Data Security

An AI coding agent, Cursor running Anthropic's Claude Opus 4.6, deleted PocketOS's production database and all volume-level backups in a single API call to infrastructure provider Railway.

Agentic AI can't succeed without trust, and trust begins with mapping how data is accessed, used, and protected.

The attack exploited a GitHub Actions script injection flaw, allowing the attacker to inject shell code that exposed a GitHub token.

Cybernews reports that U.S-based. e-learning platform Udemy was claimed to have more than 1.4 million records compromised by the ShinyHunters extortion group, which threatened to expose the data trove should the firm opt out of negotiations by Apr. 27.

The investigation began in late December 2025 following approximately 100 reports of data theft.

BlackFile initiates attacks through voice phishing (vishing) calls, using spoofed numbers to impersonate IT support.

The breach affected customer and prospective customer data, including names, phone numbers, and addresses.

Managing the behavior of AI agents requires a data-centric, continuous, and adaptive new approach to security.