All publications
Every advisory and briefing published by the AI Threat Lab, newest first.
- October 2026 | Advisory | Security for AI | IIAS-ATL-2026-001
Indirect prompt injection through retrieved content
How instructions embedded in documents, email and web content can redirect LLM-based assistants and agents, with layered mitigations across input handling, tool permissions and output validation.
- October 2026 | Advisory | Security from AI | IIAS-ATL-2026-002
Voice-clone attacks on payment-approval workflows
Attack patterns against verbal approval processes, and the out-of-band verification controls that break them.
- October 2026 | Advisory | Security with AI | IIAS-ATL-2026-003
Human oversight controls for LLM-assisted alert triage
Designing review thresholds, override logging and performance monitoring when LLMs participate in SOC decisions.
- October 2026 | Advisory | Security for AI | IIAS-ATL-2026-004
Provenance verification for open-weight models
Risks in downloading and deploying third-party model weights, and the integrity checks to apply before production use.
- October 2026 | Briefing | Security from AI | IIAS-ATL-2026-005
Synthetic identities in remote onboarding
Use of AI-generated documents and faces against remote identity proofing, and layered verification measures.
Reporting a vulnerability? Read the coordinated disclosure policy.