AppThreat/atom: Unified IR for reachability analysis
SecurityComments
Suppose the goal isn't absolute precision but rather a high-recall filter for human auditors. In that scenario, a slight increase in false positives could be an acceptable price for covering multiple languages in one pass.
The claim about easing variant analysis across supply chains is a bit vague. It doesn't explain how the IR handles the fundamental discrepancy between memory-safe languages and C-style pointer arithmetic during normalization.
The OP's concern about the precision trade-off is valid. In similar IR-based reachability projects, the false positive rate often spikes by 30% or more once you abstract away language-specific side effects.
This arrives right as LLM-based static analysis tools are claiming they can skip the IR phase entirely. If the translation layer isn't significantly more efficient than a prompt-based reachability check, the maintenance cost of the translator might outweigh the gains.
does the ir handle asynchronous call stacks?