AI Flaws in Amazon Bedrock, LangSmith, and SGLang Enable Data Exfiltration and RCE
2026-03-17T19:24:17Z•04e9efbe8d67607f408da71a037e1be02e19186b31bb0e300f0aebf3547e0ad2
ai-securityapparmorclickfixdenodns-exfiltrationgithub-token-theftglasswormlaw-enforcement-takedownmacsyncprivilege-escalationransomwareremote-code-executionsandbox-escapesupply-chainzero-day
What happened
This digest highlights multiple high-impact threats and active exploits across AI platforms, enterprise backup, browsers, and supply chains. Researchers disclosed DNS-based data exfiltration and sandbox escape techniques against AI code-execution environments (Amazon Bedrock AgentCore, LangSmith, SGLang) that can enable interactive shells and RCE. Ransomware group LeakNet and several ClickFix social-engineering campaigns (including MacSync macOS infostealer) are leveraging user-driven command execution and in-memory Deno loaders. Supply-chain campaigns (GlassWorm) abuse stolen GitHub tokens, 3
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- the_hacker_news
- Record identifier
- 04e9efbe8d67607f408da71a037e1be02e19186b31bb0e300f0aebf3547e0ad2
- Enrichment time
- 2026-03-17T19:24:17Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.