AI risk is reshaping cyber insurance underwriting and attack diversity

At a glance

What this is: Cowbell frames AI as a three-sided insurance issue of promise, peril, and vector, with autonomous agents, observability, and governance becoming underwriting inputs.

Why it matters: This matters to IAM and security teams because AI risk is now tied to access scope, orchestration, and decision autonomy, not just model performance or policy wording.

By the numbers:

• Cowbell says its platform spans 55 million entities across its global risk pool.

Context

AI is no longer just a model risk topic. It now affects how organisations assess, delegate, and govern access to data, tools, and decisions across underwriting, incident response, and operational workflows.

Cowbell's argument sits at the intersection of cyber risk and identity governance: once AI systems are allowed to act across multiple workflows, the key question becomes how much authority they hold and how tightly that authority is bounded.

Key questions

Q: How should organisations govern explainable AI systems in regulated workflows?

A: Start by assigning accountable owners, defining review points, and documenting how decisions can be challenged or corrected. Explainability only matters operationally when it supports auditability, human override, and evidence collection. If those controls are missing, the organisation may have a model that performs well but still cannot defend its decisions to regulators or customers.

Q: Why do autonomous AI workflows change cyber risk and loss modelling?

A: Because they compress the time between decision and effect. A human-centered control model assumes there is time for review, escalation, and containment, but autonomous workflows can complete actions before those checks happen. That changes both incident severity and insurer confidence in the controls that supposedly limit loss.

Q: What are the warning signs that GenAI governance is failing in production?

A: Common signs include rising token spend without clear business growth, poor traceability of prompt flows, inconsistent routing between models, and backend services that still contain long-lived API keys. If teams cannot explain who is calling what and why, governance is not working.

Q: What is the difference between AI observability and AI governance?

A: AI observability tells you what the system did. AI governance decides whether it should have been allowed to do it, who approved it, and what happens when it crosses a policy boundary. Observability is a data problem. Governance is an operating model that combines policy, ownership, evidence, and enforcement.

Technical breakdown

Why autonomous AI agents change the underwriting control model

Autonomous AI agents do not just assist a workflow. They can select actions, sequence tools, and trigger outcomes with far less human intervention than traditional automation. That matters because insurance and cyber controls often assume a person, system owner, or analyst remains the stable actor behind each decision. When the actor becomes an agent, governance has to move from periodic review to runtime authorisation, logging, and containment. The real control question becomes whether each action is explicitly scoped, attributable, and revocable before downstream effects occur. See the AI Agent Authorisation Guide.

Practical implication: Treat AI agents as governed identities with task-scoped authority, not as generic automation.

How observability and governance become risk inputs rather than afterthoughts

The article puts observability and governance inside the risk model, which is the right direction. In AI-heavy operations, visibility is not just a monitoring feature. It is the evidence layer that shows what the system did, why it did it, and whether the action stayed within approved boundaries. Without that evidence, both loss analysis and incident response become guesswork. In practice, observability has to cover prompts, tool use, data access, policy decisions, and human overrides so that insurers and operators can distinguish acceptable AI behaviour from risky drift. See the AI Agent Observability, Audit and Incident Response Guide.

Practical implication: Instrument AI workflows so every material action leaves an auditable trail.

Why AI risk now overlaps with secrets, pipelines, and orchestration

The article describes AI as a stack of compute, tokens, orchestration, data pipelines, governance, and model management. That stack creates identity-adjacent risk because each layer may depend on credentials, service access, or delegated authority. The weak point is rarely the model alone. It is the surrounding control plane that lets an AI system reach data, invoke tools, or influence decisions. This is where NHI governance becomes relevant: if an AI system can use long-lived credentials or overbroad service access, the insurance question quickly turns into a security governance problem. See the Agentic AI Identity Guide.

Practical implication: Map every AI workflow to the credentials, APIs, and service accounts it can reach.

Threat narrative

Attacker objective: Adversaries aim to expand attack speed, precision, and campaign diversity while reducing the defender's time to detect, assess, and respond.

1. Entry occurs when AI lowers the barrier for reconnaissance, phishing personalisation, malware development, and other initial attack activity.
2. Escalation follows when adaptive tooling lets adversaries scale those attacks faster than traditional defensive review cycles can absorb.
3. Impact emerges when defenders face compressed response windows, more varied attack vectors, and greater uncertainty in loss estimation and control effectiveness.

Breaches seen in the wild

• AI agent retail card theft campaign 2026: AI agents breached 27+ retailers for about $25 each, used cloud keys and a Secrets Manager dump, and stole 600,000+ payment cards.
• Spain's first AI agent data breach 2026: Spain's AEPD logged its first breach notification attributed to an attacker's AI agent, which altered personal data and accessed invoices.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

AI risk is now an identity governance problem, not just a model governance problem. Once AI systems can act across underwriting, incident response, and workflow automation, the key issue becomes who or what is authorised to do what, under which conditions, and for how long. That shifts the centre of gravity from policy statements to delegated authority, auditability, and revocation. For practitioners, the practical conclusion is to govern AI systems as actors with defined identity boundaries.

Autonomy without observability creates a pricing and control blind spot. Cowbell's framing correctly puts observability alongside autonomy and governance because risk cannot be priced or contained when actions are opaque. When AI systems make decisions, invoke tools, or move through workflows without a durable evidence trail, neither underwriting nor security teams can tell whether a loss came from misuse, misconfiguration, or expected behaviour. Practitioners should treat observability as a governance control, not a telemetry afterthought.

Closed-loop risk management will increasingly depend on whether security evidence is machine-readable. If AI is going to support continuous underwriting, the surrounding controls must produce structured signals that can be consumed by risk engines and incident workflows. That means identity context, access scope, policy decisions, and anomaly indicators need to be attributable and timely. The field is moving toward systems where control quality is measured by the quality of the signals they emit.

AI creates a new class of control debt: governance that exists on paper but not at runtime. The article's emphasis on autonomy, data pipelines, and orchestration shows why static policy is insufficient. If an AI system can continue operating after context changes, stale access and weak oversight become embedded risk rather than exceptional failure. Practitioners should close the gap between stated policy and runtime enforcement before AI-driven workflows become business-critical.

From our research library:

• 53% of security leaders expect AI to run major portions of their infrastructure autonomously within the next three years, according to the 2026 Infrastructure Identity Survey.
• Read next: AI Agent Authorisation Guide

What this signals

AI risk management is moving toward runtime governance. Organisations will increasingly need policy checks, access boundaries, and evidence trails that operate at the moment of action, not after a workflow completes. Static approval alone will not be enough once AI systems participate in core operational decisions.

Machine-readable control signals will matter more than narrative assurance. If AI-enabled underwriting and response workflows are to be trusted, they must emit structured evidence that security, risk, and finance teams can consume without manual translation. That shift will favour programmes that treat identity context, access scope, and policy enforcement as first-class telemetry.

Autonomy budget: the amount of unsupervised action a system is allowed to take before human review becomes the new control metric. As AI spreads across business workflows, practitioners will need to define where that budget sits, how it is measured, and when it is reduced. The organisations that do this well will be able to scale AI without turning control gaps into operational debt.

For practitioners

• Define AI agent authority boundaries Document what each AI system may access, which tools it can invoke, and which decisions still require human approval across production workflows.
• Instrument audit trails for AI actions Capture prompts, tool calls, policy decisions, overrides, and downstream outputs so that underwriting and incident teams can reconstruct behaviour.
• Map AI dependencies to NHI controls Inventory the service accounts, API keys, tokens, and credentials that support AI workflows and align them to least-privilege governance.
• Separate monitoring from trust decisions Use observability to detect drift, but require explicit policy enforcement before AI systems can act on high-impact workflows.

Key takeaways

• AI is changing cyber risk because it expands attack diversity while also introducing systems that can make and execute decisions across workflows.
• The article's most important governance signal is the move from static oversight to runtime control, where autonomy and observability both matter.
• For practitioners, the next step is to define bounded authority for AI systems, then prove those boundaries with auditable runtime evidence.

Related NHI Mgmt Group resources: AI Agent Authorisation Guide · Agentic AI Identity Guide · Zero Trust for AI Agents · AI Agent Observability, Audit and Incident Response Guide

Key terms

• Autonomous AI Agent: An autonomous AI agent is software that can perceive inputs, decide what to do, and act with limited or no human prompting. In identity security, it is treated as a non-human identity when it can authenticate, call tools, access data, or trigger workflows under its own runtime decisions.
• AI Risk Assessment: AI Risk Assessment is the structured process of identifying, analyzing, and prioritizing risks created by using artificial intelligence systems. It examines model behavior, data quality, security exposure, privacy impact, bias, misuse, and operational failure. In practice, it maps AI-specific threats to controls, owners, and residual risk decisions.
• Observability for AI Workflows: The ability to reconstruct what an AI system did, which tools it used, and what data or context it touched. For MCP deployments, observability is essential for incident response, audit, and proving that access stayed within approved boundaries.
• Closed-Loop Risk Management: Closed-loop risk management uses live signals, prior outcomes, and response feedback to continuously update risk decisions. In AI and cyber insurance contexts, it connects monitoring, underwriting, and incident handling so that controls improve as conditions change.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity, and workload identity for practitioners who need to govern machine actors as part of a wider security programme. It helps teams connect identity control to operational risk across modern environments.