Verification Gap
The End of the 'AI Did It' Defense
With the FTC rejecting the autonomous actor defense, enterprises face direct liability for AI agent errors. The question is no longer who built the AI, but whose risk it creates.
Step 1 — Enter your website
Free · Full report on your company
Liability Shift
The 'AI Did It' Defense Is Closing
The legal shield protecting vendors from AI agent errors is fracturing. The FTC has rejected the 'autonomous actor' defense, preventing vendors from relying on the 'AI did it' defense 1. This ruling means companies can no longer blame AI agents for errors when those agents act on their behalf.
Unstable perception is no longer just a technical glitch; it is a legal exposure.
Liability is shifting from developers to users. Congress has left liability gaps for agent-driven harm, compounding the risk for enterprises 2. A company using an AI vendor for due diligence now faces uncertainty over who holds the bag when the agent misinterprets data or violates terms.
This legal shift raises the stakes for vendor perception. If an AI agent's output drives a purchase or partnership, the stability of that output matters. Unstable perception is no longer just a technical glitch; it is a legal exposure.
Liability Shift
The End of the 'AI Did It' Defense
The Ninth Circuit ruled that users, not developers, bear the risk when AI agents access third-party websites 2.
01
Users own the access risk
When a user directs an AI agent to a website, the user is the one who 'accesses' the computer under the CFAA 2. This shifts liability from the technology provider to the business relying on the agent.
02
Agents are tools, not persons
The court held that AI agents are 'tools, not persons' for statutory purposes 3. Technical architecture determines whether the user or developer is liable for unauthorized access.
03
Stochastic risk in due diligence
Financial institutions are scaling AI-driven due diligence 4. They need independent verification to distinguish stable vendor perception from unstable AI outputs.
04
Liability gaps remain open
Congress has left liability gaps for agent-driven harm 2. Vendors can no longer rely solely on the 'autonomous actor' defense to avoid responsibility for errors.
Due Diligence Risk
Frontier models hit a 51% ceiling in complex workflows
51%
Due Diligence Score
Frontier AI models scored only 51% on a realistic private-equity due diligence simulation, revealing a capability ceiling in handling long, messy workflows 5.
51%
Capability Ceiling
The score exposes the gap between AI capability and the precision required for vendor liability under the new autonomous actor standard.
Verify vendor perception stability before the legal risk becomes your cost.
Legal Risk
The End of the 'AI Did It' Defense
The Ninth Circuit ruled that when an autonomous AI agent accesses a password-protected website, the user, not the developer, is responsible for the access 3.
VeritasLinks measures the stability of vendor perception across multiple models to separate consistent signals from stochastic noise.
Free analysis
Verify Vendor Stability Before the Ruling Hits Your Balance Sheet
Run a free analysis to see if your vendor's perception is stable or stochastic.
Enter your website
Free · No credit card
Liability Shift
The 'AI Did It' Defense Is Over
The FTC has rejected the 'autonomous actor' defense, preventing vendors from blaming AI agents for errors when those agents act on their behalf 1. This ruling shifts liability from the AI developer to the user. For enterprise risk and M&A teams, this creates a critical verification gap. If a vendor's AI agent makes a mistake during due diligence, the vendor is now on the hook. The risk is no longer abstract. It is financial and legal.
The problem is not just who is liable, but how stable the AI judgment is. LLMs are stochastic. Ask the same question 20 times, and you get 20 different answers. A tool that runs one pass and reports 'your AI visibility is 63%' reports one random draw as if it were a stable fact. VeritasLinks runs many prompts across several models with repeats, measures the spread and separates signal from noise. This separation is vital when determining vendor responsibility.
The perception spread between models is itself a risk signal. Our data shows that for a typical company, the gap between the highest and lowest model scores is 100 points on the 300 to 870 scale. For 1 in 10 companies, that gap widens to 134 points. This split perception means different models may judge the same vendor differently. When the 'AI did it' defense fails, the stability of that judgment matters. VeritasLinks measures this stability. It ensures that due diligence findings are based on consistent signals rather than stochastic noise.
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
The 'AI Did It' Defense Is No Longer Valid
What You Expect
“Vendors claim AI agents act as independent entities.”
What Is Happening
Courts reject the autonomous actor defense, holding users liable for agent actions 1.
LegalWhat You Expect
“AI data handling is a minor operational detail.”
What Is Happening
Reused customer content triggered a six-figure settlement due to confidentiality breaches 1.
FinancialWhat You Expect
“AI disclosures reflect marketing narratives about capability.”
What Is Happening
Regulators require disclosures to accurately reflect how systems are developed and deployed 1.
ComplianceWhat You Expect
“Agents are tools that act on behalf of the user.”
What Is Happening
Technical architecture determines whether the user or developer bears the access risk 3.
StructureWhen the law stops blaming the machine, you must verify the signal.
Liability Risk
The Ninth Circuit ruling shifts AI agent liability from developers to users, exposing companies to risks from unstable vendor perceptions.
The legal landscape has shifted. The Ninth Circuit ruled that when a user directs an AI agent to access a website, it is the user who 'accesses' the computer, not the AI developer 2. This means companies can no longer blame the AI when an agent causes harm. The responsibility for those outcomes now rests with the company directing the agent.
A single model response is often treated as a stable fact. It is actually just one random draw from a distribution of possible answers.
Risk teams are already navigating this uncertainty. Enterprise risk and compliance officers make critical decisions using stochastic model outputs without knowing their stability or bias 6. A single model response is often treated as a stable fact. It is actually just one random draw from a distribution of possible answers.
VeritasLinks measures the spread of answers across multiple models to separate signal from noise. It runs many prompts with repeats to determine if a vendor's perception is consistent or merely a lucky guess. The platform identifies where AI models disagree about the same company. This split-perception gap serves as a direct risk signal for compliance teams.
Stable perception is essential for liability. If a vendor's AI agent acts on flawed or inconsistent data, the company bears the cost. VeritasLinks ensures that due diligence findings are based on consistent signals rather than stochastic noise. The first analysis is free for teams to test the stability of their current vendor landscape.
Check vendor perception stability before the 'AI Did It' defense expires
- Verify that your key vendors have consistent perception scores across multiple models, not just a single random pass.
- Ensure due diligence findings rely on stable signals rather than stochastic noise from individual AI agents.
- Confirm vendor liability covers agent access errors, such as 'accessed password-protected accounts without identifying itself as an AI' 2.
- Use continuous re-runs to monitor the spread in model judgments for your specific company over time.
- Review the AI Investor Review to see how a credit officer would read your vendor's stability and bias.
- Act now while the legal landscape is shifting to protect against unstable model outputs in future contracts.
Questions
Short clarifications — positioning and proof live in the sections above.
What is the 'autonomous actor' defense?+
It is a legal argument allowing vendors to blame AI systems for errors, claiming the technology acted independently. The FTC has rejected this defense, preventing companies from using the 'AI did it' excuse. This shift places liability squarely on the vendor rather than the autonomous nature of the tool itself.
How does the FTC ruling affect third-party AI vendors?+
The ruling prevents vendors from relying on the autonomous actor defense when their AI causes harm. This increases their legal exposure and forces them to take greater responsibility for AI-driven outcomes. Vendors must now ensure their systems are reliable, as they can no longer shift blame to the technology's independence.
How does VeritasLinks measure vendor stability?+
VeritasLinks measures how large language models perceive a specific company across several AI systems. It analyzes the spread of answers from models like ChatGPT and Claude, rather than reporting a single answer. This multi-model approach captures disagreement and stability, providing a clear view of how consistently the market perceives the vendor.
Verify Vendor Stability Before the Ruling Hits Your Balance Sheet
Run a free analysis to see if your vendor's perception is stable or stochastic.
- →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
- FTC, SEC Scrutiny Raises Stakes for AI Data Disclosures · pymnts.com
- Ninth Circuit Rules on AI Agent ‘Access’ to Third-Party Websites Under CFAA · jdsupra.com
- Ninth Circuit Rules AI Agents Act on Behalf of Users, Not Developers · pymnts.com
- iDenfy upgrades its Know Your Business (KYB) verification with a Spanish-language localization feature · stevenspointjournal.com
- Finance AI Tops Out at 51% on Due Diligence: Startup Raises $30M to Fix It · techtimes.com
- Why We Should Fear The Shadow Web We’re Building For AI via @sejournal, @chrisgreenseo · searchenginejournal.com