Microsoft researchers have unveiled Lazarus, a technique capable of embedding hidden instructions into AI models by modifying fewer than 10 internal parameters without full retraining.During tests on models like Nemotron-Terminal-8B, Qwen, and Gemma, specific trigger phrases reliably activated the hidden behavior while maintaining most baseline performance.The method requires access prior to model deploy…https://en.hacks.gr/ereynites-tis-microsoft-paroysiasan-methodo-krypsis-entolon-se-montela-technitis-noimosynis/#Microsoft #ArtificialIntelligence #MachineLearning #Research
Microsoft researchers have unveiled Lazarus, a technique capable of embedding hidden instructions into AI models by modi...
