EXPLORINEER
en

UK — The Register

Token thieves costing companies $100K+ per day in illicit LLM charges

Security1 min read

Criminals are stealing access keys to companies' AI tools and running up huge bills. Google says some victims lost more than $100,000 a day, and one lost $1 million in total.

What happened

Hackers are taking over companies' access to large language models (LLMs). An LLM is the kind of AI that powers chatbots. Hackers use the company's account to run AI tools and computing power for themselves. The company then gets the bill. Google's Mandiant team, which helps companies after cyberattacks, reported this on Monday. The tactic is called LLMJacking.

Key numbers and facts

  • Some fake charges went above $100,000 a day.
  • At some companies, total charges reached $1 million before the hackers were blocked.
  • In one August case from CrowdStrike, a thief sent about 200,000 API requests in two minutes.
  • One intruder made more than 500,000 LLM API calls in under eight hours, using stolen third-party keys for a service called Jet Night.
  • Jet Night offered an unlimited plan for about $60 a month. Similar plans usually cost $100 to $250 a month.

How the thieves work

An API key is a secret code that lets a program use a service. Thieves steal these keys or find weak settings in cloud accounts. One group, UNC6993, made new keys and turned on services inside the victim's cloud project. Another thief used short-lived accounts to avoid usage limits. Some criminals now split the work. One group gets in. Another group, called LLM Mixers, sells or uses the access for profit.

Why it matters

This is a money problem for companies, and it can also help attackers. Google's expert John Hultquist warned that stolen AI power lets criminals run many cheap, automatic attacks. Companies need to lock down their AI keys and watch their bills closely so they can spot theft early.

What to watch

  • Whether more companies find stolen API keys in their cloud accounts.
  • How fast LLMJacking grows, since Google says it has risen steadily over the past two years.
  • Whether services like Jet Night keep having outages as stolen keys are cut off.

Glossary

LLM
Large language model: a type of AI that reads and writes text, like a chatbot.
LLMJacking
When criminals take over someone else's AI account to use its paid AI services.
API key
A secret code that lets one program connect to and use another service.
Mandiant
A Google-owned security firm that investigates cyberattacks and helps companies recover.
Jet Night
A token routing service that the article says used stolen keys to offer cheap AI access.

Ad

Add 150 words. Learn 10 a day.

Dicter word books send only the words you learn to your review notes.

SecurityThe Register

Thieves pilfer environment variables from Trigger AI workflow tool

Trigger.dev, a cloud-hosted platform for building AI agents and workflows, disclosed a security incident on Sunday in which an attacker gained unauthorized access to environment variables belonging to a subset of its customer organizations. The company said the exploit has been blocked and affected customers were notified of steps to take, but it has not named the affected organizations and declined to provide more details beyond a forthcoming full incident report.

SecurityITmedia

GMO press conference repeats "hourglass of life"; Internet users pile on criticism of CEO Kumagai's remarks

GMO Internet Group held a press conference on November 2 over unauthorized access to its "Onamae.com Desktop Cloud" VPS used for FX automated trading. CEO Masatoshi Kumagai repeatedly used the phrase "hourglass of life," which trended on X, while many users criticized the rambling remarks, including his discussion of AI, as hard to follow. Kumagai apologized for his recent X posts, said he would suspend his X account for a period after summarizing the conference, and did not clearly deny resigning.

SecurityThe Register

Nuclear threats and street brawls: Russia, Ukraine report attacks on their emergency alert apps

Cyberattacks on Russia's and Ukraine's emergency alert apps on Sunday evening sent fake messages about missile and nuclear strikes to the public, with officials from both countries confirming the incidents. Ukraine's Air Force quickly said there was no radiation threat, while Russia's Ministry of Emergency Situations said its app was restored about 90 minutes after an update. Russia has not explicitly blamed Ukraine for the attack, and Ukraine's Oleksander Merezhko described it as part of cyber war.

SecurityITmedia

GMO Chairman Kumagai Delivers "Kumagai-bushi" Even at Tearful Apology Briefing: "In 31 Years, I Have Never Used the Word 'Profit'"

At a briefing on the unauthorized access to GMO Internet Group's "Onamae.com Desktop Cloud" FX auto-trading VPS, Chairman Masatoshi Kumagai said the company will not pay ransom to criminals and described AI-driven changes in cyber threats. He said the industry's pace of change has become "seconds-advancing-minutes," and that the company must shift from a security firm to a resilience firm. Kumagai, 63, said he wants to make the industry a tool for smiles before he dies.