The cybersecurity world woke up to a troubling new threat this week: Dolphin X, a remote access trojan that claims to use an artificial intelligence "profiler" to score and rank infected victims. The malware, analyzed by Varonis Threat Labs and reported by BleepingComputer, marks a worrying step forward in how cybercriminals are weaponizing AI to automate their operations. Security teams now face a threat that sorts its victims for maximum damage — a capability that would have seemed futuristic just a year ago.
Inside Dolphin X's AI-Powered Targeting System
Varonis researcher Daniel Kelley spotted the trojan being advertised on a cybercrime forum by a vendor using the alias "Kontraktnik." The seller promoted Dolphin X as an all-in-one remote access trojan, and according to Varonis, the operator panel lists 329 features across ten categories. That includes a credential-stealing module that claims to target more than 300 applications — covering browsers, password managers, cryptocurrency wallets, and cloud development tools.
But Dolphin X's standout feature is its "AI Profiler." This system analyzes information collected from infected computers and assigns each victim a risk score. The seller describes it as an "AI behavioral profiler with app usage tracking, risk score, and daily summary."
Here's how it works: credential-stealing malware can scoop up login data for hundreds of online accounts per victim. Manually reviewing all that stolen data to find the truly valuable targets has always been a bottleneck for attackers. Dolphin X's AI Profiler automates this sorting process by scoring, categorizing, and ranking infected computers so attackers know which victims are worth pursuing further.
The operator panel claims the AI Profiler can process victims' application usage, risk scores and tags, browser domains, and installed software to produce ranked profiles. These scores are delivered to attackers in daily summaries containing ranked victim profiles, allowing them to prioritize machines that may provide access to valuable accounts, cryptocurrency wallets, corporate networks, cloud environments, or production systems.
Kelley confirmed to BleepingComputer that the AI Profiler is genuinely present in the operator panel. His analysis discovered technical strings supporting the profiling workflow, including Auto-Start AI Profiler, ProfilerStart, ProfilerGetData, risk_score, risk_factors, and categoryusage. These strings indicate that the ranking system actually exists and processes real data on infected machines.
Varonis could not determine which specific AI engine produces the rankings without analyzing a live Dolphin X sample in the wild. The malware builder and its network traffic were examined in an isolated lab rather than running a live agent on a test machine, so some advertised capabilities remain unconfirmed.
Beyond Credential Theft — Developer and Cloud Infrastructure at Risk
Dolphin X doesn't stop at stealing passwords. The malware's credential-stealing module targets an unusually wide range of applications: 9 Chromium and Gecko browsers, 100 cryptocurrency wallet extensions, 65 desktop crypto wallets, 10 password managers, and more than 30 cloud command-line tools. It also claims to steal .env files, SSH keys, cloud access tokens, browser login data, and other developer credentials.
That list should concern anyone working in tech operations or cloud engineering. A developer's machine infected with Dolphin X could leak SSH private keys, AWS access tokens, and GitHub personal access tokens in a single sweep — giving attackers a direct path into production infrastructure behind the corporate firewall. A single compromised cloud CLI token can undo months of security hardening and infrastructure-as-code work.
The targeting of SSH keys and .env files is particularly dangerous. These files often contain database passwords, API secrets, and service credentials that developers store locally for convenience. Security teams frequently tell developers not to commit secrets to code repositories, but the same secrets sitting in local .env files are equally exposed if a machine gets infected.
AI in Malware Is No Longer Theoretical
Dolphin X joins a growing roster of cybercrime tools that lean on AI. Threat actors have already launched services like SpamGPT for crafting convincing phishing emails, and earlier this week BleepingComputer reported on a Hermes AI agent being used to automate an attack on Thailand's Finance Ministry. What makes Dolphin X different is that it applies AI to an operational pain point — sorting stolen data at scale — rather than content generation or social engineering.
"The use of AI to sift through massive amounts of stolen credentials is a logical next step for cybercriminals," security experts note. Credential theft has always produced a firehose of data; finding the needles in that haystack used to require manual effort and human analysis. Dolphin X claims to automate that triage, making every stolen credential set potentially more dangerous than before.
The trend mirrors a broader shift in how threats operate. Attackers are moving from spray-and-pray operations to targeted, data-driven campaigns. The AI Profiler essentially acts as a force multiplier: one infected machine feeds data into the profiler, which then tells the attacker which ten machines out of a thousand deserve follow-up attention and further exploitation.
CISA issued a separate warning this week about the Russian state-sponsored hacking group Laundry Bear (also known as Void Blizzard) exploiting a zero-click Zimbra collaboration vulnerability. The group is using CVE-2025-66376, a cross-site scripting flaw in Zimbra's Classic UI, to automatically steal emails, passwords, and two-factor authentication tokens from victims who simply view a crafted message in their inbox. The overlap between AI-powered commodity malware and state-sponsored espionage operations paints a picture of an increasingly automated threat environment.
The Defenders' Dilemma — Guardrails vs. Guard Dogs
While cybercriminals are freely adopting AI for their operations, legitimate security researchers face growing obstacles from the same technology. A detailed report from TechCrunch this week highlighted how AI guardrails — safety restrictions designed to prevent malicious use of language models — are increasingly frustrating the work of offensive security professionals whose job is to find and fix vulnerabilities before criminals can exploit them.
Mark Dowd, a veteran security researcher with decades of zero-day experience, told TechCrunch that "it's not really comfortable to me that these random large companies are making arbitrary decisions about what is safe in security and what's not." Chris Anley, chief scientist at NCC Group, pointed out that a prompt as innocent as "fix this code" is both a defensive essential and an offensive roadmap, and the two uses "can't really be unpicked."
Paolo Stagno, chief technology officer at Crowdfense, agreed with Dowd, saying AI companies "essentially treat customers like children who need babysitting" with their vetted programs and guardrails. Stagno said his team uses frontier models only for reverse engineering, avoiding cloud-based AI for vulnerability discovery work due to the risk of leaking sensitive data into training runs. For actual exploit development, they use open-source models run locally.
The paradox is stark: while AI companies restrict access to frontier models through vetted programs like OpenAI's Trusted Access for Cyber and Anthropic's Cyber Verification Program, criminals face no such restrictions. Open-source models — including Chinese models like GLM that can be downloaded and run locally with zero guardrails — are readily available to anyone with a GPU. As Chris Thompson, CEO of RemoteThreat, told TechCrunch, "responsible researchers are being pushed away from U.S.-governed systems to foreign-owned systems."
Thompson warned of "a big storm coming — attacks at speed and scale like never before." Dolphin X, with its AI-powered victim ranking, may be an early wave of that storm.
The message for defenders is clear: AI is now a double-edged sword wielded by both attackers and defenders. The question is which side wields it more effectively. For more on cybersecurity threats and defense strategies, check out our Cybersecurity coverage.