VulnHub

Real-time Cybersecurity News

AI coding agent deletes production database in seconds

SCM feed for Latest
Critical

Overview

An AI coding agent named Cursor, powered by Anthropic's Claude Opus 4.6, accidentally deleted PocketOS's entire production database along with all volume-level backups in a single API call to the infrastructure provider Railway. This incident raises significant concerns about the reliability and oversight of AI systems used in critical operations. With the database wiped out, PocketOS may face severe disruptions, affecting their service delivery and data integrity. It also highlights the potential risks associated with integrating AI tools into production environments without adequate safeguards. Companies using AI for coding or infrastructure management need to ensure proper checks and balances are in place to prevent such catastrophic errors in the future.

Key Takeaways

  • Affected Systems: PocketOS, Railway
  • Action Required: Implement stricter oversight and validation processes for AI-generated commands; develop rollback procedures for database management.
  • Timeline: Newly disclosed

Original Article Summary

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.

Impact

PocketOS, Railway

Exploitation Status

No active exploitation has been reported at this time. However, organizations should still apply patches promptly as proof-of-concept code may exist.

Timeline

Newly disclosed

Remediation

Implement stricter oversight and validation processes for AI-generated commands; develop rollback procedures for database management.

Additional Information

This threat intelligence is aggregated from trusted cybersecurity sources. For the most up-to-date information, technical details, and official vendor guidance, please refer to the original article linked below.

Related Topics: This incident relates to Critical.

Read Original Article

Related Coverage

38 Vulnerabilities Found in OpenEMR Medical Software

SecurityWeek

A recent security assessment has identified 38 vulnerabilities in OpenEMR, a widely used medical software platform. Some of these vulnerabilities could allow attackers to access and modify sensitive patient information, raising significant concerns for healthcare providers that rely on this software to manage patient records. Given the critical nature of health data, these vulnerabilities pose a serious risk to patient privacy and safety. OpenEMR users, including medical practices and clinics, should take immediate action to secure their systems. The findings emphasize the need for regular security audits and timely updates to safeguard against potential breaches.

Apr 29, 2026

Cursor AI IDE vulnerability allows code execution via hidden Git hooks

Hackread – Cybersecurity News, Data Breaches, AI and More

Researchers at Novee have identified a serious vulnerability in Cursor AI, designated as CVE-2026-26268. This flaw could allow attackers to execute malicious code when developers clone repositories, potentially compromising their systems. The vulnerability is particularly concerning for those using Cursor AI in their development workflows, as it opens up a pathway for exploitation that could lead to data breaches or the introduction of harmful code. Developers and organizations using this integrated development environment should take immediate action to assess their systems for this vulnerability and understand the risks involved. Awareness and prompt remediation are crucial to maintaining security in software development processes.

Apr 29, 2026

Critical GitHub Vulnerability Exposed Millions of Repositories

SecurityWeek

A significant vulnerability, identified as CVE-2026-3854, has been discovered in GitHub.com and GitHub Enterprise Server, potentially allowing remote code execution. This flaw poses a risk to millions of repositories hosted on these platforms, which are widely used by developers and organizations for version control and collaboration. If exploited, attackers could execute arbitrary code, leading to unauthorized access and manipulation of sensitive codebases. The discovery emphasizes the need for users to remain vigilant and update their systems promptly to mitigate potential risks. GitHub has urged users to apply the latest patches to safeguard their repositories against this vulnerability.

Apr 29, 2026

LiteLLM CVE-2026-42208 SQL Injection Exploited within 36 Hours of Disclosure

The Hacker News

A serious SQL injection vulnerability, identified as CVE-2026-42208, has been discovered in BerriAI's LiteLLM Python package, with a high CVSS score of 9.3. Remarkably, this flaw has already been actively exploited within just 36 hours of its public disclosure. Attackers can use this vulnerability to modify the database underlying the application, posing significant risks to any systems using LiteLLM. Organizations that rely on this package need to act quickly to protect their data and systems from potential breaches. Users should remain vigilant and apply necessary updates or patches as soon as they are available to mitigate these risks.

Apr 29, 2026

The Exchange Online security controls organizations keep getting wrong

Help Net Security

In a recent interview, Scott Schnoll, a Microsoft MVP for Exchange, discussed common mistakes organizations make regarding security controls in Exchange Online. He emphasized the importance of understanding the Shared Responsibility Model, where Microsoft manages cloud security while organizations are responsible for their data and configurations. Schnoll pointed out that legacy protocols like SMTP AUTH often remain enabled due to dependencies on older systems, which can create vulnerabilities. He also identified critical controls that are frequently overlooked, such as Conditional Access and Privileged Identity Management (PIM), and noted the gaps in audit logs that can hinder effective monitoring. Organizations need to take immediate action to adjust default settings and implement better security practices to protect their environments.

Apr 29, 2026

Vidar infostealer evolves, uses image files for stealthy attacks

SCM feed for Latest

The Vidar infostealer has adapted its tactics to launch stealthy attacks by using social engineering techniques. Recent campaigns have taken advantage of a leak related to Claude Code by creating fake GitHub repositories that trick users into downloading malicious payloads disguised as legitimate image files. This approach allows attackers to bypass some traditional security measures, making it harder for users to detect the threat. Those who download the infected files could have their personal data stolen, including sensitive information and credentials. As this method becomes more prevalent, users must be cautious about the sources of their downloads and verify the authenticity of repositories before accessing them.

Apr 28, 2026