Fraudsters have been caught gaming Canva’s AI credits system, exploiting a weakness in its account management to accumulate Canva AI tokens and sell them in bulk at deep discounts to customers that are increasingly questioning their uncontrolled spending on AI tools.

Using a ploy known as ‘seat cycling’, fraudsters exploited linkages between Canva and Leonardo.ai, a generative AI (genAI) platform that Canva bought in 2024 for $320 million.

As Canva combined the services, Canva Business trial users were given access to Leonardo.ai’s Essential plan, which provided 8,500 fast tokens per month and a 25,500 token rollover bank for access to genAI features from Leonardo.ai and partners like SeeDance, Veo Flux.2 Pro, and Kling.

Because each eligible team member received their own allotment of credits, fraudsters could continuously create new seats, sell access to those accounts, then delete the depleted accounts and create new ones with a new allocation of tokens – a process they now call the ‘Canva method’.

Canva pushed back by limiting Leonardo.ai access to paid Canva Business members, telling Capital Brief – which identified the practice and reported sellers listing compromised accounts as recently as 17 August – that it had “recently made some updates to address fraudulent activity.”

That activity, it said, “[involved] accounts being used to access AI services beyond normal limits…. These updates are part of our ongoing work to prevent abuse and protect our services.”

AI costs are taking their toll

The situation hints at a growing problem for AI vendors working to reconcile their hunger for user growth – and stickiness – with the high costs of building and offering genAI services at large scale.

High-end ‘frontier AI’ models, capable of complex problem solving and inference, are feted by tech giants that have a vested interest in maximising customers’ use of the expensive models – yet the emerging reality of AI ‘tokenomics’ has forced businesses to reconsider their AI spending.

Video generation, in particular, can burn through AI tokens extremely quickly, with a high-quality 8-second video clip costing around 2,500 tokens – meaning that the normal Essential plan allotment could be depleted by making just three videos.

Unfettered use of genAI platforms costing heavy users up to $10,450 (US$7,449) per employee per month – and with AI agents using up to 24 times as many AI tokens as standard queries, Goldman Sachs expects they will drive global token usage to an estimated 120 quadrillion tokens per month.

Canva founders [L to R]: Cliff Obrecht, Melanie Perkins, and Cameron Adams. Image: Supplied

Many businesses, however, want to rein in AI expenses – and most can halve AI agent token costs, and cut overall costs by up to 60 per cent, without affecting quality, Gartner advised in a recent research note that recommends “stable, repetitive tasks” be directed to “rightsized models”.

API gateways like LiteLLM and OpenRouter monitor AI platform costs and automatically direct AI queries to the cheapest or most appropriate model – but this model can be exploited by fraudsters, who register cheap AI models on the sites and sell tokens at massive discounts.

Yet many of these discounts have a sting in the tail, security experts have warned, routing user data through inexpensive, largely China-based AI models whose security risks were laid bare when DeepSeek’s emergence last year led to broad bans across Australia.

AI giants are adjusting their expectations

Like most companies offering ‘freemium’ pricing, Canva has built its business model around giving users enough flexibility to experience the benefits of its platform, while preserving incentives for them to upgrade for more powerful features or use in business teams.

AI companies’ ability to convert users to premium plans is key to their success, Gartner said, noting that “two AI companies with identical revenue and identical model providers can sit 35 gross margin points apart…. The difference is in how many tokens reach the most expensive models.”

Yet with a recent Stripe analysis estimating that 7.4 per cent of AI account signups are related to abuse of free accounts, AI firms’ financials are being compromised by the market for AI tokens, API credits, subscriptions and account permissions driven by widespread grey-market arbitrage.

As users become more careful about their AI token usage – and more willing to explore grey-market alternatives – AI investors are reassessing the returns on the hundreds of billions of dollars’ investment being poured into AI infrastructure – with dramatic consequences.

This month, for example, Canva cut its internal valuation by 20 per cent, reporting to investors that its business is now worth $43.9 billion – down by over $10 billion in the past year – even as major investor Blackbird Ventures slashed its valuation of Canva to $49.5 billion (US$34.9 billion).

It’s an unintended consequence for AI innovators like Canva, which is investing in AI to enhance its core design tools even as ever more powerful AI models obviate the need for those tools and other specialised software applications.

In June, a mass sell-off of tech and AI shares spooked global share markets amidst concerns that the billions being spent on AI infrastructure won’t generate enough revenues – particularly as SpaceX outlined a bond issuance to access more cash – just months after its record IPO.