Analysis
AI in the patent industry: The sneaky practice of prompt injection (and what to do about it)
Publication date unknown · Discovered ipkitten.blogspot.com
Page published
Publication date unknown · First observed: 13 Oct 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Prompt injection via documents fed to AI systems is a practical manipulation vector for domains like patent examination, where LLMs process adversarially-authored third-party text.
The IPKat blog explains prompt injection as a risk for the patent industry, where AI tools increasingly read, summarise and draw conclusions about third-party documents such as patent applications and prior art. The piece describes how hidden instructions in input documents can manipulate an LLM to reach a desired conclusion, leak information, or recommend a particular outcome, and notes the growing danger from connectors like MCP servers.