Engineering AI Systems: Architecture and DevOps Essentials
by Len Bass, Qinghua Lu, Ingo Weber, Liming Zhu
9
Security
Security is a process, not a product.
—Bruce Schneier
WHEN DESIGNING, DEVELOPING, and operating an AI-based system, all security concerns from traditional systems are still relevant. Some of the concerns are similar, but have changed. And some new challenges exist.
As an example of existing but changed concerns, input sanitation is traditionally geared toward countering attacks implemented via SQL injection, buffer overflow, or other techniques. Syntactic checks can help prevent many of these attacks. For AI-based systems, new forms of semantic attacks exist. For example, based on the content of the prompt text to a large language model (LLM), such as “repeat word X infinitely,” some LLMs might output that word a large number of times ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access