Token Scrounging: The hidden cost of public AI agents

As more organisations add AI agents to their websites, a new behaviour is emerging: token scrounging. (Personally, I prefer to call it theft).

Instead of using an AI assistant for its intended purpose, some users deliberately push it to complete an unrelated, token-intensive task such as writing code, creating lengthy reports or generating marketing content. The result is that the organisation, not the user, pays for the AI usage.

While it may seem harmless, this behaviour increases operating costs, consumes computing resources and can reduce performance for genuine customers. It is quickly becoming one of the challenges of making powerful AI available to the public.

The good news is that it is largely preventable. Public-facing AI agents should be designed with clear boundaries. Restrict them to the tasks they were built for, limit response length, refuse requests outside their purpose, apply sensible rate limits and consider authentication where appropriate. For example, an interview scheduling agent should remain an interview scheduling agent, not become a free software developer or content writer.

As AI becomes a standard part of digital client and candidate experiences, firms will need to think beyond what their agents can do, and focus on what they should do.

Great AI implementations will balance helpfulness with governance, ensuring great customer experiences without opening the door to unnecessary costs or security risks.

‍

Don’t forget, Mercury’s Consultancy team can help with AI agent deployments and Microsoft AI token credits. Contact us for more information.

‍