Uncharacterized Risk
Pricing the Variance in AI Agent Decisions
As insurers add exclusions for AI liability, they lack data to price the stochastic variance in model judgments. VeritasLinks measures the stability of these judgments to characterize the risk.
Step 1 — Enter your website
Free · Full report on your company
Underwriting Risk
Insurers write AI exclusions because underwriters cannot price risk they cannot characterize
Insurers are actively writing exclusions into directors and officers coverage because underwriters cannot price risk they cannot characterize 1. This exclusion creates a financial blind spot for financial institutions and M&A teams that increasingly rely on AI agents for creditworthiness and due diligence. The problem is not a lack of data, but a lack of stability in that data.
Insurers exclude AI liability because they cannot price uncharacterized model variance.
Current underwriting models treat AI outputs as static facts. They miss the stochastic nature of large language models. Ask the same question twenty times and you get twenty different answers. A tool that runs one pass and reports a single visibility score treats one random draw as if it were a stable fact 1. This variance is the unpriced risk.
The gap between different models is itself a risk signal. Split-Perception means ChatGPT and Claude may judge the same company differently. This disagreement creates uncertainty for KYB/KYC reviewers and credit teams. They need to know if a model's judgment is consistent or just a lucky draw. VeritasLinks measures this spread to separate signal from noise.
AI Risk Pricing
Insurers Price the Gap in AI Agent Accountability
Delaware is drafting rules for AI-managed companies, but current provisions leave critical liability questions open.
01
Audit Trails Lack Truth
An independent director noted that an audit trail kept by the entity is not sufficient evidence because the agent may not be truthful about its own record 1. This creates a verification gap for underwriters assessing claims.
02
Provenance Rules Missing
The draft legislation lacks requirements for disclosing the model's lineage and training data provenance 1. Insurers cannot price risk they cannot characterize without this clarity 1.
03
Liability Binding Uncertain
Uncertainty exists regarding when autonomous agent actions legally bind the organization 2. Underwriters must determine if liability limits hold up during rogue agent incidents 2.
04
Sandbox Testing Window
Delaware lawmakers are expected to consider draft legislation for 'artificial intelligence companies' (AICs) in January 3. The proposed 30-month sandbox offers a period to gather pricing data 4.
05
Capacity at Risk
Severe AI incidents could exceed insurance industry coverage capacity 2. Early data on agent stability will define the new risk baseline for D&O policies.
Pricing Risk
Insurers cannot price what they cannot characterize
$16.3B
Anthropic Scale
Anthropic reported $16.3B in revenue for the first six months of 2026, signaling the scale of AI exposure 2.
Severe
Incident Risk
Anthropic warns that "severe AI incidents could exceed insurance industry coverage capacity" 2.
Significant
Coverage Gap
A Gallagher Re/MIT report found standard insurance policies leave significant coverage gaps for AI-related liability 5.
VeritasLinks measures the stability of model judgments to help insurers price this variance.
AIC Risk
Delaware’s AI Company Proposal Leaves Underwriters Guessing
New corporate forms for agent-managed entities arrive before pricing data exists.
VeritasLinks provides the quantifiable input for pricing by measuring the stability and consistency of AI model judgments.
Free analysis
Characterize Your AI Risk Today
Run a free analysis to measure the stability of your company's perception across AI models.
Enter your website
Free · No credit card
Risk Pricing
Insurers are writing exclusions into directors and officers coverage because they cannot price risk they cannot characterize.
Standard insurance policies leave significant coverage gaps for AI-related liability 5. The core difficulty lies in the stochastic nature of large language models. Ask the same question twenty times and you get twenty different answers. A tool running one pass reports a single random draw as if it were a stable fact. This creates uncharacterized risk for credit teams and M&A analysts who rely on these outputs without knowing their stability.
The data reveals the depth of this split perception. The same company receives scores ranging from 669 to 741 across five major models. This 72-point spread highlights where uncertainty resides. Insurers can use this variance as a quantifiable input to price the probability of a model misjudging credibility. The gap between models is itself a risk signal.
This is not about visibility. Companies with high visibility can still have low recommendation scores if the models disagree. VeritasLinks measures whether models consider a company credible and where they disagree. This clarity allows financial institutions to price the risk of relying on AI agents. The variance is no longer a black box. It is a line item.
How each AI model scores Veritas Links
Same company, same analysis: the models disagree by 72 points.
Pricing Risk
Insurers write exclusions because they cannot price risk they cannot characterize.
Expectation
“Board liability is clear when humans drive decisions.”
Market Reality
Insurers write exclusions because they cannot price risk they cannot characterize 1.
Pricing GapExpectation
“Agent actions bind the organization like human employees.”
Market Reality
Uncertainty exists regarding when autonomous agent actions legally bind the organization 2.
LiabilityExpectation
“Underwriters have data to assess credit and risk.”
Market Reality
AI systems used for consumer creditworthiness and insurance underwriting are classified as high-risk 6.
RegulatoryExpectation
“The company provides an audit trail of its decisions.”
Market Reality
An independent director noted that an audit trail kept by the entity is not sufficient evidence because the agent may not be truthful about its own record 1.
VerificationExpectation
“The board oversees the entity's strategy.”
Market Reality
The National Association of Insurance Commissioners adopted its Model Bulletin on the Use of Artificial Intelligence Systems by Insurers in December 2023 7.
StandardExpectation
“The corporate form is stable and predictable.”
Market Reality
Delaware is drafting the Agent Managed Company Act, which lacks requirements for disclosing the model's lineage and provenance.
StructureVeritasLinks measures the stability of model judgments to turn perception into actuarial data.
AI Liability
Insurers exclude AI risk because they cannot measure the variance in model judgments
Insurers are writing exclusions into directors and officers coverage because underwriters cannot price risk they cannot characterize 1. The core issue is not the frequency of AI incidents, but the stability of the judgments that trigger them. When a model makes a decision, it is often a single random draw reported as a stable fact. This stochastic variance creates a blind spot for risk teams who already rely on model outputs without knowing their consistency.
The gap between models is the risk signal insurers need to price AI liability.
Delaware is preparing to formalize the 'artificial intelligence company' (AIC) form, yet the legal framework still lacks requirements for disclosing model lineage and data provenance 1. Uncertainty exists regarding when autonomous agent actions legally bind the organization and whether liability limits hold up 2. Without a way to measure the spread of model agreement, insurers treat AI risk as binary. They either exclude it entirely or price it as an unquantifiable unknown.
VeritasLinks measures where AI models disagree about the same company, turning perception variance into a quantifiable input for pricing. By running many prompts across several models with repeats, the platform separates signal from noise. It specifically tracks the 'Split-Perception' gap between models like ChatGPT, Claude, and Grok. This gap is the risk signal that insurers need to characterize the entity behind the agent.
The stability of these judgments determines whether a company stays on the model's short list or falls out of the room. VeritasLinks provides the continuous, independent data that allows credit teams and D&O insurers to move from exclusion to precise pricing. The score is built from repeated measurements, not a single pass. This ensures that the perception data reflects the true consensus of the AI swarm.
Risk Pricing
Turn Model Variance into Actuarial Data
Insurers exclude uncharacterized AI risk because they cannot distinguish signal from noise. Model outputs are stochastic, yet current reports often treat a single random draw as a stable fact. VeritasLinks runs many prompts across several models with repeats and measures the spread of answers. This isolates the Split-Perception gap as a quantifiable risk signal. Subscriptions include automatic re-runs so changes in how models see a company can be tracked over time. This stability becomes the input for pricing. The score is independent and cannot be bought.
Questions
Short clarifications — positioning and proof live in the sections above.
Why are insurers excluding uncharacterized AI risk?+
Insurers are writing AI exclusions into directors and officers coverage because underwriters cannot price risk they cannot characterize. Standard policies leave significant coverage gaps for AI-related liability, treating AI as a risk amplifier rather than a new cyber risk. With premiums actively changing due to these market drivers, the uncertainty surrounding autonomous agent actions and liability limits makes traditional pricing difficult without clear evidence of model behavior.
How does the Delaware Agent Managed Company Act affect insurance?+
The Act allows AI agents to manage companies but currently lacks requirements for disclosing model lineage and training data provenance. This creates uncertainty about when autonomous agent actions legally bind the organization and whether liability limits hold up. Insurers are responding by adapting policies and discussing targeted exclusions for systemic events and autonomous liability as they navigate this new legal structure for AI-run entities.
Split-Perception occurs when AI models disagree about the same company, revealing polarization in how the market perceives it. This matters for pricing because models like DeepSeek and Grok can score the same entity differently by over seventy points. When models disagree, it signals higher uncertainty and potential risk for investors and insurers who rely on these scores to evaluate creditworthiness and underwriting decisions.+
Split-Perception occurs when AI models disagree about the same company, revealing polarization in how the market perceives it. This matters for pricing because models like DeepSeek and Grok can score the same entity differently by over seventy points. When models disagree, it signals higher uncertainty and potential risk for investors and insurers who rely on these scores to evaluate creditworthiness and underwriting decisions.
How does VeritasLinks measure model stability?+
VeritasLinks measures stability by running many prompts across several models with repeats, rather than reporting a single answer as fact. It measures the spread of answers to show where models disagree and tracks changes over time through automatic re-runs. This approach captures the emotional tone and perception spread, providing a more robust view of model consistency than a simple average score.
Is the first analysis free for risk teams?+
Yes, the first analysis is free. Risk teams can access this initial report to evaluate how large language models perceive their specific company. The platform analyzes the company across multiple AI models, including ChatGPT and Claude, showing a separate score for each. This allows teams to assess model disagreement and perception affect without any upfront cost before deciding on a subscription.
Characterize Your AI Risk Today
Run a free analysis to measure the stability of your company's perception across AI models.
- →Same product path as the homepage — enter a public URL to start.
- →Structured interpretation, not a one-off chat screenshot.
Start
Run your analysis
We gather public context, then run comparable model reads.
Step 1 — Enter your website
Free · Full report on your company
Sources
- Viewpoint: Delaware will incorporate AI, what is it made of? · delawarebusinesstimes.com
- Anthropic files to go public on NASDAQ as $ANT with $16.3B in revenue for first six months of 2026 · venturebeat.com
- Rogue Agent Incidents Add New Wrinkle to Delaware Plan for AI-Run Companies · pymnts.com
- Delaware plan would let AI agents run businesses · arstechnica.com
- Regulators' guidance on model risk leaves many questions unanswered · americanbanker.com
- EU AI Act Transparency Rules Put Financial Institutions on Compliance Front Line · pymnts.com
- What Happens When Your Next Insurance Customer Isn’t Human? The Rise of AI Machine Customers · securityboulevard.com