The rapid integration of Artificial Intelligence (AI) into virtually all sectors of society presents a paradox: immense opportunities alongside ethical and security challenges of increasing complexity. This dynamic underscores a critical gap between the speed of technological innovation and governance capacity, making the discussion about developing trustworthy AI systems imperative. Safety and ethics, far from being secondary considerations, are inseparable foundations that must be integrated from the conception of AI systems to sustain trust in the technology.

AI Ethics is defined as a set of principles and values guiding the development and use of technology to reinforce responsibility, justice, security, and transparency, aiming to benefit society and mitigate unintentional harm. Complementarily, Responsible AI translates these ethical principles into concrete guidelines and practices, focusing on building trust and equitably distributing benefits. AI Security, in turn, protects AI systems from malicious attacks and ensures the accuracy and reliability of their decisions.

One of the most pressing ethical challenges is algorithmic bias, which is not inherent to the algorithm itself, but a direct consequence of training data and model design. If historical data reflects existing societal prejudices, the AI model can amplify them or perform poorly for underrepresented groups. Real-world examples include Amazon's recruitment tool that systematically discriminated against women and the COMPAS algorithm, which classified Black defendants differently from White defendants, illustrating how algorithmic bias perpetuates gender and racial prejudices. The opacity of these models, often referred to as "black boxes," makes identifying and correcting these biases a significant challenge.

In the cybersecurity landscape, AI systems are targets for adversarial attacks. These include poisoning attacks, where incorrectly labeled data is injected during the training phase to corrupt the model, as seen with Microsoft's Tay chatbot. There are also evasion attacks, which occur after the model has been trained, seeking incorrect classifications of new inputs, such as deceiving a spam filter with specific word combinations.

Beyond being targets, AI systems can also be tools for malicious threats. Deepfakes enable convincing manipulation of videos and audios for financial fraud, such as the US$ 25.6 million case in Hong Kong. AI also facilitates the creation of more advanced automated malware and phishing, which can evade traditional security systems. Model stealing, where proprietary AI models are replicated, exemplified by the suspicion involving the DeepSeek startup and OpenAI, demonstrates the complexity of technological espionage. The interconnection between security failures and governance is illustrated by the Therac-25 tragedy, where a software bug combined with human negligence resulted in fatalities, emphasizing that ultimate responsibility lies with the humans who design and oversee AI.

To mitigate these risks, AI governance frameworks are crucial. The Databricks AI Governance Framework (DAGF) proposes five pillars: AI Organization, Legal and Regulatory Compliance, Ethics, Transparency, and Interpretability, Data, AIOps, and Infrastructure, and AI Security. In parallel, global regulation advances with initiatives like the European Union's AI Act, which classifies AI systems by risk level—unacceptable, high, limited, and minimal/none—imposing distinct requirements for each category.

Technically, Explainable AI (XAI) is a fundamental discipline to make "black box" models understandable, increasing trust by allowing users to comprehend their conclusions. Complementarily, algorithmic audits emerge as a vital evaluation process to ensure algorithms operate fairly, ethically, and transparently, identifying biases and ensuring compliance. These solutions, both structural and technical, require a profound cultural and organizational shift.

In conclusion, AI possesses immense potential for the good of humanity, but its uncontrolled advancement can exacerbate inequalities and create systemic vulnerabilities. It is imperative that companies, governments, and society adopt a proactive governance approach, investing in "Ethics and Security by Design" from the initial development phases. Collaboration and education are essential to maintain a delicate balance between innovation and regulation, ensuring that the AI revolution leads to a just, secure, and equitable future for all.