AI Cybersecurity Reinvented: 13 Essential Principles for Future Safety

AI Cybersecurity Reinvented: 13 Essential Principles for Future Safety

• Why AI security needs its own rules: AI models and the systems around them bring risks that older security frameworks were not written for.
• Standards Initiative: ETSI TS 104 223, published in April 2025, sets baseline security requirements for AI models and systems.
• 13 Key Principles: The standard groups 13 principles across the AI lifecycle, from design to disposal.
• Applications in Industry: Banks and hospitals are among the users who stand to gain the most.
• Future Trends and Challenges: What comes next for AI security and where the hard problems are.

A baseline for AI security

AI now runs inside fraud engines, help desks and hospital scheduling systems. Securing it is a different job from securing ordinary software. The European Telecommunications Standards Institute (ETSI) addressed that in April 2025 with TS 104 223, “Securing Artificial Intelligence (SAI); Baseline Cyber Security Requirements for AI Models and Systems.” Here is what the standard covers and how its 13 principles break down.

How the standard is built

Understanding ETSI TS 104 223

ETSI TS 104 223 sets out baseline requirements for protecting AI systems from attack, grouped by stage of the AI lifecycle: secure design, secure development, secure deployment, secure maintenance and secure end of life. ETSI positions it as an international benchmark that developers, vendors and operators can all work from. The aim is the familiar one for any security standard: keep systems and their data intact, confidential and available.

Why a shared standard helps

AI security frameworks address real attacks, from poisoned training data to prompt injection. Misused AI can cost companies money and expose private data. A common standard gives developers and the companies deploying AI one set of practices to check against, which makes it easier to judge whether a given system can be trusted.

The 13 Essential Principles for AI Cybersecurity

The 13 principles run from raising awareness of AI threats at the design stage to disposing of data and models safely at the end. Five of them give a sense of the range:

• Evaluate threats and manage risks: Assess the risks to an AI system before and during development.
• Secure the supply chain: Check the models, data and components that come from outside suppliers.
• Document data, models and prompts: Keep clear records so problems can be traced and fixed.
• Enable human responsibility: Make sure people stay accountable for how the system behaves.
• Conduct appropriate testing and evaluation: Test AI systems for security before release and after changes.

Implementing Principles in Industry

Banks lean on AI for fraud detection and risk scoring, so they have strong reasons to lock down the models making those calls. Hospitals are in a similar spot: they rely on AI tools while having to protect patient records and keep clinical systems running, and a published baseline gives them something concrete to hold vendors to when contracts come up for renewal.

Future Trends and Challenges

Emerging Trends

AI security keeps changing as the technology does. One trend is AI-driven defense, tools that spot and stop threats in real time. Another is shared threat intelligence about attacks on AI systems.

Potential Challenges

The hard problems remain hard. AI threats change quickly, and using AI in security roles raises ethical questions about oversight and error. Developers and policymakers will keep wrestling with how to patch AI systems against new weaknesses without cutting corners on those standards.

Where to start

AI is now part of the infrastructure in many industries, and it needs security controls built for it. ETSI’s 13 principles give teams a practical starting point. Map your AI projects against the five lifecycle stages, find the gaps, and expect to revisit the work as both the attacks and the standard change.

John King, CISSP, PMP, CISM

John King currently works in the greater Los Angeles area as a ISSO (Information Systems Security Officer). John has a passion for learning and developing his cyber security skills through education, hands on work, and studying for IT certifications.