Collected molecules will appear here. Add from search or explore.
A benchmarking toolkit and library for implementing and evaluating adversarial attacks against deep neural networks.
Defensibility
stars
7
forks
22
GTAttackPod is an academic repository from the Georgia Tech Distributed Data Intensive Systems Lab (DISL). Despite its pedigree, the project is essentially dormant with a velocity of 0.0/hr and an age exceeding six years. With only 7 stars, it lacks any significant community or commercial traction. Its primary value was as a collection of 'state-of-the-art' attacks at the time of its release (likely focusing on CNNs and classical adversarial perturbations like FGSM or PGD). Today, it has been entirely superseded by industry-standard libraries such as IBM's Adversarial Robustness Toolbox (ART), CleverHans, and Advertorch, which offer much broader support for modern architectures (Transformers, LLMs) and deep integration with current ML frameworks. Frontier labs like OpenAI and Anthropic have moved beyond these pixel-level attacks to focus on alignment, jailbreaking, and multimodal safety, rendering this specific toolkit a relic of early adversarial ML research. The high fork-to-star ratio (22 forks to 7 stars) is characteristic of academic projects used internally by students for specific course assignments rather than broad adoption.
TECH STACK
INTEGRATION
library_import
READINESS