Towards Trustworthy AI: A Review of Ethical and Robust Large Language Models

Multiple trust challenges require simultaneous attention: The paper identifies five key trust issues in LLMs—unintended harms, lack of transparency, vulnerability to attacks, misalignment with human values, and environmental impact—suggesting that trustworthy AI cannot be achieved by addressing just one concern in isolation.

Societal biases and opaque decision-making undermine user trust: The review highlights that LLMs can perpetuate societal biases and operate as "black boxes," making it difficult for users to understand how these systems reach their conclusions, which is particularly problematic in sensitive sectors like healthcare, finance, and education.

A multi-stakeholder approach is essential: Rather than relying solely on technology companies, the paper recommends combining ethical oversight, industry accountability, regulation, and public involvement—requiring collaboration across technology, ethics, law, and policy fields to create meaningful safeguards.

Ethics must be integrated throughout development, not added afterward: The review emphasizes that responsible AI requires reshaping development norms and aligning incentives to embed ethical considerations from the beginning of the machine learning process, rather than treating ethics as an afterthought.

Rapid deployment in high-stakes domains outpaces oversight capabilities: As LLMs are increasingly used in sensitive areas like healthcare and policy, the paper warns that the speed of technological advancement is creating significant challenges for adequate oversight and governance.