← All concepts

prompt injection via capability drift

1 articles · 4 co-occurring · 0 contradictions · 0 briefs

The model's stated capability constraints appear to be overridden or forgotten by later turns, creating an effective prompt injection where the model ignores its own stated limitations.

The model's stated capability constraints appear to be overridden or forgotten by later turns, creating an effective prompt injection where the model ignores its own stated limitations.

query this concept
$ db.articles("prompt-injection-via-capability-drift")
$ db.cooccurrence("prompt-injection-via-capability-drift")
$ db.contradictions("prompt-injection-via-capability-drift")