Glossary

AI Compliance

Meeting the laws that apply to an AI system: AI-specific regulation, the sector rules you already follow, and privacy law. For most companies it comes down to documentation, logging, human oversight and knowing where the data goes.

General definition

AI Compliance is the subset of AI governance that deals with legal obligations. Three bodies of law usually apply at once. AI-specific laws, led by the EU AI Act, classify systems by risk and attach duties to each tier. Sector rules written before LLMs existed still apply to them: HIPAA when an agent touches health information, FINRA and SEC supervision and record-keeping rules when a chatbot talks to investors, consumer protection law when an agent sells. Privacy law applies whenever the inputs or outputs are personal data: the GDPR’s transparency duties, its Article 22 limits on solely automated decisions with legal effect, and its data minimisation and storage limitation principles.

Regulators have said repeatedly that their rules are technology neutral. FINRA’s Regulatory Notice 24-09 (2024) reminded firms that supervision, record-keeping and communications rules cover generative AI tools exactly as they cover any other technology. HIPAA does not mention AI, but a vendor whose model processes protected health information is a business associate and needs a BAA. In the United States, state laws are adding AI-specific duties on top, for example disclosure and impact assessment requirements for high-risk automated decisions.

In practice, compliance for a deployed AI chat system reduces to a short list of evidence:

  • Documentation: what the system does, what model and data it uses, its intended purpose and known limitations, and the risk assessment behind it (a DPIA where personal data is involved)
  • Logging: an audit log of inputs, outputs, model versions and operator actions, kept for the period the applicable rule requires
  • Human oversight: a person who can review, override and stop the system, with human-in-the-loop approval for consequential actions
  • Transparency: telling people they are talking to an AI, and explaining automated decisions where the law requires it
  • Data handling: lawful basis, retention limits, deletion, residency and contracts with every provider in the chain

The hard part is rarely the policy; it is proving it afterwards. A regulator or auditor asks for the transcript of a specific conversation, the version of the model that produced it and who approved that version. Systems deployed without those records cannot be made compliant retroactively.

In the Ethora ecosystem

Ethora supplies the evidence layer that AI compliance depends on, inside a platform built for regulated industries. Conversations with AI agents are ordinary messages in rooms, so they land in the server-side archive with configurable retention and per-user erasure, and in the compliance audit trail with immutable log export to S3. Each agent has a declared system prompt, model and knowledge base that can be documented in a technical file. The Trust and Safety system adds a control point for blocking or quarantining content an agent should not send, and a human member of the room can be kept in the loop before anything consequential happens.

For healthcare customers, Ethora signs a Business Associate Agreement on Enterprise and dedicated plans and provides the technical safeguards; the covered entity still owns its own compliance. A DPA covers GDPR processing, and dedicated or self-hosted deployments keep transcripts and prompts in the region you choose. Ethora does not hold SOC 2, ISO 27001 or ISO/IEC 42001 certification; what it offers is an architecture that makes those controls yours to run and to evidence.

Get started

Chat and AI built for regulated industries

Ethora gives you audit trails, retention controls, a Trust & Safety layer and dedicated deployments in your own cloud. Talk to our team.

Start Free
Free tier available Enterprise SLA No vendor lock-in