The rise of AI‑powered infostealers: threat actors automate victim profiling
Introduction
Illicit forums are increasingly hosting infostealers featuring built‑in automated AI profiling of compromised systems. The BI.ZONE Threat Intelligence team has identified two such listings: the Windows‑based Bee Stealer and AVENGER, a comprehensive malware agent designed for macOS.
The baseline feature set of modern infostealers has remained largely unchanged for years. A standard data collection profile typically includes:
- browser data: passwords, active session cookies, autofill data, browsing history, and saved banking card details
- crypto assets: browser wallet extensions, desktop wallets, and crypto exchange account credentials
- messengers and email: Telegram, Discord, WhatsApp
, and Viber sessions, and email clients such as Outlook and Thunderbird - system and network credentials: SSH keys, configuration files, cloud credentials, password managers (KeePass, Bitwarden, etc.), and FTP credentials
- files and environment: screenshots, webcam snapshots, system information, documents matching specific filename patterns in user directories
Modern infostealers support over 100 browser versions and bypass current protection mechanisms, including Google Chrome’s App‑Bound Encryption. Additional core functionalities include establishing persistence, clipboard hijacking, and self‑deletion.
A relatively new but now established capability deserves special mention: the theft of AI agent data. Modern infostealer variants exfiltrate authentication data and chat histories from AI coding/development assistants and LLM clients. This data commands a premium on the dark web. Compromised sessions are resold to provide unauthorized access to paid subscriptions and API keys. Meanwhile, chat histories often contain source code, internal documents, tokens, and sensitive account details inadvertently shared by users. This intelligence provides threat actors with deep insights into the victim and their employer.
Against this backdrop, automated AI profiling emerges as the next stage in the evolution of malware.
Next-generation infostealers
How does AI profiling work? The exfiltrated data is automatically fed into a language model, which generates a concise dossier on the user of the compromised device. In the case of Bee Stealer, a significant portion of the stolen log is processed by a neural network directly on the attacker’s control panel, with the results saved as ProfileRu.txt and ProfileEn.txt.
The resulting report automatically populates fields describing the victim’s likely profession, areas of interest, frequently visited websites, and the purpose of key domains found in the log. The malware’s author provides an example where a compromised user was flagged as an “IT specialist and web developer with an interest in security,” alongside highlighted development‑related activities and selected high‑value domains.
In contrast, the multifunctional macOS AVENGER malware analyzes browser and command‑line history, messages, email, and configuration files to construct a victim profile that explicitly identifies their workplace, primary banks, cryptocurrency exchanges, and digital assets.
The attacker´s advantage
This new capability allows threat actors to radically accelerate their workflow when sorting through vast log arrays that contain thousands of stolen cookies, accounts, and domains. Manually assessing whether a collected log belongs to an average user or an enterprise employee with access to critical infrastructure takes a lot of time.
AI profiling addresses this bottleneck by delivering several key advantages:
- speeds up the analysis of large log datasets
- enables attackers to quickly prioritize and filter corporate employees, developers, and high‑net‑worth crypto holders before the compromised sessions expire or the user resets their credentials
- generates clean, structured profile reports tailored for secondary markets, as logs can be immediately sorted and sold to targeted buyers as “developer access” or “financial access” rather than as an unprocessed data dump
- lowers the skill barrier, with threat actors no longer needing advanced technical expertise to evaluate the collected data or determine its market value
Conclusion and outlook for the future
AI profiling enables adversaries to rapidly assess the value of a compromised device and immediately exploit the harvested intelligence. However, practical defense recommendations remain unchanged: upon detecting a compromise, organizations should immediately terminate active sessions, revoke security tokens, and reset all credentials. Because AI‑driven analysis happens post‑exfiltration on the attacker’s infrastructure, monitoring outbound network traffic is equally critical.
The appearance of multiple listings for infostealers with AI profiling capabilities indicates where this malware category is headed. We project a shift from passive victim profiling to the automation of the attack lifecycle. Future AI models integrated into malware will likely independently select the most lucrative monetization strategy: hijacking high‑priority active sessions, restoring access via stolen tokens, logging into corporate cloud services, or conducting automated spearphishing campaigns directly from the compromised employee’s account.
To effectively secure your business and stay one step ahead of adversaries, it is crucial to track these shifts in the cyber threat landscape. All critical insights—including threat research and deep‑dive articles—are continuously aggregated in our Threat Zone hub.