With the advent of AI, productivity has increased significantly. But it also opens the door for threat actors to exploit new technologies for malicious purposes.
As organizations adopt new AI technologies, their attack surface expands exponentially. The cybercrime arena that once required skilled operators with advanced hacking skills is now accessible to anyone with a laptop who can effectively use AI prompts. The scope of AI crimes is wide-ranging and poses several risks to businesses, including financial fraud, data poisoning, and malware.
But by investing in AI security tools, fostering a culture of validation, implementing strong governance, and participating in the global security community, companies can create an environment where AI acts as an asset rather than a security risk. These steps can help businesses become more resilient to AI attacks and prevent and manage their impact.
AI crime and its impact on business
AI crime is a cyber attack that uses AI technology as a weapon or accelerator to attack organizations or individuals. AI crime is more dangerous than traditional cybercrime. For example, AI-generated malware may change its code signature multiple times, unlike regular static malware. This makes it very difficult for detection systems to detect.
AI in cybercrime has evolved from simple spam email automation to attackers carrying out fully autonomous attacks, from reconnaissance to exploitation. Before 2023, attackers focused on using machine learning to bypass automated spam filters and automate vulnerability scans on IT infrastructure. GenAI has since become widely available, allowing cybercriminals to use LLM to generate customized email messages in multiple languages, generate basic malware code, and power social engineering attacks.
Agent AI now rules the world. The AI agent can perform the entire sequence of steps without human intervention. Rather than hackers running separate tools one by one to accomplish a single function, AI agents can plan and execute steps throughout the attack lifecycle with minimal human oversight.
AI-powered crime cannot be considered an extension of regular cybercrime, as it exceeds it in terms of scale, speed, and credibility. The impact of AI crime on businesses can be severe, from operations to finance to strategy. Consider the following impacts of AI crime on your business:
- Financial fraud. Threat actors can use GenAI to perform advanced attacks and create convincing voice clones and deepfake videos. These could allow an attacker to bypass traditional validation controls and perform malicious actions, such as approving an emergency wire transfer using a CFO’s voice notes.
- It undermines digital trust. When customers can’t differentiate between real and fake communications, such as emails, support chats, and executive messages, they lose trust in your brand. This could negatively impact our business and lead to loss of customers, market share and partner relationships.
- Data breach. Cybercriminals can automate phishing emails, malware generation, credential stuffing, and vulnerability discovery to extract sensitive business data. These attacks have a high success rate because attackers can customize their strategies for each target.
- Damage to reputation. Threat actors can generate large amounts of content, such as fake news, reviews, and executive quotes, and spread it across social media platforms to tarnish a targeted company’s public image within hours.
Types of AI crimes
Rapid advances in AI are introducing new and advanced threats to the cybersecurity environment. Understanding these threats is essential to proposing mitigation strategies to counter them.
AI-powered phishing and social engineering
This is the most immediately and economically destructive form of AI crime. GenAI allows attackers to generate clean, grammatically correct messages in multiple languages at scale. Generate custom emails according to specific target profiles in minutes. These messages can trick victims into providing credentials, authorizing payments, or revealing sensitive data.
According to a study conducted by security firm Barracuda in collaboration with researchers from Columbia University and the University of Chicago, 51% of all spam is now generated by AI. Until the arrival of ChatGPT in late 2022, this percentage was close to zero. Attackers often use spam emails to send malware to unsuspecting victims.
Targeted email attacks such as spear phishing and business email compromise are becoming more widespread and effective with the help of AI. AegisAI’s report, “The State of AI Threats in Email: 2025,” found that AI-generated spear-phishing emails can bypass traditional spam filters more than 50% of the time.
deepfake scam
Deepfake scams are becoming the most threatening content hackers generate using AI. Deepfakes use synthetic media, such as AI-generated audio, video, and images, to create a realistic impersonation of an individual. While the use of deepfake technology is increasing, its cost continues to fall, making it accessible to criminals with low technical skills.
According to research by VPN provider Surfshark, the cost of losses from deepfake attacks in 2025 is estimated to triple to $1.1 billion from $360 million in 2024, and a nine-fold increase from $128 million from 2020 to 2023.
In one high-profile incident, attackers targeted a British multinational engineering company and were able to persuade a financial representative to hand over $25 million. The employee was attending a video conference call with the alleged CFO and other staff members. However, the attackers used deepfake technology to generate other members of the call.
AI-powered malware and ransomware
Threat actors also use AI to create new malware strains. Using AI in malware code and delivery mechanisms can improve attackers’ accuracy and ability to evade detection. According to SQ Magazine, as of 2025, 41% of ransomware families will include AI components for adaptive payload delivery.
Autonomous malware such as PromptLock uses GenAI to carry out attacks. Such malware runs locally accessible AI language models to generate harmful Lua scripts that run on Windows, Linux, and macOS in real time. These tools independently decide whether to steal or encrypt the data they find based on the text prompts you set.
Data poisoning and model manipulation
Data poisoning and model manipulation threaten the accuracy and integrity of AI systems. Data poisoning involves injecting malicious data into a model’s training data set (or its supply chain data set) to affect its output. Poisoning data can have a significant impact on AI model decisions. IACIS’ Data Poisoning 2018-2025: A Systematic Review of Risks, Impacts, and Mitigation Challenges study states that 0.001% perturbation of training data can reduce model accuracy by up to 30%. Injecting malicious data into the training data set can also create backdoors that threat actors can exploit.
This issue can be problematic for companies that are building their own internal AI models. Suppose a company trains a customer service chatbot on malicious data. This bot can provide inappropriate results to customers, and businesses can continue to operate without discovering this for long periods of time.
Crime LLM
The Crime LLM is the latest addition to the AI crime ecosystem. Threat actors exploit LLMs for malicious purposes, sometimes using legitimate LLMs such as ChatGPT and Gemini, to remove protections set by developers (i.e., jailbreaking). This can provide malicious output to attackers or develop criminal LLMs specifically for malicious purposes.
There are various crime LLM tools such as WormGPT, GhostGPT, KawaiiGPT, etc. The final tool is a free, open-source malicious LLM that can bypass the safety restrictions of standard AI models. This allows users to create unlimited amounts of malicious output for cyberattacks.
How to build resilience against AI crime
As AI crimes become more sophisticated, organizations need to go beyond traditional defenses and create comprehensive security models to counter AI-powered cyberattacks. Building such systems requires a multi-layered approach that combines technical defenses and controls, informed human resources, strong governance, and collaborative networks.
Invest in AI security tools
Technical controls remain the first line of defense against AI threats. Organizations must invest in a new generation of security tools that can understand and stop AI-specific threats. For example, traditional firewalls and antivirus software may not be able to detect prompt injection or deepfake manipulation attacks. The table below lists some tools that can help organizations detect the most common types of AI threats.
Employee training and awareness
Humans remain the weakest element in any cyber defense strategy, and AI amplifies this risk by making deception much more convincing. Companies should upgrade their traditional security awareness programs to go beyond basic phishing email detection and include lessons on AI threat detection. Employees also need to understand how to detect deepfakes and highly customized spear-phishing emails.
For high-risk work operations, such as performing wire transfers, resetting credentials, or sharing sensitive data, businesses should set up multichannel verification mechanisms. Suppose an employee receives a payment request via email. Verification must be through a known phone number or secure internal channel, not through one provided in a message. This verification prevents threat actors from controlling the entire communication chain.
Strengthening governance policy
AI governance provides the necessary guardrails to safely deploy AI and ensure that security is not compromised. Strong governance includes several areas, including creating clear rules for the use of AI within work environments, ensuring complete visibility into AI systems, and ensuring humans are kept informed as AI makes important decisions.
A key element of governance policy is having an acceptable use policy (AUP) for AI. This is a formal document that establishes the rules for employees to use AI in their workspaces. AI AUPs must address the risks of inappropriate use of AI, including data breaches, piracy, and bias in decision-making.
Companies should also maintain an AI bill of materials, a document that lists all the components used to build an AI model. This should include AI-specific components such as models, datasets, and prompts. As enterprises rely on AI to run their workloads, it’s important to understand the components that make up the system so you can answer important questions like “Where did this model come from?” or “What data influenced that behavior to give a particular response or decision?”
Collaborate with industry experts
AI crimes are becoming more sophisticated and are extremely difficult to defend against alone. To stay ahead of emerging AI threats, it’s important to collaborate with other companies and government agencies. Threat intelligence feeds provide real-time information about new AI-related attacks and indicators of compromise, such as new prompt injection payloads and deepfake signatures. Adopting standards such as the OWASP Top 10 for LLM Applications and the NIST AI Risk Management Framework can help companies establish a language for exchanging information about AI attacks and best countermeasures.
Nihad A. Hassan is an independent cybersecurity consultant, digital forensics and cyber OSINT expert, online blogger, and author with over 15 years of experience in information security research. He is the author of six books and numerous articles on information security. Nihad is deeply involved in security training, education, and motivation.
