about a decision, the customer can ask a person to review it. The request lands in the Approvals inbox with a reference, a due date, the conversation and the decision log. The employee upholds or changes the decision, and the customer hears back on the same channel.

Asked for by

Stop button

One button stops an agent on every channel at once; a second one stops every agent of the tenant. Open conversations go to a live agent or get a fixed message, and scheduled tasks and campaigns pause. Who stopped it, when and why goes into the audit log.

Asked for by

Keep each agent within its purpose

Guardrails

Rules, blocked topics and a judge model check what users ask and what the agent answers. A content scanner runs on every assistant. The guard models run inside the platform, so these checks send no data to an outside provider.

Asked for by

Answers from your own sources

Agents answer from a knowledge base you approve and cite the source document. A policy can force the agent to search it before every answer.

Asked for by

Testing and monitoring

Test suites run real conversations against an agent before each release, and scheduled monitors repeat them in production. Analytics, anomaly review and user feedback show how the agent behaves.

Asked for by

Bias check

A test suite template sends the same request many times, changing only the name, gender, age or nationality. The report shows whether the outcome differs per group. Schedule it once a year and after every material change.

Asked for by

Version history and dry run

Every change to an assistant or prompt is a version you can roll back. A dry run shows what an agentic task would do before it acts.

Asked for by

The records your regulator asks for

Decision log

For every conversation: the routing decisions, each tool call with masked inputs and results, guardrail checks, approvals and the model used. Export it as JSON or stream it over OpenTelemetry to your SIEM. A separate audit log records every change platform users make.

Asked for by

AI register

One table of every agent, assistant and workflow in your tenant. The platform fills in the model, provider, region, channels and tools; you add the owner, purpose, risk class and next review date. Export it as CSV for your regulator.

Asked for by

Retention periods

Set how long conversations are kept, for the tenant and per agent, and how long the decision log is kept, from 30 days to ten years. A nightly cleanup deletes what is older, including attachments, transcripts and recordings.

Asked for by

Notice of model changes

When a model provider retires a model, every tenant that uses it hears from us as soon as the date is known, and again 14 days before. The notice names the agents it affects, lets you run their test suites on the new model, and stays in a change history you can show.

Asked for by

Data, and Blits.ai as your supplier

Privacy Shield and consent

Names, account and card numbers, national IDs and more are masked before they are stored or sent to a model. Consent is asked and recorded before a conversation starts, a forget me request deletes a user's data, and conversation history exports as CSV.

Asked for by

Data location and model choice

Run in a managed region in the EU, Qatar, the UAE, Singapore, the US or Australia, in your private cloud or on premises. Limit which AI providers may receive your data, use your own keys or model endpoints, and set a fallback model.

Asked for by

Access and security

62 separate permissions checked on every request, roles, tenants kept apart, multi factor sign in and encrypted secrets. Custom code runs in a sandbox.

Asked for by

Continuity

Provider fallback keeps agents answering when a model provider fails. The SLA, the status page and handover to live agents cover the rest.

Asked for by

Exit and data return

Export assistants, conversation history, consent records and decision logs at any time, in formats another system can read.

Asked for by

Certified supplier

Blits.ai is certified for ISO 27001, SOC 2 Type II, PCI DSS Level 1 and ISO/IEC 42001. Certificates and reports are available for your due diligence through the Trust Center.

Asked for by

Which control covers which rule?

Which control helps with which rule. Hover or tap a check to see the article.

ControlEU AI ActGDPRDORAISO/IEC 42001NIST AI RMFMASQCBCBUAE
AI noticeArticle 50(1) and (5)Annex B, B.8.2MAP 5.1Paragraph 5.6Section 20.1Guidance Note 4(a) and 4(b)
Approval per toolArticles 14(4) and 26(2)Article 22(3)Annex B, B.9.2 to B.9.4GOVERN 3.2; MAP 3.5Paragraph 5.9Sections 13.1 and 13.4Guidance Note 7(a)
Review by a personArticle 86Article 22(3)Section 21.2Guidance Note 7(c)
Stop buttonArticle 14(4)(e)MANAGE 2.4Paragraph 5.11Section 13.4.4Guidance Note 6(f)
GuardrailsArticle 5(1)(a) and (b)MEASURE 2.7Paragraph 5.15Sections 13.6.2 and 13.7.5Guidance Note 7(d)
Answers from your own sourcesArticle 26(4)Annex B, B.7.4MEASURE 2.5Guidance Note 7(d)
Testing and monitoringArticle 26(5)Annex B, B.6.2.4 to B.6.2.6MEASURE 2.3, 2.4 and 3.3Paragraphs 5.14 and 5.23Sections 15.6 and 19.1Guidance Note 6(a), 6(c) and 6(d)
Bias checkMEASURE 2.11Guidance Note 3(c)
Version history and dry runArticle 9(4)(e)Annex B, B.6.2.5
Decision logArticles 12, 19 and 26(6)Articles 5(2) and 33(5)Articles 10 and 17Annex B, B.6.2.8MANAGE 4.1 and 4.3Paragraphs 5.9(d) and 5.23Sections 19.3.1 and 19.3.3Enabling Technologies Guidelines 3.97(a)
AI registerArticle 28(3)GOVERN 1.6Paragraph 4.9Sections 10.1 and 10.6Guidance Note 2(f)
Retention periodsArticle 26(6)Article 5(1)(e)
Notice of model changesArticle 30(3)(b)Annex B, B.10.3MANAGE 3.1Paragraph 5.11(d)
Privacy Shield and consentArticles 5(1)(c), 7(1), 15, 17, 20 and 25Annex B, B.7.4 and B.7.5MEASURE 2.10Paragraph 5.4(e) and (f)Sections 17.2 and 17.9Consumer Protection Regulation 6.1.2.5
Data location and model choiceArticles 28(2) and 44Articles 28(8), 29 and 30(2)(b)Annex B, B.4.4GOVERN 6.1 and 6.2Paragraph 5.11(c) and (e)Cloud Computing Regulation 21.4Consumer Protection Standards 6.1.6.3
Access and securityArticle 32(1)Article 9(4)(c) and (d)Paragraph 5.16(b)
ContinuityArticles 11 and 30(3)(a)GOVERN 6.2
Exit and data returnArticle 20Article 30(2)(d)
Certified supplierArticle 28(5)Annex B, B.10.3Paragraph 5.11(a)Sections 9.5.2 and 12.2Guidance Note 9(a)

What does each rule ask of your AI agents?

What each rule asks, what the platform covers and what stays with you. Every rule has a full guide with the timeline, the duties per role and the sources.

EU AI Act

European Union. Transparency duties for chatbots and voice agents apply from August 2, 2026; duties for high risk uses such as credit scoring from December 2, 2027.

The platform helps with

Stays with you

Classify each use case, carry out the fundamental rights impact assessment where Article 27 requires it, and train your staff in AI literacy.

Read the full guide

GDPR

European Union. Applies to every agent that handles personal data of people in the EU.

The platform helps with

Stays with you

Choose a lawful basis per purpose, carry out the DPIA, keep your record of processing and notify breaches within 72 hours.

Read the full guide

DORA

European Union. Applies to banks, insurers, payment institutions and other financial entities, and to their ICT suppliers.

The platform helps with

Stays with you

Map the agent in your ICT risk framework, keep the register of information, classify and report incidents, and keep a tested exit plan.

Read the full guide

ISO/IEC 42001

International standard. Voluntary, and often asked for in tenders and supplier reviews.

The platform helps with

Stays with you

Run your own AI management system: the policy, the risk and impact assessment per agent and the management review.

Read the full guide

NIST AI RMF

United States. A voluntary framework that many US banks and their supervisors use as a reference.

The platform helps with

Stays with you

Set your risk tolerance, decide whether each agent goes live and involve independent assessors.

Read the full guide

MAS

Singapore. Guidelines for financial institutions supervised by the Monetary Authority of Singapore, issued in October 2026.

The platform helps with

Stays with you

Get board approval, rate the materiality of each AI use and test third party AI in your own use case.

Read the full guide

QCB

Qatar. Guideline for financial institutions licensed by Qatar Central Bank.

The platform helps with

Stays with you

Keep an AI strategy, classify each system, seek QCB approval where the guideline asks for it and send the register to QCB every year.

Read the full guide

CBUAE

United Arab Emirates. Guidance for financial institutions licensed by the Central Bank of the UAE.

The platform helps with

Stays with you

Keep the inventory and risk rating, run the yearly bias test, keep records for five years and get written board approval for telemarketing, also by AI.

Read the full guide

Questions about AI compliance

Talk to us about your regulator

Tell us which rules apply to you, and we show you the controls and records for each one on a live agent.

Book a demo