Beware the token trap: cutting costs could boost security risks

The prolific token consumption of agentic AI can leave organisations facing substantial and unexpected costs. Yet attempts to reduce those expenses without properly assessing the security implications could prove even more costly.

Token charges can appear negligible when considered individually. However, those costs can rise rapidly as AI agents take on more tasks, conduct longer reasoning processes and interact repeatedly with tools, applications and other agents. The result can be an AI bill far larger than was anticipated.

This is creating a difficult balancing act for CISOs. Efforts to contain inference costs, driven largely by the volume of tokens consumed and generated by AI agents, can encourage organisations to use smaller models or reduce the depth of automated security checks.

The problem is compounded by fragmented pricing across the AI market. Organisations must navigate an expanding range of options, from highly optimised nano models designed for narrow, specialised tasks to premium reasoning engines capable of handling complex, multi-step workflows.

The hidden cost of AI agents

Erratic pricing has been a trademark of generative AI, almost from the beginning. About two years after OpenAI released ChatGPT, the Chinese company DeepSeek shook up the AI market with the release of a powerful, open-weighted large language model whose training parameters were publicly available, allowing users to customise the model and build on the cheap, compared with other generative AI models.

ChatGPT-maker OpenAI and other AI companies started doing the same, and suddenly, the costs of using GenAI systems dropped off a cliff. In fact, prices fell faster for GenAI than for any other technology in history.

The emergence of agentic AI, however, has introduced some stealth costs into the equation. The costs of agentic software range from free for open-source models, to enterprise agents with prices that vary from one-time fees (roughly $15,000 for basic models to more than $1 million for global enterprise models) to monthly subscriptions (which can range from a few thousand to $13,000 or more).

The hidden cost of AI tokens

Tokens are consumed whenever an AI agent asks a generative AI model to process information. On the surface, the expense can appear negligible. Input tokens typically cost between US15c and US$5 per million, while output tokens, which require more computing power, range from about US60c to US$25 per million.

However, those modest charges can escalate rapidly. AI agents operate at speed, with considerable autonomy and often unpredictable behaviour. They are designed to interact with systems, applications and other agents across an enterprise, meaning a single task can trigger scores of calls to a large language model.

As agentic AI adoption accelerates, so too does token consumption. Recent estimates suggest the technology has already pushed some IT budgets about 20% higher, turning inference costs into a material financial consideration for organisations.

This pressure is encouraging CISOs to seek savings wherever possible, including by selecting LLMs with the lowest per-token prices.

However, while a cheaper LLM may deliver immediate budget relief, any savings can quickly evaporate if the model introduces vulnerabilities or lacks adequate governance and oversight. For CISOs, the challenge is therefore not simply to minimise token costs, but to balance those costs against security and operational risk.

Yet better security doesn’t necessarily have to cost more. Depending on what they’re using agentic AI for, organisations may find they don’t always have to trade security for lower token costs.

Getting costs (and risks) under control

There are a number of things organisations can do to help stop token costs from getting out of hand, including:

• Matching agents and LLMs to the job at hand: Commodity AI systems can cost little or nothing, but they lack the deep reasoning for complex security synthesis. Yet not every application or function within an organisation requires a reasoning engine. Agents can work with low-cost LLMs on low-risk projects, while preserving higher-cost LLMs for critical tasks.
• Factoring in risk scores: The security implications of using AI can’t be ignored. When developing a budget plan, include risk factors.
• Monitoring workflows: Maintain a close watch on workflows to help track costs and performance, allowing a better understanding of which tools work best in which situations.
• Maintaining human oversight: Despite agentic AI’s autonomy — in fact, because of agentic AI’s autonomy — forgetting about the importance of the human element is risky business. Teams need thorough upskilling in secure development, with clearly defined ownership roles.
   

Agentic AI is becoming central to enterprise operations, making cost control essential. However, choosing tools on token price alone risks creating technical debt and weakening security. CISOs must assess cost, performance and risk together within a modern security framework and AI governance policy. This will allow organisations to scale agentic AI without blowing budgets or exposing the business to unacceptable risk.

Top image credit: iStock.com/Just_Super