Cybercriminals are exploiting the widespread adoption of artificial intelligence tools to infiltrate corporate networks, compromise sensitive records, and seize control of critical enterprise accounts. According to an examination of 1,258 public sandbox submissions recorded between April 15 and October 5, 2026, threat actors are deploying fraudulent download portals, malicious installers, and spoofed authentication prompts that impersonate established AI systems, including Claude, ChatGPT, GitHub Copilot, and Gemini. While the figures reflect recorded analyses rather than verified victims, geographical tracking across seven major campaigns showed 422 country-based records. India registered the highest volume with 86 submissions, followed closely by the United States with 69 records, representing roughly 16 percent of the total.
Impersonation Campaigns Exploit Trusted Cloud Infrastructure
Activity peaked in May 2026, when researchers logged 358 separate analyses, before settling into a steady volume ranging between 108 and 151 monthly incidents through September. Counterfeit variants of Claude and OpenAI offerings accounted for roughly 81 percent of all investigated cases, with 601 records tied to Claude and 418 linked to ChatGPT and Codex. Attackers have expanded their delivery mechanisms beyond conventional phishing emails and malicious attachments. Operators are now leveraging sponsored search engine listings, deceptive search results, and public AI conversations, while hosting fraudulent landing pages on trusted platforms such as GitHub, GitLab Pages, Cloudflare Pages, and Google Sites to bypass enterprise web filters.
Terminal Commands and Hidden Browsers Facilitate Data Theft
A significant portion of the campaigns relies on social engineering to bypass browser protections entirely. Approximately 45 percent of the investigated cases involved instructions directing victims to manually copy and execute commands in PowerShell or macOS Terminal, split between 341 fake command-line setups and 220 rogue download sites. In multiple instances, spoofed Claude portals instructed Windows users to paste malicious commands directly into the system Run dialog, using native administrative utilities to download payloads without dropping an installer in browser folders. Other operations bundled legitimate software with hidden malware that scheduled background instances of Microsoft Edge to exfiltrate cookies and extension data, granting attackers access to active cloud sessions without triggering multi-factor authentication checks.
Targeted Attacks on Corporate Accounts Demand Stricter Controls
The risks extend across technical and commercial operations, with marketing departments facing targeted lures such as fraudulent portals promising direct ad management through Claude, Gemini, Perplexity, or Manus. Interacting with these platforms surfaces spoofed Google sign-in prompts designed to capture administrative access, exposing corporate budgets and customer management portals. Developer workstations remain equally vulnerable to the loss of source code, cloud keys, and API tokens. Security specialists advise enterprises to restrict unauthorized script execution, maintain vetted catalogues of sanctioned AI tools, and train staff never to run terminal commands copied from search ads or shared chats, while monitoring endpoints for irregular browser tasks and anomalous network commands.
Follow for daily updates on cybercrime, corporate fraud, DFIR, hacking, investigations, and digital forensics
