On July 23, 2026, Varonis Threat Labs uncovered a new strain of RAT that markets itself as an AI profiling malware solution. The Dolphin X trojan, advertised on a cybercrime forum by a vendor called “Kontraktnik,” claims to score and rank infected users, handing attackers a prioritized list of high‑value targets.
Key Takeaways
- Operator panel lists 329 features across ten categories, including a credential‑stealing module that targets over 300 applications.
- The “AI Profiler” purports to analyze app usage, browser domains, and installed software to assign a daily risk score.
- Varonis confirmed the presence of strings like
ProfilerStartandrisk_score, indicating the profiling workflow is built into the panel. - The specific AI engine remains unknown because Varonis never ran a live Dolphin X agent on an infected host.
- Security teams log only 54% of successful attacks and alert on just 14%, leaving many of these prioritized victims undetected.
Historical Context
Malware that claims to use artificial intelligence is not brand new. Earlier in the decade, threat actors began experimenting with large language models to automate phishing content and generate malicious code snippets. Those efforts focused on the creation stage—making the attack itself more convincing.
Dolphin X flips the script. Instead of using AI to write the payload, it promises to let operators sort the loot after the infection. The shift mirrors a broader industry trend where automation moves from the front end of an attack to the back end, where data is abundant but time is scarce.
When Varonis first saw the advert, the team recognized a pattern that had emerged in other forums: vendors touting “AI‑enhanced” capabilities to differentiate their tools. The claim of a “behavioral profiler” echoes earlier promises made by ransomware‑as‑a‑service platforms, which suggested predictive targeting but rarely delivered a demonstrable model.
That history matters because it sets expectations. Operators have grown accustomed to hype that never materializes. In Dolphin X’s case, the presence of profiling strings gives the claim some weight, yet the lack of a live sample keeps the true sophistication hidden.
AI Profiling Malware: Dolphin X’s New Threat
Varonis researcher Daniel Kelley spotted the ransomware‑like advert on a forum and immediately requested the operator panel for analysis. That’s how the team got a look at the claimed AI‑driven ranking system without ever executing the payload on a real machine. “Beyond credential collection, the panel includes a surveillance tab containing the AI Profiler. The seller describes it as an ‘AI behavioral profiler with app usage tracking, risk score, and daily summary’,” the report says.
What’s Inside the Operator Panel
The panel is a sprawling dashboard that enumerates 329 features split into ten categories. Among them, credential‑stealing is the most heavily advertised capability, with the panel boasting support for more than 300 applications. That includes nine Chromium‑based browsers, ten password managers, and a host of cryptocurrency wallets.
But the real headline is the AI Profiler tab. It claims to ingest data from infected hosts, then churn out a daily summary that ranks victims by perceived value. The summary supposedly highlights machines that might hold cryptocurrency, corporate credentials, or cloud tokens.
How the AI Profiler Claims to Work
According to the panel, the AI Profiler processes “application usage, risk scores and tags, browser domains, and installed software” to produce a ranked profile. The idea is that attackers can skip the tedious manual review of thousands of stolen credentials and jump straight to the juicy ones. “In practice, the feature appears designed to help operators triage victims,” Kelley explained.
Technical strings found by Varonis—Auto-Start AI Profiler, ProfilerGetData, risk_factors, and categoryusage—suggest the workflow is baked into the panel. However, Varonis couldn’t confirm which machine‑learning model or rule set powers the rankings because they never observed a live sample generating scores.
“Beyond credential collection, the panel includes a surveillance tab containing the AI Profiler. The seller describes it as an ‘AI behavioral profiler with app usage tracking, risk score, and daily summary’,”
Credential‑Stealing Capabilities and Claims
While the AI Profiler gets most of the buzz, the underlying RAT still functions as a conventional credential stealer. The operator panel lists more than 300 applications as targets, ranging from popular browsers to niche developer tools.
Targeted Applications
- 9 Chromium and Gecko browsers
- 100 cryptocurrency wallet extensions
- 65 desktop crypto wallets
- 10 password managers
- 30+ cloud command‑line tools
It also claims to harvest.env files, SSH keys, cloud access tokens, and other developer credentials. The panel says it can pull data from browser login stores, cryptocurrency wallets, and even server configuration files.
Because Varonis analyzed only the panel and network traffic—never a live malware instance—the advertised collection capabilities haven’t been independently verified. That’s a gap that makes it hard to gauge the true impact of the threat.
Limitations of the Analysis
The biggest blind spot is the unknown AI engine. Varonis saw the strings that would trigger profiling, but without a live sample they couldn’t extract the model or see the actual risk scores being generated. That means the claimed “AI” could be a simple rule‑based system rather than a sophisticated neural net.
Another limitation is the lack of real‑world infection data. The forum advert says the tool is sold as an all‑in‑one solution, but we don’t know how many criminal groups have actually deployed it, or how effective the ranking really is in practice.
Why AI in Malware Is Growing
AI isn’t new to cybercrime. Services like SpamGPT and autonomous attack agents have already shown how large language models can be weaponized. Dolphin X’s approach is different: it uses AI to solve an operational problem—sorting massive amounts of stolen data—rather than generating new exploits.
That shift matters because it lowers the barrier for less‑skilled actors. If a cheap RAT can automatically hand over a prioritized list of high‑value victims, the economics of ransomware and credential‑theft campaigns could tilt even further toward automation.
Security teams already struggle: they log 54% of successful attacks but only alert on 14%. An AI‑driven ranking system could make the remaining 86% of alerts even harder to triage, letting attackers slip through unnoticed.
What This Means For You
Developers should treat any.env, SSH key, or cloud token as high‑value data. If a trojan like Dolphin X lands on a machine, those files could be the very data the AI Profiler flags for attackers. Implement strict access controls, rotate secrets frequently, and monitor for abnormal file access patterns.
Security teams need to adapt their detection playbooks. Because the AI Profiler only surfaces in the operator panel, traditional endpoint telemetry might miss the profiling step entirely. Look for the network calls tied to the strings Kelley identified—especially those that query for ProfilerGetData or push daily summaries.
In short, AI’s growth‑enabled profiling forces defenders to think beyond “did the malware run?” and ask “what does the malware do with the data it gathers?”
Concrete Scenarios for Different Roles
For a DevOps engineer, the threat translates to a need for tighter secret management. Imagine a compromised build server that stores deployment keys in a.env file. The AI Profiler could tag that host as a top‑priority target, prompting attackers to move quickly before the key is rotated.
For a C‑suite executive, the headline number—only 14% of attacks trigger alerts—highlights a governance gap. An organization that relies on manual ticket triage may miss the handful of high‑value victims that the AI Profiler surfaces. Investing in automated alert enrichment could close that gap.
For a SOC analyst, the presence of strings like ProfilerGetData offers a concrete hunting indicator. Craft a detection rule that flags outbound connections to unknown domains shortly after a host contacts a known command‑and‑control server. That pattern could expose the profiling phase even if the actual scoring never reaches the analyst’s view.
Competitive Landscape
Other malware families have flirted with AI, but most remain on the periphery of hype. SpamGPT, for instance, generates phishing content on demand, while autonomous attack agents can pivot between exploits without human direction. Dolphin X distinguishes itself by embedding a data‑ranking component into its operational flow.
That distinction creates a niche that could inspire copycats. If one group demonstrates a functional profiling engine, competitors may adopt similar dashboards to stay relevant. The marketplace for “AI‑enhanced” RATs could expand, pushing the baseline expectation for threat actors to include some form of automated data prioritization.
Defenders should watch for overlapping feature sets. The 329‑feature panel covers credential theft, surveillance, and profiling—all in one interface. Future samples may trim the list but retain the core ranking capability, making the profiler a reusable module across different malware families.
Key Questions Remaining
- What exact algorithm or rule set drives the daily risk score?
- How many active deployments exist in the wild, and what is their success rate?
- Can defenders reliably emulate the profiling process to generate decoy data and waste attacker resources?
- Will law‑enforcement agencies prioritize tracking the vendor “Kontraktnik,” or focus on the infrastructure used to deliver the RAT?
Answers to these questions will shape the next wave of defenses. Until then, organizations must assume the profiler is functional and act accordingly.
Sources: BleepingComputer, original report
