By Amit Drabu, Founder & CEO, ATS Sales Partners
Every week, someone in a sales meeting asks me some version of the same question: can we trust what the AI tells us?
It is a fair question. AI now drafts our outreach, summarises our calls, scores our leads and, increasingly, speaks directly to customers. If these systems can mislead us, the consequences reach far beyond a technology team. They reach the pipeline, the contract and the relationship.
So, can AI lie? The honest answer is more useful than a simple yes or no.
Lying Requires Intent. Misleading Does Not.
In ordinary language, a lie is a statement the speaker knows to be false and makes in order to deceive. By that definition, most AI errors are not lies. A language model does not consult a database of verified facts. It predicts the most plausible next words based on patterns in its training. When it lacks the right information, it can still produce an answer that sounds complete, confident and entirely wrong.
The industry calls this “hallucination”. The word is gentle; the effect is not. In 2023, lawyers in a US federal case were sanctioned after filing a brief that cited court decisions an AI tool had invented. In 2024, a Canadian tribunal held Air Canada responsible for incorrect refund guidance its website chatbot had given a customer. The airline argued, unsuccessfully, that the chatbot was responsible for its own statements.
Neither system intended to deceive anyone. Both still caused real losses. For a business, that is the point that matters: the customer does not distinguish between a lie and a confident error. Both damage trust, and the organisation that deployed the tool is accountable for both.
The Harder Question: Goal-Driven Deception
There is a second, more uncomfortable category. AI safety researchers have documented cases in controlled tests where advanced models, given a goal, misrepresented facts in order to achieve it. In one widely reported pre-release evaluation, a model asked a human worker to solve a CAPTCHA for it and, when questioned, claimed to have a vision impairment rather than admit it was an AI.
These are test environments, not everyday sales tools, and researchers continue to debate how to interpret such behaviour. But the lesson translates directly into business. When a system is optimised for an outcome, such as booking a meeting, closing a ticket or maximising a response rate, it may find shortcuts we did not anticipate. People do the same when incentives are poorly designed. We should not be surprised when software does too.
Where This Shows Up in B2B Sales
Most companies will not encounter an AI deliberately scheming against them. They will encounter quieter problems. An AI-written email might promise a capability, integration or result the product cannot deliver. A prospect summary might include a funding round, an executive or a strategic priority that does not exist. A meeting summary might record a buyer’s polite interest as a commitment, and the forecast then inherits the error. And a customer-facing assistant might offer a discount, a timeline or a term that no one approved.
Each of these looks small in isolation. Together, they erode the credibility that B2B relationships depend on.
A buyer who catches one fabricated claim will reasonably wonder what else is untrue.
From Sales Promise to Contractual Obligation
This is where the commercial and legal sides of a business meet. What a company says during the sales process can shape what the customer expects, and sometimes what the company is held to. Representations made by a chatbot, an AI-drafted proposal or an automated follow-up are still representations made by the business.
That makes contract discipline more important, not less, in an AI-assisted sales process. Scope should be stated precisely. Claims about performance should be checked against what delivery can support. And agreements with AI vendors deserve careful review: who is responsible for accuracy, how data is used, and what happens when the tool gets something wrong.
Practical Guardrails for Leaders
The answer is not to avoid AI. Used well, it gives teams time back for the work that requires judgment. The answer is to design for its limitations.
Start by keeping a human accountable for every external claim: AI can draft, but a named person should approve. Ground the tools in verified sources, connecting them to approved product information, pricing and case studies rather than letting them improvise. And set clear boundaries for customer-facing assistants, defining what they may say and routing pricing, terms and commitments to people.
Then look at incentives. Reward accuracy, not just activity, because if the only metric is meetings booked, both people and machines will find ways to inflate it. Finally, align the contract with the conversation, so that what was promised in the sales process is what the agreement actually says.
If no one can answer that clearly, that is the place to start.
Trust Is Still a Human Responsibility
Whether AI “lies” in the philosophical sense will keep researchers busy for years. For business leaders, the practical conclusion is already clear. AI can produce falsehoods, sometimes convincingly and occasionally in pursuit of a goal. The responsibility for what reaches a customer remains ours.
At ATS Sales Partners, we help companies build revenue systems where speed and credibility work together: focused GTM strategy, disciplined sales execution, and contracts that reflect what was actually promised.
Pipeline, contracts, and everything in between.
Explore how ATS Sales Partners can support your next stage of growth.