Navigating Security Tradeoffs of AI Agents

2026-03-19T08:51:44Z236b98d5d3ba95bf5f7667ba466a4c7f14ba2c955cd4a72375365b7fb6a098b2
AI securityBeyondTrustBoggy SerpensCL-UNK-1068CVE-2026-0628CVE-2026-1281CVE-2026-1340CVE-2026-1731China espionageChrome GeminiHandala HackIranian threat actorsIvanti EPMMLLM guardrailsOT securityQR-code phishingagentic browsersmobile phishingprompt fuzzingprompt injectionweb shellswiper attackszero-day exploitation

What happened

Unit 42 published a series of March 2026 posts covering two parallel risk trends: (1) rapidly evolving GenAI/LLM attack surface — prompt injection, prompt-fuzzing (genetic-algorithm-based evasion), vulnerable "AI judges" and agentic browsers — and (2) active, high-impact cyber campaigns and exploited vulnerabilities. Notable technical findings include Chrome Gemini panel hijack (CVE-2026-0628), a BeyondTrust RCE exploited in the wild (CVE-2026-1731), and widespread exploitation of Ivanti EPMM zero-days (CVE-2026-1281, CVE-2026-1340). Unit 42 also highlights Iranian actor activity (Handala Hack

Why it matters

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

Evidence and limitations

Source ID
paloalto_unit42
Record identifier
236b98d5d3ba95bf5f7667ba466a4c7f14ba2c955cd4a72375365b7fb6a098b2
Enrichment time
2026-03-19T08:51:44Z
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.

Record · Navigating Security Tradeoffs of AI Agents · Baitaphish