The development of deep learning frameworks is a major step forward for AI and has many potential applications. However, security risks associated with deep learning systems are a key impediment to their widespread use. Any attempt by malicious insiders or outsiders to compromise deep learning frameworks will have far-reaching consequences for people's everyday lives. We get things off with a rundown of the deep learning algorithm structure and a careful analysis of its vulnerabilities and dangers. Here, we provide a comprehensive categorization technique for security worries and defensive methods in deep learning frameworks, and we establish connections between different types of threats and the countermeasures that may be taken against them. We also look at a real-world scenario where security flaws in deep learning were present. We conclude with a discussion of future directions and challenges for deep learning architectures. We hope that our efforts will pique the attention of the academic and business sectors in furthering the study of and developing solutions for the security challenges presented by deep learning frameworks.

adversarial examples; deep learning frameworks; defensive techniques; security concerns