Updating the taxonomy of failure modes in agentic AI systems: What a year of red teaming taught us

2026-06-04T20:52:22Z302352d2d8c3165777fb10e5f95b09b9c672fada969532ffe80a632d53dececb
CI/CDConfluenceF5 BIG‑IPGentlemenKerberos relayMiasmaScreenConnectagentic AIcredential theftcryptojackingdependency confusiondetection and mitigationdeveloper toolinggoal hijackingnpmransomwarered teamingself‑propagationsupply chaintyposquatting

What happened

Microsoft Security Blog posts (late May–early June 2026) describe multiple high‑impact threats and defensive updates: an updated taxonomy of failure modes in agentic AI systems based on a year of red teaming (introducing seven new modes such as supply‑chain compromise and goal hijacking, with practical mitigations); large npm supply‑chain campaigns (the Red Hat “Miasma” compromise affecting 90+ @redhat‑cloud‑services versions, credential‑stealing and worm‑like republishing); dependency‑confusion and typosquatting npm campaigns that profile developers and steal cloud/CI/CD secrets; analysis of

Why it matters

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

Evidence and limitations

Source ID
microsoft_security_blog
Record identifier
302352d2d8c3165777fb10e5f95b09b9c672fada969532ffe80a632d53dececb
Enrichment time
2026-06-04T20:52:22Z
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 · Updating the taxonomy of failure modes in agentic AI systems: What a year of red teaming taught us · Baitaphish