Tag: AI security

Researchers have identified security vulnerabilities in Amazon Bedrock, LangSmith, and SGLang that could enable data exfiltration, account takeover, and remote code execution in AI environments.

China’s CNCERT warns that OpenClaw AI agent security weaknesses could enable prompt injection attacks, endpoint compromise, and sensitive data exfiltration.

As enterprises deploy AI agents that read and act on information from internal systems and the internet, prompt injection is emerging as a new cybersecurity risk that can manipulate machine reasoning, expose sensitive data, and influence automated workflows.

A compromised npm publish token was used to release Cline CLI version 2.3.0, triggering unauthorized installation of OpenClaw on developer systems during an eight hour supply chain attack window.

CrowdStrike plans to acquire Seraphic, expanding its Falcon platform into browser runtime security to provide real-time protection across devices, SaaS, and AI interactions.

Microsoft has released an AI Security Risk Assessment white paper to help organizations evaluate AI-related risks using existing frameworks as a starting point for secure AI adoption.