AI cyber threats in 2026: what the FBI and ENISA report

This list groups every AI-related passage in two public reports into six themes: the FBI IC3 2025 Annual Report (2025 complaints) and the ENISA Threat Landscape 2026 (incidents from 1 January 2025 to 31 December 2025). The themes are selected by that criterion, so they are not ranked. Each theme says who reports it.

What do the FBI and ENISA say about AI-enabled threats?

The FBI IC3 report says AI is a tool that can serve legitimate or criminal purposes. It adds that AI-enabled synthetic content is becoming increasingly difficult to detect and easier to make. ENISA reports the continued integration of artificial intelligence into malicious activity by cybercriminal operators, State-nexus intrusion sets and Information Manipulation Sets (IMS). These are the FBI’s headline numbers for 2025.

The FBI also notes that overall investment scam losses exceeded $8 billion, and says many victims do not realize the extent AI may be involved.

Theme 1: How is AI used in fraud and scams? (FBI and ENISA)

The FBI describes several scam types. In business email compromise (BEC), chat generators can quickly create official-sounding emails mimicking a company’s CEO. Businesses reported losses over $30 million to BEC scams involving AI. In confidence and romance scams, AI chat generators produce fake profiles and scripts. Victims lost over $19 million to scams with a likely AI-nexus.

In investment scams, the FBI says subjects use AI to quickly generate thousands of conversations that appear different to each prospective victim. Its complaint table lists 4,356 AI references for investment and 1,764 for extortion. ENISA reports that AI further increases the tempo of fraud and scam activity because it allows better customization of deceitful content.

Theme 2: What do the reports say about voice cloning and deepfakes? (FBI and ENISA)

The FBI reports that voice cloning can mimic a loved one in distress scams, with victims claiming losses over $5 million in 2025. It also describes voice spoofing, or potentially voice deepfakes, in online job interviews. In those cases the actions and lip movement of the person on camera do not completely coordinate with the audio. The FBI says the goal generally appears to be gaining access to private computer networks, and reports almost $13 million in losses to AI-involved employment scams.

ENISA reports an increase of generative AI and deepfake videos used to impersonate trusted contacts in social engineering. It also describes the group Famous Chollima using fictitious profiles with generative AI-created text and fake profile images. A check derived from the FBI’s description: can your hiring team notice when audio and video do not line up? Read more on generative AI in our glossary.

Theme 3: How do attackers use AI in malware and intrusions? (ENISA, FBI)

ENISA reports that attackers primarily use consumer-grade AI tools to augment existing skills rather than to achieve breakthrough capabilities. Reported uses include LLMs that craft tailored phishing messages at scale, reconnaissance, script development and post-exploitation. ENISA also reports likely AI-generated malware in real-world use and “turnkey” AI-assisted toolkits resold on underground marketplaces. ENISA also reports that operators of a ransomware group used AI coding assistants to develop the group’s Ransomware-As-A-Service (RaaS) panel. One AI-assisted cloud intrusion reportedly achieved administrative access within 8 minutes.

ENISA cites a report that a state-nexus group used an agentic AI system in a campaign, with a limited number of successful compromises. See our posts on agentic AI in offensive attacks and agentic ransomware, and the glossary entry on AI agents. For comparison, the FBI’s table lists 42 AI references for malware and 16 for ransomware.

Theme 4: Are AI systems themselves targets? (ENISA)

ENISA says the integration of AI systems into enterprise environments creates a new attack surface. Threat groups show consistent interest in AI systems both as tools and as targets. AI applications are increasingly targets, particularly where they have access to files, credentials, browser sessions or development environments. ENISA adds that AI services may also provide a path into trusted environments, and that a fake website for an AI service was used to distribute malware.

A question derived from these facts: which AI tools in your organization can reach files, credentials or browser sessions, including ones nobody approved? See shadow AI. ENISA also notes the European Commission published an EU Action Plan on Cybersecurity and Artificial Intelligence.

Theme 5: How does AI appear in information manipulation? (ENISA)

Drawing on the EEAS, ENISA reports that 27% of incidents detected in 2025 involved AI-related techniques, tactics and procedures. That is up from 41 to 147 cases, a growth of approximately 259%. ENISA says Russian and Chinese actors have fully embedded AI tools into their operations. It also says much of the AI-generated material remains low-quality, with limited overall impact, and that one Russian infrastructure is suspected of “LLM grooming”.

Theme 6: What does ENISA expect next?

ENISA states its outlook in careful probability terms. The wording below is quoted as written.

Elsewhere ENISA says AI will “highly likely continue to primarily act as a force multiplier”. It also says that AI directly enabling more phases would be “likely to hinder detection and imputation for networks defenders”.

Two linked pieces cover this from other angles: the OWASP LLM Top 10 and hacktivism and DDoS trends.