Threat Research

In March 2026, Anthropic accidentally exposed the full source code of its Claude Code AI agent through a misconfigured npm package that included a large JavaScript source map file. The leak revealed hundreds of thousands of lines of unobfuscated code, exposing internal architecture, agent orchestration logic, and security-related components....

Between late February and March 2026, TeamPCP launched a calculated series of escalating supply chain attacks. They compromised trusted open-source security tools like Trivy, KICS, and the AI gateway LiteLLM. The campaign also targeted the official Python SDK of Telnyx. Malicious infostealer payloads were injected into GitHub Actions and PyPI registries....

Researchers uncovered and analyzed the full source code of an AI-driven AiTM phishing platform called “UPMI ULTIMATE,” linked to a group named “Team Unlimited.” The code was retrieved from an exposed central server that manages licensing, intelligence sharing, and remote control for all client instances....

EvilTokens is a newly identified phishing-as-a-service (PhaaS) kit that enables large-scale Microsoft device code phishing attacks, leveraging social engineering techniques and rapidly adopted by cybercriminals for Adversary-in-the-Middle (AitM) and Business Email Compromise (BEC) operations....

A cyberespionage campaign discovered in early 2026 involved three distinct threat clusters Paper Werewolf, Versatile Werewolf, and Eagle Werewolf targeting victims using malware disguised as Starlink registration services and drone training applications....

A growing share of cyber incidents now stems from supply chain attacks. Attackers use tactics like malicious open-source libraries or hijacked developer accounts. These compromised libraries spread widely, affecting countless applications and services. In March 2026, a trojanized LiteLLM Python library was uploaded to PyPI, infecting systems....

A supply chain attack compromised the LiteLLM AI proxy package on PyPI, with malicious versions delivering a multi-stage payload that harvested credentials, enabled Kubernetes lateral movement, and established persistent backdoor access for remote code execution....

A large-scale malware campaign leveraged AI-driven “vibe coding” to generate malicious code components, lowering the barrier for threat actors to create and distribute malware. The campaign used hundreds of malicious ZIP files impersonating popular software—such as AI tools, game mods, and utilities—to deliver multiple variants of the WinUpdateHelper.dll payload....

A financially motivated threat group tracked as Hive0163 has been observed using a likely AI-generated malware called Slopoly during ransomware attacks, marking an early example of AI-driven malware development in real-world operations....

Boggy Serpens (also known as MuddyWater), an Iranian state-linked threat group associated with MOIS, continues to conduct cyberespionage campaigns targeting diplomatic entities and critical infrastructure sectors such as energy, maritime, and finance....