AI Threat Landscape Digest March-April 2026

MITRE Techniques

• [T1059 ] Command and Scripting Interpreter – Used as AI-generated command execution during exploitation and post-exploitation workflows (‘5,317 AI-executed commands’ and ‘shadow file extraction and timestamp cleanup’).
• [T1105 ] Ingress Tool Transfer – The attacker built tunnel chains and moved data through attacker-controlled infrastructure during the Mexico breach (‘build tunnel chains’).
• [T1068 ] Exploitation for Privilege Escalation – Claude Code helped escalate privileges in the Mexico campaign (‘escalate privileges’ and ‘after gaining root’).
• [T1087 ] Account Discovery – EvilTokens extracted account-related data from stolen emails, including account numbers and routing numbers (‘extracting account numbers, routing numbers’).
• [T1114 ] Email Collection – EvilTokens processed large batches of stolen emails for intelligence extraction (‘ingests up to 5,000 emails’).
• [T1589 ] Gather Victim Identity Information – The platform synthesized victim communication style and business context for BEC (‘imitate sender style’ and ‘reference real email threads’).
• [T1566 ] Phishing – EvilTokens used device-code phishing pages impersonating Adobe, DocuSign, and SharePoint to harvest Microsoft OAuth tokens (‘Device-code phishing pages impersonating Adobe, DocuSign, and SharePoint’).
• [T1657 ] Financial Theft – EvilTokens generated BEC drafts and payment-change deception to support wire fraud (‘mask payment changes behind “plausible business reasons”’).
• [T1027 ] Obfuscated Files or Information – The campaign used header fingerprint randomization, DKIM signing, and CSS randomization to disguise delivery (‘header fingerprint randomization, DKIM signing, and CSS randomization’).
• [T1090 ] Proxy – The report describes privacy-routed proxies used to avoid provider monitoring (‘privacy-routed proxies’).
• [T1552 ] Unsecured Credentials – API keys were harvested from compromised .env files and malicious proxy redirection could steal authorization headers (‘API keys’ and ‘potentially steals API keys’).
• [T1204 ] User Execution – Malicious repository and settings files were used to trigger actions when developers opened projects (‘a malicious settings file embedded in a pull request’).
• [T1195 ] Supply Chain Compromise – Malicious agentic configuration files in repositories and compromised codebases were used to compromise developer machines (’embedded in a pull request, honeypot repository, or compromised codebase’).
• [T1056 ] Input Capture – The malicious proxy intercepted authorization headers in transit (‘intercepts authorization headers’).
• [T1065 ] Uncommonly Used Port – The report references proxying and service redirection behavior through custom endpoints (‘redirects ANTHROPIC_BASE_URL to a malicious proxy’).