TL;DR
Get smart everyday buys delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
A Financial Times report says underground markets are selling stolen access to AI models and computing resources, sometimes at discounts of up to 97%, citing Google Threat Intelligence researchers. The report describes a growing market that could lower attackers’ costs as companies also use AI to defend against fraud and cyberattacks.
Illegal access to AI models is becoming a commodity in cybercriminal markets, according to a Financial Times report published Sept. 27 that cited Google Threat Intelligence researchers. The activity includes selling stolen login credentials and taking computing resources to run models without paying, potentially lowering attackers’ costs for extortion, warfare and espionage.
John Hultquist, chief analyst at Google Threat Intelligence Group, told the FT that the company had seen a major increase in so-called LLM-jacking this year. The term covers efforts to obtain unauthorized access to large language models, including through pilfered credentials for public AI tools or stolen computing capacity used to run models. The report did not provide a count of incidents or define the period used to measure the increase.
Dark web marketplaces are offering access to models from Anthropic, OpenAI and Google at discounts of up to 97%, the FT reported, citing information from Google Threat researchers. The report compared the market with subscriptions for advanced versions of ChatGPT and Claude, which can cost as much as $200 per user each month. It did not specify how many listings were observed or how the discount figure was calculated.
Hultquist said access to AI at a lower cost can give attackers a financial or efficiency advantage over organizations that must defend themselves using similar tools. The report also cited PYMNTS Intelligence research showing that 42% of surveyed companies use three or more AI defense tools, while half identified AI-generated vendor impersonation emails as their leading threat. Those figures concern the separate PYMNTS survey, not the scale of AI access theft.
Lower AI Costs for Attackers
If stolen credentials or computing capacity let criminals use expensive AI tools at a steep discount, they may be able to reduce the cost of carrying out or scaling some cyber operations. Hultquist described this as an economic or efficiency advantage; the report does not quantify how much it changes attack frequency, success rates or the resources needed to defend against them.
The development also puts AI on both sides of cybersecurity. Companies are adopting AI tools to detect fraud and other threats while, according to Hultquist, threat actors are using AI as well. That overlap makes unauthorized access to commercial models relevant to organizations managing account security, cloud resources and fraud controls.
Top picks for "acces becom commodity"
As an affiliate, we earn on qualifying purchases.
Companies Face AI-Driven Fraud
The reported market exists alongside legitimate, paid access to commercial AI services. The FT report says advanced ChatGPT and Claude subscriptions can cost up to $200 per user per month, while illicit listings may offer access at much lower prices. A discounted listing does not by itself establish whether a buyer receives reliable access, which account or resources were compromised, or how the access was obtained.
In research cited by PYMNTS, finance leaders also reported challenges verifying vendors. The report said close to 90% of finance leaders considered vendor verification a moderate or major burden, and identified AI-generated impersonation emails as a top concern among surveyed companies. These findings offer context on defensive pressures, but they do not establish a direct connection between vendor fraud and the marketplace listings described by the FT.
Scale and Sources Remain Unclear
The report does not state how many stolen accounts or computing resources have been identified, how many criminal groups are involved, or whether the reported increase reflects more incidents, more observed activity, or both. It also does not identify the marketplaces, list the specific models offered in each listing, or explain how researchers verified the advertised discounts.
It remains unclear how often buyers obtain working access, how long that access lasts, and what proportion of AI-related criminal activity relies on stolen credentials or computing resources. The cited reporting describes a trend and researchers’ observations; it does not quantify the resulting impact on victims or establish that every threat actor uses AI.
Monitoring Access and Account Abuse
Google Threat Intelligence researchers are tracking the reported growth in LLM-jacking, according to Hultquist’s comments to the FT. The source material does not give a schedule for future findings or identify a specific policy, enforcement action or security measure that will follow.
For companies, the immediate issue raised by the report is whether their AI accounts and computing resources are protected from unauthorized use while they continue to address AI-assisted impersonation and other fraud. The available reporting does not say which safeguards are most effective against the underground market or how quickly providers and defenders can limit access obtained through stolen credentials.
Key Questions
What does LLM-jacking mean in this report?
It refers to obtaining unauthorized access to large language models, including through stolen login credentials or computing resources taken over to run models without paying.
Which AI providers were named?
The Financial Times report named Anthropic, OpenAI and Google among the companies whose models were being offered through dark web marketplaces.
How large were the reported discounts?
The report said some marketplace access was offered at discounts of up to 97%, citing Google Threat Intelligence researchers. It did not state how many listings were reviewed or how the discounts were calculated.
Does the report show that AI has increased cyberattacks?
No. It reports that Google Threat Intelligence has seen a major increase in LLM-jacking this year and relays concerns about attackers’ access to AI. It does not quantify any change in attack frequency or success rates.
What remains unknown about the market?
The number of compromised accounts and resources, the size of the market, how often listings work and the effects on victims are not specified in the reported material.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
