Prompt injection manipulates LLM outputs through direct or indirect methods, requiring mitigations like restricting model capabilities and applying guardrails.
Malicious or biased training data can poison a model's worldview, necessitating careful data screening and keeping models up-to-date.
LLM outputs can exploit downstream systems like databases or frontends, so they must be treated as untrusted user input and sanitized accordingly.
Sensitive data used for training can be extracted from models, highlighting the need to redact or anonymize information before it reaches the LLM.
Just like in traditional application security, achieving 99% security is still a failing grade because attackers will find and exploit any existing vulnerability.
msg
Ismaning, Germany
27:32Sebastian Schrittwieser
27:10Georg Dresler
29:00Alex Soto
30:36Mackenzie Jackson
26:30Simon A.T. Jiménez
35:37Liran Tal
27:19Nura Kawa
27:11Emanuele Fabbiani
Jobs that call for the skills explored in this talk.
Lumen Technologies, Inc
CommandLink
Collaboration Betters The World GmbH
Palo Alto Networks
Ai-powered
Motion Recruitment Partners LLC.
NTT DATA, Inc.
Jersey City, United States of America
