Hi I'm currently implementing a symbolic execution engine for EVM in Rust. Though my current implementation is just a translation of hevm, I would like to add some novel features. For example, I'm considering using A* search or multi-armed bandit to enhance scalability. Any kind of feedback, ideas, and requests are super welcome!!!
AIJack is an easy-to-use open-source simulation tool for testing the security of your AI system against hijackers. It provides advanced security techniques like Differential Privacy, Homomorphic Encryption, K-anonymity and Federated Learning to guarantee protection for your AI. With AIJack, you can test and simulate defenses against various attacks such as Poisoning, Model Inversion, Backdoor, and Free-Rider. We support more than 30 state-of-the-art methods. For more information, check our documentation and start securing your AI today with AIJack.
Hello, Hacker News community! I am excited to introduce a new project called Gymbo, a Proof of Concept for a Gradient-based Symbolic Execution Engine. Gymbo is designed to push the boundaries of symbolic execution by leveraging recent advancements in gradient descent to tackle SMT-like formulas. This approach allows Gymbo to discover input values that satisfy each path constraint during symbolic execution.
Gymbo is entirely implemented in C++ and relies only on standard libraries.
What sets Gymbo apart from other symbolic execution tools is its simplicity and compactness in implementation. I believe that this project will help individuals better understand the core principles of symbolic execution and SMT problem-solving through gradient descent.
Things like:
prompt injection → goal hijacking
agents going rogue due to misalignment
unintended/unsafe tool use
It feels like we're starting to see repeatable patterns, not just isolated bugs.
I'm collecting cases + papers here:
https://github.com/h5i-dev/awesome-ai-agent-incidents
If you've seen interesting incidents, weird failures, or relevant research, I would love to add them.