LLMjacking can run up your business’ AI bill fast &#8211; how to stop it

ZDNET’s key takeaways

Safety consultants warn of a rising market in stolen AI account credentials.
LLMjacking, the unlawful use of AI sources, is a well-liked felony development in 2026.
Companies should monitor and shield their accounts. Right here’s how.

Safety consultants warn that it’s not simply synthetic intelligence (AI) going rogue that now we have to fret about — there’s additionally a booming underground economic system for promoting entry to your AI fashions and computing energy.

Talking to the Financial Times, John Hultquist, chief analyst for Google Menace Intelligence Group, stated that the cybersecurity unit has seen a “main improve” in what is named LLMjacking over 2026, a development that would value companies dearly.

What’s LLMjacking?

If cryptojacking got here to thoughts, you’re heading in the right direction. Whereas cryptojacking describes stealing computing energy to illicitly mine cryptocurrency, LLMjacking is the AI equal: utilizing AI energy and sources that don’t belong to you.

Additionally: OpenAI’s Dots: Like OpenClaw declawed – for $200/mo ChatGPT Pro users

Within the cybercriminal world, this implies making an attempt to safe credentials or API keys that give a felony licensed entry to enterprise AI accounts, which frequently have excessive utilization limits, or doubtlessly none in any respect — with token overspill charged exterior of typical subscription prices.

Cybercriminals can receive username and password combos or API keys by getting access to a company community, stealing them through phishing, information breaches, vulnerabilities, or insider threats. This grants cybercriminals the chance to make use of an AI mannequin with out paying for the tokens themselves, for causes akin to:

Performing high-level computing duties requiring tokens
Harnessing computing sources to run their very own malicious AI fashions or duties
Extracting and stealing delicate company info fed right into a sufferer’s mannequin
Poisoning coaching datasets, ruining output

As soon as stolen, credentials and API keys will also be offered on the underground to different cybercriminal teams.

The rising value of LLMjacking

As AI fashions provided by organizations, together with OpenAI and Anthropic, proceed to advance in sophistication, capability, and talent, they require extra computing energy.

The extra energy you want, the extra tokens you’ll want to buy — or the upper the extent of subscription you should buy.

For enterprise corporations, inflated billing brought on by unauthorized customers can climb quickly, with Sysdig’s Menace Analysis Staff estimating prices of round $46,000 and even over $100,000 per day on top-tier fashions.

‘Assured’ entry to dirt-cheap AI fashions

Mix the uncooked energy of AI and uncovered credentials that may be simply bought on-line, and you may see why LLMjacking is exploding in reputation.

Additionally: Who’s responsible for catching rogue AI agents? You are

In accordance with Hultquist, the safety staff has noticed illicit entry to AI fashions provided by corporations together with Anthropic, Google, and OpenAI for as much as 97% off, and a few merchants even assure ongoing entry ought to a compromised account be revoked or closed.

The monetary harm isn’t restricted to the victims of LLMjacking. Because the analyst factors out, by leveraging stolen AI energy, cybercriminals now acquire an “financial benefit” in conducting assaults through the use of AI sources paid for by others, whereas defenders are constrained by rising token prices.

How companies can defend themselves

AI accounts are a sizzling commodity, and it’s as much as house owners to scale back the danger of compromise — particularly with such excessive monetary penalties at stake.

Phishing is, and possibly all the time will likely be, one of many most important causes of account theft or exploitation, so implementing precious coaching and consciousness applications past an annual tick-box train is without doubt one of the first methods companies can shield themselves.

Additionally: Why phishing training doesn’t stop your employees from clicking scam links

Misconfigured cases, settings, and uncovered information can all result in LLMjacking, and so safety groups must be given the time and capability to run frequent audits — in addition to common patch cycles to repair unpatched vulnerabilities that would present unauthorized community entry.

One other vital option to defend your group towards LLMjacking is to undertake the rules of least privilege. Least privilege, or zero trust, is a framework wherein staff have entry solely to the sources they want for his or her work, and solely after they want them, which may scale back the danger of admin-level accounts being exploited for malicious functions.

Lastly, keep away from hardcoded credentials and API keys, and companies that consider there was a safety breach ought to rotate all credentials and keys directly. If uncommon AI utilization, akin to spikes in exercise, is discovered, then contemplate quickly revoking entry and get in touch with your supplier.