Inference-Engine Fingerprinting Attacks Are Practical: Exploring Model-Driven Environmental Discovery, Exploitation, and Escape

Frontier AI models are rapidly gaining the ability to exploit vulnerabilities in complex pieces of software. The risk is not theoretical, as evidenced by recent sandbox escapes performed by frontier models at OpenAI and Anthropic. Discussions of how to sandbox inference stack components often foc...

Incorrect Condition for Validating Input and Output Tokens in CurveTricryptoAdapter::_determineComputeType Function

Lines of code Vulnerability details Impact The contract contains a bug in the determineComputeType function, where the condition for checking if the input and output tokens are valid for swapping is incorrect. This bug may lead to unexpected behavior and incorrect computation of the compute type...

Potential Loss of Funds Due to Zero Slippage Hardcoding in TalosBaseStrategy#deposit

Lines of code Vulnerability details Impact In the deposit function within the TalosBaseStrategy contract, both slippage for two tokens amount0Min and amount1Min are hardcoded to zero. This can have severe implications as users may unintentionally accept a minimum of zero output tokens from a swap...