Guarding AI memory
2026-06-23T14:52:19Z•24d6d521c2d2e1621c89e22fc15381fa4bbb730fe8d5af3ae5b565f76f5c18e5
agentic-vulnerability-detectionai-memoryai-securityautogen-studioautojackbackdoorclipboard-theftcrypto-clipperedremail-security-benchmarkinglocalhost-trustmastramcp-websocketmdashmicrosoft-defendernpmparallel-threat-actorspostinstall-payloadransomwareremote-code-executionsapphire-sleetsupply-chaintor-c2worm-like-propagationxdr
What happened
Microsoft Security Blog roundup (June 15–22, 2026): multiple high-risk findings across AI agents, supply chain, malware campaigns, and enterprise telemetry. Notable items: AutoJack — a novel exploit chain where a single malicious webpage can achieve host remote code execution by abusing localhost trust, missing authentication, and unsafe parameter handling in AutoGen Studio’s MCP WebSocket; Mastra npm supply‑chain compromise (Sapphire Sleet) that poisoned a package and impacted 140+ projects via postinstall payloads; a crypto clipper campaign combining clipboard theft, wallet replacement, Tor‑
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- microsoft_security_blog
- Record identifier
- 24d6d521c2d2e1621c89e22fc15381fa4bbb730fe8d5af3ae5b565f76f5c18e5
- Enrichment time
- 2026-06-23T14:52:19Z
- 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.