# Silent AI is the new cyber as exclusions rise and standalone cover emerges

> AI is starting to look a lot like cyber: silent exposure, quick-spreading exclusions, and a push for standalone cover. Expect sharper underwriting and clear, testable triggers.

- Type: AI news
- Page: https://completeaitraining.com/news/silent-ai-is-the-new-cyber-as-exclusions-rise-and/
- Published: 2026-02-24
- Updated: 2026-02-24
- Author: Jeroen Erne
- Jobs: Insurance
- Reading time: 4 min

### AI is the next cyber: why insurers may carve out standalone cover

AI is spreading across insureds faster than policy language can keep up. That gap creates "silent AI" exposure - risk that may be sitting across multiple lines without being priced, modelled, or explicitly asked about.

The result is the same tension we saw with cyber. Clients expect cover; contracts hide ambiguity. Claims will test definitions, causation and exclusions.

#### Why this is surfacing now

Accumulation is the headache. If thousands of insureds lean on the same models, libraries or vendors, a single defect or legal theory can scale into correlated loss.

That's why the market is starting to cap ambiguity, sharpen underwriting questions, and - where needed - move exposure into products built for AI risk.

#### From ambiguity to exclusions

Early AI exclusions are appearing in some markets. Once an exclusion gets templated, it can spread quickly through guidelines, broker playbooks and renewal packs - often before insureds catch up.

Industry reporting also points to new general liability exclusionary forms aimed at generative AI with 2026 edition dates. If these standardise, expect a fast cascade across lines.

#### The coverage gap many will miss

AI incidents rarely sit in one tower. A single event can look like misrepresentation (GL/advertising injury), a professional error (PI/E&O), a privacy issue (cyber), or a governance failure (D&O).

If exclusions land across more than one of those at once, you get a multi-policy no man's land - exactly when a dispute or class action heats up.

#### Echoes of cyber - and the path to standalone

Cyber started as a silent pickup in traditional forms, then was carved out and rebuilt as a dedicated product with its own pricing, controls and limits. AI is tracking the same pattern.

"We're already looking at a standalone AI insurance product," said Eric Lowenstein, CEO of Tego. Expect DIC/DIL-style solutions to drop down where traditional policies retreat.

#### First waves of "affirmative AI" cover

Some Lloyd's markets are backing targeted AI solutions tied to performance degradation and model failure, not just a one-off mistake. Other offerings are exploring cover for "mechanical underperformance" - think hallucinations, model drift and tool malfunctions - alongside third-party liability.

This is the market doing what worked for cyber: define the peril, set clear triggers, and demand controls. That's the bridge from "we won't cover this" to "we will - under conditions we understand."

### What to do now

#### For brokers

- Map client AI use. Build an AI inventory: use cases, autonomy level, user counts, criticality, vendors, and shared dependencies.
- Pre-empt underwriting. Add an AI supplemental to submissions with controls, providers, monitoring and rollback detail.
- Compare exclusions across GL, PI/E&O, cyber and D&O. Look for concurrent gaps and consider DIC/DIL or endorsements to coordinate triggers.
- Tighten contracts with AI vendors. Push for indemnities, SLA remedies tied to performance degradation, change notifications, audit rights and log retention.
- Pre-wire claim narratives. Where would a hallucination, biased output, defamation, IP misuse, or a provider outage actually land? Align towers before the loss.
- Educate buyers on grey areas. Misadvice from a chatbot may be treated differently than an operational outage - set expectations early.

#### For underwriters

- Exposure mapping: What decisions does AI touch? Customer-facing vs internal, human-in-the-loop, and the business processes at risk.
- Dependency risk: Which models, APIs, clouds and open-source components are common across your book? Watch vendor concentration.
- Controls and governance: Model testing, red-teaming, drift monitoring, evals, rollback/kill-switches, change management and incident response.
- Data posture: PII/PHI usage, copyright exposure, consent, RAG source governance, and prompt/content filtering.
- Auditability: Versioning, prompts, guardrails, logs and the ability to reconstruct an event for forensic and claims purposes.
- Legal posture: Disclaimers, product representations, bias/discrimination safeguards, jurisdictional exposure and IP risk management.
- Loss models: Frequency/severity for misstatement, discrimination, IP claims, privacy events and business interruption from provider outages.

#### For risk managers and insureds

- Build an AI register: owner, purpose, model/provider, data flows, integrations, and criticality. Keep it current.
- Adopt an AI control framework (e.g., the [NIST AI RMF](https://www.nist.gov/itl/ai-risk-management-framework?ref=completeaitraining.com)) and prove it with documentation.
- Strengthen vendor management: performance SLAs tied to model quality, change controls, notification windows, indemnities and meaningful caps/carve-outs.
- Keep evidence: prompts, outputs, model versions and decision logs. Claims depend on reconstructing what happened.
- Run tabletop scenarios: defamation from a chatbot, biased screening, pricing drift, bad RAG sources, provider outage. Check which policy would respond.
- Stress-test wordings at renewal. Ask directly about AI exclusions, performance-degradation triggers and any systemic risk terms.

#### What a standalone AI policy could include

- Third-party liability: algorithmic errors, misleading outputs/hallucinations, bias and discrimination, defamation, IP infringement, product misstatement.
- First-party: rectification costs, forensic costs, business interruption from AI/tool outage or degraded performance, extra expense, data restoration.
- Regulatory and investigations: defense costs, fines/penalties where insurable, consumer protection actions.
- Clear triggers: performance degradation thresholds, drift events, rollback activations - not just discrete "errors."
- Conditions: minimum governance, model monitoring, human oversight, vendor SLAs, logging and change controls.
- Aggregation controls: sublimits for vendor/systemic events and accumulators for widely used models or providers.

#### 12-24 month outlook

Expect more exclusions, supplemental questionnaires and pilots of standalone cover. Brokers who map AI exposure across towers and pre-wire claims will win renewals.

Insurers will write AI under clearer triggers and documented controls. The market will move from silent exposure to affirmative risk - with pricing, limits and governance to match.

**Want to go deeper on practical AI use across claims and underwriting?** Explore [AI for Insurance](https://completeaitraining.com/tag/insurance/).

---

This is the Markdown form of https://completeaitraining.com/news/silent-ai-is-the-new-cyber-as-exclusions-rise-and/. The whole site is mapped at https://completeaitraining.com/llms.txt.
