TL;DR

Flaws in Amazon Bedrock, LangSmith, and SGLang allow attackers to exfiltrate data using DNS queries and gain RCE by enabling interactive shells within the sandbox environment.

What happened

['AI flaws enable data exfiltration via DNS', 'Interactive shell enabled through sandbox mode']

Why it matters for ops

['DNS-based data exfiltration possible', 'Risks of compromised AI environments']

Mitigation

  • Implement strict inbound and outbound firewall rules
  • Monitor and restrict DNS queries from AI environments

Action items

  • Review security configurations of AI services
  • Apply patches for affected systems

Detection IOCs

  • Unusual DNS requests
  • Unexpected outbound network traffic

Source link

https://thehackernews.com/2026/03/ai-flaws-in-amazon-bedrock-langsmith.html