Attackers systematically exploit legitimate AI tools and popular developer infrastructure as attack vectors while deliberately minimizing traditional security signals.
15 compromised JetBrains plugins masquerade as AI assistants and steal plaintext API keys over unencrypted HTTP connections to IP address 39.107.60.51.
At least 15 malicious plugins in the JetBrains Marketplace were designed to steal AI API keys from developers and gain access to internal corporate services.
Miasma replicates autonomously across Git repositories and automatically deletes user data when its GitHub token is blocked, with the now-public source code expected to lead to further variants.
Legitimate AI agents inherently satisfy all three criteria of the “lethal trifecta” (data access, external content, external communication), so security must shift from architectural design to runtime monitoring.
Over 400 Arch Linux AUR packages were compromised with infostealer malware, posing a data exfiltration risk to all systems that installed these packages on or after June 11, 2026.
Publicly available supply-chain attack kits, commercialized RAT infrastructures, and empirically demonstrated phishing vulnerability of AI agents mark a professionalization of the threat landscape.
VerdantBamboo strategically exploits Linux appliances in under-protected network positions as an access bridge to compromise high-value targets and bypass network security mechanisms.
Attackers operate highly ranked fake pages for tools like Ghidra and dnSpy on Google, redirect users through TDS-controlled JavaScript to malware servers, and evade security analysis by filtering VPNs, data centers, and repeated access.
A developer deliberately placed sabotage code in jqwik 1.10.0 to manipulate AI agents into deleting code, revealing a new security vulnerability in the open-source software supply chain.
