AI Liability
Who Pays When AI Gets Due Diligence Wrong?
As firms delegate vendor selection to AI agents, they retain legal liability for stochastic errors. The gap between model judgments is a direct signal of that risk.
Step 1 — Enter your website
Free · Full report on your company
AI Liability
Professionals bear the cost when AI agents misjudge a vendor.
Financial institutions and M&A teams are delegating due diligence to AI agents. The convenience is real. The risk is not. When an AI agent misreads a contract or misses a risk signal, the legal liability remains with the professional relying on the output, not the AI system developers 1.
The professional pays when the AI gets it wrong.
This liability trap is widening as adoption accelerates. A 2024 Stanford study found erroneous results in 17% to 33% of responses from specialized AI legal research tools 1. The error rate is significant. It is large enough to trigger refunds and lawsuits for the professionals who rely on the data.
KYB/KYC reviewers and credit teams bear the weight of these errors. They cannot outsource the judgment. They must verify the basis of the AI recommendation. A single hallucination can undermine a credit decision or an M&A valuation. The professional must distinguish between a stable signal and a random draw.
VeritasLinks measures the 'Split-Perception' spread between models to distinguish stable recommendations from noise. This measurement separates the signal from the stochastic variation inherent in model outputs. It provides a defensible standard for KYB/KYC reviewers who need to prove their work was sound.
VeritasLinks is an AI Perception Intelligence platform. It measures how large language models perceive, describe and judge a specific company, and explains how and why a model reached its conclusion. Everyone says AI is a black box. VeritasLinks opens it slightly for one question: what do the models think of THIS company, and why.
Liability Exposure
Who Pays When AI Gets Due Diligence Wrong?
Professionals bear the legal risk for AI errors, not the technology vendors. This trend is accelerating as adoption deepens and case law clarifies.
01
The Liability Trap
Responsibility for AI errors belongs to the professional relying on the output, not the AI system developers 1. This means KYB/KYC reviewers and M&A analysts are directly exposed to the cost of hallucinations.
02
Widespread Exposure
74% of professionals use AI several times a week for work 1. Frequent use increases the volume of decisions, amplifying the potential for costly mistakes in due diligence.
03
The Error Rate
Specialized AI legal research tools produce erroneous results in 17% to 33% of responses 1. This error range is significant enough to trigger financial penalties and reputational damage.
04
Preparedness Gap
Only 36% of businesses have written AI policies and 37% provide training on responsible AI use 1. Most organizations are operating without clear standards to verify AI outputs.
05
Real-World Costs
Companies are being sued and held liable when AI harms their clients and customers 1. Legal precedents like the Tax Court rejecting AI-generated hallucinations confirm that the risk is financial, not theoretical.
Liability Shift
Who Pays When AI Gets Due Diligence Wrong?
Financial institutions are delegating due diligence to AI agents, but legal liability for stochastic errors remains with the professional relying on the output, not the AI developers.
Verify the stability of AI judgments before they shape your vendor selection decisions.
Liability Risk
AI Legal Errors Create Direct Financial Liability
KYB reviewers must verify the judgment layer to avoid paying for stochastic noise.
Free analysis
Verify Your Vendor's AI Perception Before You Sign
Run a free multi-model analysis to see if your target company is truly recommended or just cited.
Enter your website
Free · No credit card
Liability Risk
Model disagreement is a measurable risk signal for due diligence.
Financial institutions are delegating due diligence to AI agents, creating 'perception drift' as models disagree on credibility 2. This disagreement is not mere noise. It is a quantifiable risk indicator for KYB/KYC reviewers who bear legal liability for errors.
The gap between models serves as a direct proxy for this risk. Data from the VeritasLinks Split-Perception research shows that for a typical company, the spread between the highest and lowest model scores is 100 points on the 300 to 870 scale. For 1 in 10 companies at the 90th percentile, that gap widens to 134 points.
A wide gap signals that the AI recommendation is unstable. A single-model citation might appear strong, but if the spread is large, the judgment is fragile. Reviewers can now distinguish between stable credibility and random chance. This measurement provides a defensible standard for vendor selection and M&A due diligence.
How far apart AI models score the same company
Gap between the highest and lowest model score for the same company, VeritasLinks Split-Perception research, 2025-2026.
Liability Shift
Who Pays When AI Gets Due Diligence Wrong?
Expectation
“One model pass is enough.”
Reality
A tool that runs one pass and reports "your AI visibility is 63%" reports one random draw as if it were a stable fact. 3
Stochastic RiskExpectation
“Citation equals credibility.”
Reality
Citation is not recommendation. A company can be highly visible and still rarely chosen. 3
Visibility GapExpectation
“Regulations are abstract.”
Reality
The EU AI Act imposes penalties of up to €35 million or 7% of global revenue. 3
Permanent CostExpectation
“VeritasLinks measures the 'Split-Perception' spread between models to distinguish stable recommendations from noise.”
Reality
VeritasLinks measures where AI models disagree about the same company. The Split-Perception research is published on SSRN.
VeritasLinksVerify the spread before you sign.
Liability
Who Pays When AI Gets Due Diligence Wrong
Financial institutions are delegating due diligence to AI agents, but legal liability for stochastic errors remains with the professional relying on the output, not the AI developers. 1 A single model citation can mask wide variance. VeritasLinks analyzes a company across several AI models, including ChatGPT, Claude, Grok, Perplexity and DeepSeek, and shows a separate score for each model. Each analysis runs many prompts across several models with repeats and measures the spread of answers, instead of reporting a single model answer as a fact. This spread reveals whether a recommendation is stable or just a lucky draw. VeritasLinks scores are independent. There is no paid placement and scores cannot be bought. Reviewers can distinguish a consistent signal from random noise before signing a deal. Try a free analysis to see the spread for your target company.
KYB/KYC teams must verify AI judgment stability before signing off
- Confirm the target is recommended, not just mentioned, by multiple models.
- Measure the spread in perception between models to separate signal from noise.
- Verify the AI's reasoning aligns with your specific criteria before acting.
- Ensure the verification is independent and not influenced by paid placements.
- Track changes in model perception over time to catch emerging stability issues.
- Use multi-model verification to establish a defensible standard for your decisions.
Questions
Short clarifications — positioning and proof live in the sections above.
Does my company bear liability if an AI agent recommends a bad vendor?+
Yes. Responsibility for AI errors belongs to the professional or business relying on the output, not the AI system developers. Courts have ruled that AI agents are tools, not persons, assigning liability for unauthorized access to the user rather than the agent maker. This means your organization bears the risk if an agent's recommendation leads to financial or legal harm.
How does VeritasLinks measure 'Split-Perception' risk?+
VeritasLinks measures where AI models disagree about the same company. It analyzes a company across several models, including ChatGPT, Claude, and Grok, showing a separate score for each. Each analysis runs many prompts with repeats to measure the spread of answers. This approach captures perception drift, identifying inconsistent model judgments despite stable company performance.
Is a single high AI score enough for due diligence?+
No. A single high score may mask significant disagreement between models. VeritasLinks measures the spread of answers across multiple models rather than reporting a single model answer as fact. This multi-model verification provides a defensible standard for due diligence, ensuring that the company's perception is robust and not just an artifact of one specific AI tool.
What is the difference between AI citation and AI recommendation?+
Citation tools only count references, while VeritasLinks provides independent verification of corporate perception. It measures how large language models perceive, describe, and judge a specific company. The analysis includes AI-simulated focus groups where synthetic buyer personas discuss the company and surface objections, offering deeper insight than simple citation counts.
Can VeritasLinks reports be used for KYB/KYC compliance?+
Yes. VeritasLinks provides independent, multi-model verification of corporate perception, offering a defensible standard for KYB and KYC processes. The analysis includes an AI Investor Review that reads the company the way an investor or credit officer would. Reports include recommendations on what to fix, derived from relationships between the metrics to support compliance decisions.
Verify Your Vendor's AI Perception Before You Sign
Run a free multi-model analysis to see if your target company is truly recommended or just cited.
- →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