Book a demo
Topics · GOVERNANCE-AGENTS

Insurance Agent Identity DORA | Moring

DORA reaches insurance agents by function rather than by name: Regulation (EU) 2022/2554, applicable to insurers and reinsurers since January 2025, governs ICT-supported critical or important functions, and an agent that changes records or moves payments qualifies. Agent identity is the translation layer, giving every action an attributable actor, complete action logs, revocable credentials, and incident evidence that reconstructs without archaeology.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading
  5. Sources

Short answer

DORA reaches insurance agents by function rather than by name: Regulation (EU) 2022/2554, applicable to insurers and reinsurers since January 2025, governs ICT-supported critical or important functions, and an agent that changes records or moves payments qualifies. Agent identity is the translation layer, giving every action an attributable actor, complete action logs, revocable credentials, and incident evidence that reconstructs without archaeology.

How it works

DORA never says AI. Its obligations are function-based: any system, including an AI model or an autonomous agent, that supports a critical or important insurance function falls within scope because of what it does, not what it is called. That framing matters for identity, because the regulation's requirements attach to actions and systems, and actions can only be governed when each carries an attributable actor.

The framework applies to in-scope financial entities, including insurance and reinsurance undertakings, since 17 January 2025. Its ICT risk management chapter requires a documented, board-approved framework covering identification, protection, detection, response, and recovery for ICT systems supporting critical or important functions. An agent in claims triage, automated decisioning, or underwriting support sits inside that perimeter the moment it touches the function.

Five obligations translate directly into identity requirements. ICT risk management needs an inventory of actors, which means agents must be registered identities with owners, not anonymous processes. Logging and event reconstruction requires complete action logs, which means every agent action must be attributable to a specific identity and stored in a form that supports reconstruction without archaeology. Resilience testing includes stopping mechanisms, which means credential revocation and session termination must work under drill conditions. Third-party risk covers model and tool providers, which means contract terms should reflect how agent credentials and logs are handled. Incident reporting requires evidence assembled quickly, which depends entirely on attribution quality.

Identity also strengthens the database-level requirements that DORA implies for agent workloads. Append-only audit storage with per-agent identity attribution on every write, strong consistency, and tested failover turn the regulation's evidence expectations into storage properties rather than application promises. An agent record that cannot be altered after the fact is what makes event reconstruction credible to an examiner.

The incident scenario shows why identity is load-bearing. When an agent misbehaves in production, responders need to answer three questions quickly: which agent acted, under what authority, and what else shares that authority. With per-agent identities and time-limited credentials, revocation is targeted and the blast radius is known. With shared service accounts, the default response halts every workload behind the account, converting one incident into several.

The practical alignment path is concrete: register every agent as an identity within the ICT risk framework, enforce scoped and revocable credentials, retain tamper-evident action logs, rehearse a stop drill, and ensure provider contracts address logging and credential control. Carriers that operate this way treat DORA not as an AI regulation to interpret but as an operations discipline their agents already satisfy.

Frequently asked questions

Does DORA apply to AI agents in insurance at all?

Yes, by function. DORA, Regulation (EU) 2022/2554, applies to in-scope financial entities including insurers and reinsurers since 17 January 2025, and its scope is defined by what a system does rather than what it is called. An agent that supports claims triage, automated decisioning, or underwriting within a critical or important function is an ICT-supported function in every sense that matters, so it inherits the framework's risk management, logging, resilience, and incident obligations.

Which DORA obligations touch agent identity most directly?

Four: ICT risk management, which needs agents inventoried as registered identities with owners; logging and event reconstruction, which requires every action attributable to a specific agent and stored for reconstruction; resilience testing, which implies credential revocation and session termination that work under drill conditions; and incident reporting, where evidence quality depends entirely on attribution. Third-party risk provisions also reach the model and tool providers behind agent credentials and logs.

How does agent identity improve incident response under DORA?

When an agent misbehaves, responders must quickly determine which agent acted, under what authority, and what else shares that access. Per-agent identities and time-limited credentials make revocation targeted and the blast radius known, so containment stops one workload. Shared service accounts force the opposite: halting every process behind the account and converting a single incident into several. Attribution quality in the identity layer directly determines both response speed and the completeness of incident evidence.

What storage properties support DORA evidence for agent actions?

Append-only audit storage enforced at the storage layer, per-agent identity attribution on every write, strong read consistency, and tested failover. These properties turn the regulation's reconstruction expectations into system guarantees rather than application-level promises, so a tamper-evident agent action record survives operational stress. Combined with retained policy decisions, the storage layer lets an insurer reconstruct any automated underwriting or claims event in the form examiners expect.

Sources

  1. cockroachlabs.com
  2. trussed.ai
  3. moring.ai
  4. dataiku.com
  5. getagentid.com
  6. omadaidentity.com
  7. akamai.com
  8. - content.naic.org
  9. - artificialintelligenceact.eu
  10. - eur-lex.europa.eu