Fooling AI Agents: Web-Based Indirect Prompt Injection Observed in the Wild

2026-03-04T21:26:55Z3297f00aabe3a5472f8454b32bb79eeea8db480135800857cf397a4cf78ed236
CVE-2025-0921CVE-2026-0628CVE-2026-1281CVE-2026-1340CVE-2026-1731ai-securitybeyondtrust-cve-2026-1731chrome-cve-2026-0628cloud-detectioniconics-cve-2025-0921iranivanti-cve-2026-1281ivanti-cve-2026-1340muddled-libranation-statenotepad++-supply-chainot-securityphishingprompt-injectionqr-code-phishingsparkRATvshellweb-shells

What happened

Palo Alto Unit 42 published a collection of March 2026 posts detailing multiple high-impact threats and exploited vulnerabilities. Key findings include real-world web-based indirect prompt-injection against LLM agents, a high-severity Chrome Gemini extension hijack (CVE-2026-0628), active exploitation of BeyondTrust RCE (CVE-2026-1731) with VShell and SparkRAT, widespread exploitation of Ivanti EPMM zero-days (CVE-2026-1281, CVE-2026-1340) deploying web shells/backdoors, and a privileged-file-system SCADA DoS (CVE-2025-0921). Unit 42 also covers rising Iranian cyber activity, supply-chain (Not

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
paloalto_unit42
Record identifier
3297f00aabe3a5472f8454b32bb79eeea8db480135800857cf397a4cf78ed236
Enrichment time
2026-03-04T21:26:55Z
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.