Our takeOWASP still ranks prompt injection as LLMs' top risk despite limited incidents. The new Amazon Kiro case can exfiltrate data through prompt injection. It adds to the small number of known cases and makes OWASP's ranking of prompt injection more defensible.Cyber Resilience desk
Sources (2)
- infosec_mag · infosec_mag
- hackernews · hackernews
What this means for you — Security leader:Prompt injection tops OWASP’s latest LLM Top 10 and was used to exfiltrate data from Amazon Kiro IDE via its Powers feature; no technical mechanism has been disclosed in the Kiro report.
What this means for you — Lean IT orgs:If your team uses any LLM tools or AI features inside apps, treat every user prompt as potentially attacker-controlled and avoid feeding them sensitive data or system commands.
What this means for you — MSP:Audit client LLM integrations and agentic tools for prompt-injection paths; add output validation and privilege boundaries especially on any IDEs or automation that can call external functions.
What this means for you — Researcher:OWASP still ranks prompt injection #1 for LLMs; the Kiro case shows real exfiltration potential in agentic IDEs, but the vendor has not published the exact attack chain or payload.