Technical

Secure AI

Protection against malicious manipulation and unauthorized access.

How the laws define this

Official international frameworks establish that secure AI requires protecting systems against malicious misuse, cyber threats, unauthorized manipulation, and adversarial attacks to maintain data integrity and system resilience. Cybersecurity controls and access protections form the shared foundation across these instruments. However, the texts diverge regarding the explicit scope and specific targets of those protections. The Belgrade Ministerial Declaration on Artificial Intelligence uniquely extends its definition beyond AI systems to cover their supporting infrastructure, while explicitly demanding measures to protect confidentiality, availability, and system reliability to preserve trust. In contrast, the UN General Assembly Resolution on Artificial Intelligence (A/78/L.49) explicitly highlights model and data protection alongside safeguards against exploitation. Meanwhile, the Declaration: Towards the Safe, Secure, and Trustworthy Development and Deployment of Artificial Intelligence in the Americas: The Importance of Governance, Regulatory, and Institutional Frameworks frames the concept around protecting against malicious actors without detailing specific infrastructure or access control mechanisms.

Synthesised from the 3 statutory definitions below. Each one is quoted in full, with its source.

How jurisdictions differ

What is distinctive in each definition, binding law in force first. Showing 2 of the 3 instruments that define this term; all are listed under Definitions below.

Jurisdiction	Instrument	Status	What is distinctive
OAS	Declaration: Towards the Safe, Secure, and Trustworthy Development and Deployment of Artificial Intelligence in the Americas: The Importance of Governance, Regulatory, and Institutional Frameworks	In Force	Emphasizes protection against malicious actors, cyber threats, and misuse, with a focus on data integrity and system resilience, in a regional framework balancing innovation and human rights.
Serbia	Belgrade Ministerial Declaration on Artificial Intelligence	Adopted	Explicitly covers supporting infrastructure, with detailed measures like cybersecurity practices, adversarial robustness, and access controls for 44 GPAI states and the EU.

What it triggers

The OAS Declaration sets a regional framework for safe, secure, and trustworthy AI development in the Americas, balancing innovation with human rights.source
The Belgrade Ministerial Declaration guides 44 Global Partnership on Artificial Intelligence member states and the EU toward safe and human-centric AI development.source

Definitions (3)

AI systems and their supporting infrastructure that incorporate measures to protect integrity, confidentiality, and availability from adversarial attacks, data breaches, and unauthorized manipulation. This includes cybersecurity practices, adversarial robustness, and access controls to preserve system reliability and trust.

Belgrade Ministerial Declaration on Artificial IntelligenceDefinition 1 of 3

AI systems implemented with protections against unauthorized access, manipulation, and exploitation, including cybersecurity controls, model and data protection, and resilience to adversarial attacks and malicious misuse.

UN General Assembly Resolution on Artificial Intelligence (A/78/L.49)Definition 2 of 3

AI systems protected against malicious actors, cyber threats, and misuse, ensuring data integrity and system resilience.

Declaration: Towards the Safe, Secure, and Trustworthy Development and Deployment of Artificial Intelligence in the Americas: The Importance of Governance, Regulatory, and Institutional FrameworksDefinition 3 of 3

Not to be confused with

Security (AI)

Security focuses on malicious attacks, while Secure AI also addresses cyber threats, misuse, and infrastructure integrity.

Safe AI

Safe AI prevents unintended harm across the lifecycle, whereas Secure AI targets malicious manipulation, cyber threats, and unauthorized access.

Trust/security

Trust/security includes privacy, data governance, robustness, and cybersecurity, but Secure AI specifically addresses malicious actors, adversarial attacks, and system resilience.

Questions and answers

What makes the OAS definition of Secure AI distinctive?

It protects AI systems against malicious actors, cyber threats, and misuse while ensuring data integrity and system resilience within a framework balancing innovation and human rights.

How does the Belgrade Declaration define Secure AI?

It defines Secure AI as systems and infrastructure with measures protecting integrity, confidentiality, and availability from adversarial attacks, data breaches, and unauthorized manipulation, including cybersecurity, robustness, and access controls.

What frameworks promote Secure AI?

The OAS Declaration provides a regional framework for the Americas, and the Belgrade Ministerial Declaration guides GPAI member states and the EU.