AI on the dark web: the cybersecurity threats explained

01How AI is used on the dark web

The dark web is the World Wide Web content that exists on darknets that use the Internet, but require specific software, configurations, or authorization to access. Through the dark web, private computer networks can communicate and conduct business anonymously without divulging identifying information, such as a user's location. The dark web forms a small part of the deep web, the part of the web not indexed by web search engines, although sometimes the term deep web is mistakenly used to refer specifically to the dark web.

The dark web has become a distribution platform for AI-powered criminal tools. Forums and marketplaces hosted on Tor hidden services now offer automated phishing kits, deepfake generation services, and AI-assisted malware builders. The same generative AI technologies that power legitimate applications are being repurposed for fraud, impersonation, and exploitation.

What makes this development significant is the lowering of technical barriers. Previously, conducting sophisticated cyberattacks required specialized knowledge in coding, networking, and social engineering. AI tools available on dark web markets now automate much of this work, allowing less-skilled actors to launch attacks that once required expert-level capabilities.

02Automated phishing and social engineering

Cybercrime encompasses a wide range of criminal activities that are carried out using digital devices and/or networks. It has been variously defined as "a crime committed on a computer network, especially the Internet". Cybercriminals may exploit vulnerabilities in computer systems and networks to gain unauthorized access, steal sensitive information, disrupt services, and cause financial or reputational harm to individuals, organizations, and governments.

AI has transformed phishing from a volume game into a precision weapon. Large language models can generate convincing, personalized phishing emails in bulk, scraping target information from social media and corporate websites to craft messages that are far more likely to deceive. The telltale signs of phishing, such as poor grammar and generic greetings, are no longer reliable indicators when the text is AI-generated.

Beyond email, AI-powered voice cloning and deepfake video technology have enabled new forms of social engineering. There have been documented cases of fraudsters using AI voice cloning to impersonate executives, convincing employees to authorize large wire transfers. The technology to clone a voice from a few seconds of audio is now widely available and inexpensive.

Dark web marketplaces sell phishing-as-a-service kits that include AI-generated email templates, landing page generators, and credential harvesting tools. These kits lower the cost of launching a phishing campaign to a fraction of what it would cost to build from scratch, democratizing access to what was once a specialized criminal skill.

03AI-generated malware and evasion

The use of AI in malware development is still emerging but growing rapidly. Researchers have demonstrated that large language models can generate functional malicious code, including keyloggers, ransomware payloads, and encryption scripts. While the output is not yet as sophisticated as hand-crafted malware by experienced developers, the gap is narrowing as models improve.

AI is also being used for evasion. Machine learning models can analyze antivirus signatures and behavioral detection patterns to generate polymorphic malware that alters its code structure to avoid detection. Some dark web tools advertise AI-driven packing and obfuscation services that automatically modify malware to bypass common security products.

Artificial intelligence (AI) and its subfields have been used in applications throughout industry and academia. Machine learning has been used for various scientific and commercial purposes, including language translation, image recognition, decision-making, credit scoring, and e-commerce. Since the 2020s, massive advancements have been made in the field of generative artificial intelligence (GenAI), which generates text, images, music, videos, and other forms of data. The dual-use nature of these technologies means that the same advancements improving legitimate security products are simultaneously being adapted for offensive use by criminal actors.

04The dark web marketplace economy

Dark web marketplaces operate as sophisticated e-commerce platforms, complete with vendor ratings, escrow services, dispute resolution, and customer support. Transactions are conducted in cryptocurrency, primarily Bitcoin and Monero, providing a degree of financial anonymity. The market for cybercrime tools and stolen data is estimated to be worth tens of billions of dollars annually.

AI-powered tools command premium prices on these markets. A custom deepfake video can sell for several hundred dollars, while a phishing-as-a-service subscription can cost hundreds per month. The economics incentivize continued development: skilled developers can earn significant income creating AI tools for criminal use, often more than they would earn in legitimate cybersecurity roles.

Law enforcement takedowns, such as the seizure of the Hydra marketplace in 2022 and Wall Street Market in 2019, temporarily disrupt these markets but do not eliminate them. New marketplaces emerge to replace those that are shut down, often with improved operational security and decentralized infrastructure that makes takedowns more difficult.

05How law enforcement is responding

Law enforcement agencies are increasingly using AI themselves to combat cybercrime. The FBI, Europol, and other agencies employ machine learning to analyze dark web forum data, identify patterns in criminal activity, and track cryptocurrency transactions. AI-powered tools can process vast amounts of data from dark web monitoring, flagging threats that human analysts might miss.

International cooperation has intensified. Joint operations between the FBI, Europol, and national police forces have led to significant arrests and marketplace seizures. The takedown of the LockBit ransomware group in 2024 demonstrated that even well-organized criminal operations with sophisticated infrastructure are vulnerable to coordinated law enforcement action.

However, the asymmetry of the conflict favors attackers. Cybercriminals can operate from jurisdictions with weak cybercrime enforcement, while law enforcement must navigate complex international legal frameworks. The speed at which AI tools evolve also outpaces the regulatory and legislative processes needed to address them.

06What organizations should do to protect themselves

Defending against AI-powered cybercrime requires a multi-layered approach. Traditional perimeter defenses, such as firewalls and antivirus, remain necessary but are no longer sufficient. Organizations should implement zero-trust architecture, where every access request is verified regardless of its origin, reducing the blast radius of compromised credentials.

Employee training is critical, as social engineering remains the most common entry point for attacks. Training should specifically address AI-generated phishing and deepfake impersonation, teaching employees to verify unusual requests through secondary channels. Organizations should establish verification protocols for financial transactions and sensitive data access that do not rely solely on email or phone communication.

Investment in AI-powered security tools is becoming essential. Threat detection platforms that use machine learning to identify anomalous behavior, automated incident response systems, and continuous dark web monitoring services can provide early warning of emerging threats. The cost of these tools must be weighed against the cost of a breach, which for many organizations far exceeds the investment in prevention.

07The future of AI-powered cybercrime

The trajectory of AI-powered cybercrime points toward increasing sophistication and automation. As generative AI models continue to improve, the quality of phishing content, deepfakes, and AI-generated malware will approach levels that are indistinguishable from human-created content. Automated attack pipelines could eventually operate with minimal human oversight, scanning for vulnerabilities, crafting exploits, and deploying payloads in near real-time.

The cybersecurity industry is responding with AI-powered defenses, creating an arms race between offensive and defensive AI. Automated threat hunting, behavioral analysis, and predictive defense systems are becoming standard in enterprise security. The question is whether defensive AI can keep pace with offensive innovation, given that attackers need only find a single vulnerability while defenders must protect all of them.

Regulatory and policy responses will play a crucial role. Governments are grappling with how to regulate AI development to prevent misuse while not stifling innovation. Export controls on AI technology, requirements for safety testing of AI models, and international agreements on AI governance are all under discussion. The effectiveness of these measures will depend on global cooperation and enforcement, both of which remain uncertain in the current geopolitical climate.