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 cybercriminals are increasingly trading stolen AI credentials and computing access, with some dark web listings offering access to major models at discounts of up to 97%. Google Threat Intelligence Group analyst John Hultquist says the activity could give attackers a cost advantage, while companies are also using AI to defend against fraud.
Cybercriminals are increasingly trading stolen AI credentials and computing access, according to a Financial Times report published Sept. 27, a development that could lower the cost of using advanced models for extortion, espionage and other attacks. John Hultquist of Google Threat Intelligence Group told the FT that researchers have seen a major increase in so-called LLM-jacking this year, though the report does not quantify the increase.
The activity described includes selling pilfered login credentials for public AI services and taking computing resources to run models without paying for access. The FT report, as summarized by PYMNTS, cites Google Threat Intelligence researchers as saying dark web marketplaces offer access to models from Anthropic, OpenAI and Google at discounts of up to 97%. The report does not specify how many listings were examined or how the discount figure was calculated.
Access to advanced AI subscriptions can be expensive: the source report says the highest-level versions of ChatGPT and Claude can cost as much as $200 per user each month. Hultquist told the FT that attackers who obtain access at lower cost could gain an economic or efficiency advantage over organizations paying to use comparable tools for defense. His assessment is a warning about potential effects; the report does not establish how much stolen access has changed attackers’ capabilities or outcomes.
The same source material points to companies’ growing use of AI in cybersecurity. PYMNTS Intelligence research found that 42% of surveyed companies use three or more AI defense tools in a layered stack. In that research, half of the companies surveyed identified AI-generated vendor impersonation emails as their leading threat. PYMNTS also reported that close to 90% of finance leaders called vendor verification a moderate or major burden. Those figures describe the survey respondents and the specific research; they do not measure all companies or all cyber threats.
The Cost Gap in AI Defense
Cheap access could change the economics of cybercrime by letting attackers use paid models or computing capacity without bearing the full cost. Hultquist’s concern is that defenders may have to spend more to match capabilities that criminals can obtain through stolen accounts or resources. The report describes this as a potential cost and efficiency advantage; it provides no estimate of financial losses caused by LLM-jacking.
The risk sits alongside the expanding use of AI in fraud and security work. Companies are deploying multiple AI defenses, while also reporting AI-generated impersonation as a leading concern. That overlap means access controls, account security and vendor verification may matter as much as the models themselves. The cited research shows that verification already takes time and money for many finance teams, but it does not establish that stolen model access is behind the impersonation emails respondents reported.
For readers and organizations, the development signals that AI access is itself becoming a target and a traded resource. Security teams may need to account for compromised credentials and unauthorized computing use when assessing threats. The available reporting supports concern about an emerging market, but leaves its scale and direct impact open.
Top picks for "acces becom commodity"
As an affiliate, we earn on qualifying purchases.
From Paid Models to Stolen Access
Large language models require accounts, subscription payments or computing resources. The FT report describes a criminal market in which attackers seek to bypass those costs by acquiring credentials or using computing capacity without authorization. It refers to this activity as LLM-jacking. The source material does not provide a technical breakdown of how each intrusion works, nor does it say that every listing offers direct access to a model provider’s systems.
The report places the activity within a wider shift in which threat actors are using AI tools. Hultquist told the FT that AI is now used by “every threat actor,” a broad characterization attributed to him rather than a measured finding presented in the article. PYMNTS’ separate research describes the defensive side of this competition: companies are adopting AI tools even as they identify AI-assisted impersonation as a major concern. These strands provide context, but the material does not show that the surveyed firms were affected by the specific access market described by Google researchers.
“Catching these fraud attempts is hard work, costing firms time and money.”
— PYMNTS Intelligence, in “Prevention First: Building a Smarter Defense Against Payments Fraud”
Scale and Impact Still Unknown
The available account does not give the number of compromised accounts, marketplace listings or affected organizations. It also does not explain the time period or calculation behind the reported discounts of up to 97%. Google Threat Intelligence Group’s characterization of a major increase is attributed to Hultquist, but no comparison baseline or numerical measure is supplied.
It remains unclear how often the access is used for extortion, warfare or espionage, and what proportion of listings lead to successful model use. The source material does not identify specific buyers or sellers, describe confirmed incidents tied to the marketplace activity, or quantify resulting harm. The reported corporate survey figures concern AI defenses and impersonation threats, not the prevalence of LLM-jacking.
Security Teams Track Access Abuse
The report offers no specific enforcement action, investigation timeline or forthcoming release of additional data. The next developments to watch are further findings from Google Threat Intelligence Group on the scale and methods of unauthorized AI access, and whether providers disclose changes to account protections or abuse monitoring. Those steps are possibilities to monitor, not announced actions in the source material.
For organizations, the immediate issue described is protecting AI accounts and computing resources while assessing how AI tools affect both attacks and defenses. More evidence would be needed to determine whether the reported underground market is growing in volume, driving measurable incidents, or changing the balance between attackers and defenders.
Key Questions
What does “LLM-jacking” mean in this report?
It refers to the unauthorized use of AI access, including selling stolen login credentials or taking computing resources to run models without paying for them.
Which AI companies are named?
The report says dark web marketplaces offer access to models from Anthropic, OpenAI and Google. It does not identify particular sellers or listings.
Are the reported discounts verified across the market?
The report cites Google Threat Intelligence researchers for discounts of up to 97%. It gives no listing count, calculation method or market-wide comparison, so the figure should be read as a reported maximum.
How much of a cost advantage do attackers gain?
Hultquist says cheaper access could create an economic or efficiency advantage, but the source material provides no quantified estimate of the savings or resulting impact.
What do companies say about AI-related cyber threats?
In separate PYMNTS Intelligence research, 50% of surveyed companies named AI-generated vendor impersonation emails as their leading threat. That survey finding does not measure LLM-jacking incidents.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
