Recent Insights

Start Your AI or Digital Project with MultiViews

Contact our Melbourne team for AI transformation, website development, API integration, or creative design support.

Anthropic Cursor Extension Backdoored for Nearly 12 Hours

Anthropic Cursor Extension Backdoored for Nearly 12 Hours

Security researchers reported that a verified Anthropic VS Code-style extension distributed through Cursor was backdoored for nearly twelve hours. Attackers somehow obtained access to Anthropic’s verified publisher account on the platform; the exact method remains undisclosed. During that window, users who installed or updated the extension could have received malicious code mixed into an otherwise trusted AI coding assistant.

Why this matters for Melbourne product and engineering teams

Melbourne’s growing cohort of SaaS, fintech and digital agencies relies heavily on AI-assisted editors such as Cursor. Many local squads treat marketplace extensions as low-friction productivity tools rather than third-party code running with full workspace privileges. A twelve-hour window is long enough for automated CI machines, overnight builds and developers working across Australian time zones to pull a tainted update. Once inside an IDE, malware can exfiltrate source, tokens, .env files and cloud credentials—assets that Australian privacy and APRA-regulated firms cannot afford to lose.

Practical steps for Australian organisations include locking extension auto-updates behind change control, pinning known-good versions in developer golden images, and routing all IDE traffic through monitored egress proxies. Teams should also maintain an internal allow-list of publishers and require cryptographic verification where the marketplace supports it. MultiViews Australia recommends treating every AI coding plugin with the same supply-chain scrutiny applied to npm or PyPI packages used in production pipelines.

Until marketplace operators harden publisher-account recovery and add stronger anomaly detection, the burden sits with engineering leads. A short compromise of a trusted AI vendor is a reminder that “verified” badges reduce friction for attackers as much as they do for legitimate users. Local firms that document extension inventories and rehearse credential-rotation playbooks will recover faster when the next similar incident appears.