Patch LLMMathChain exec vulnerability #1119
Closed
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
As referenced in other issues (see #814, #1026), the current implementation of
LLMMathChainallows for prompt injection attacks which can execute arbitrary code using Python'sexecmethod. As a quick patch, I have modified the chain to use the slightly saferevalmethod and modified the prompt template accordingly. From my quick testing, this doesn't seem to hurt the chain's mathematical ability and prevents the exploit outlined in #814.NB: This is intended as a quick patch to mitigate the current security risks. Future developments, as in #1055, should further help with solving this vulnerability and others.