The United Kingdom and the United States, together with international partners from 16 other countries, they have published new guidelines for the development of safe artificial intelligence (AI) systems.
What do these guidelines for the development of artificial intelligence consist of
“The approach prioritizes ownership of safety outcomes for customers, embraces transparency and radical accountability and establishes organizational structures where safe design is a top priority“, has declared the United States Cybersecurity and Infrastructure Security Agency (CISA).
The goal is to increase the levels of AI cybersecurity and ensure that the technology is designed, developed and implemented securelyhas added the National Center for Cyber Security (NCSC).
The guidelines are also based on ongoing efforts of the United States government to manage the risks posed by Artificial Intelligenceensuring that new tools are adequately tested before their public distribution, with adequate protections to address societal harms, such as bias and discrimination, and privacy concerns, and establishing robust methods for consumers to identify content generated by the ‘AI.
This is used for safe design guidelines It also commits companies to encouraging the discovery and reporting of vulnerabilities in their AI systems by third partiesthrough a system of bug bountiesso that such vulnerabilities and various security and privacy flaws can be identified and resolved quickly.
The last guidelines “they help developers ensure that cybersecurity both an essential condition for the safety of the AI system and an integral part of the development process from the beginning and throughout the life cycle“the NCSC said.
This includes secure design, secure development, safe implementation and safe operation and maintenance, covering all significant areas in AI system developmentrequiring organizations to model threats to their systems and protect their supply chains and infrastructure.
The objective, the agencies underlined, is also to combat the attacks by cyber criminals aimed at artificial intelligence and machine learning (ML) systems which aim to cause unintentional behavior in various ways, including influencing the classification of a modelallow users to perform unauthorized actions and extract sensitive information.
“There are multiple ways to achieve these effects, such as fast injection attacks in the large language model (LLM) domain, or the deliberate corruption of training data or user feedback (known as ‘data poisoning‘)“the NCSC finally said.
Some brief considerations
We need to make some brief considerations on these new guidelines reflect the growing commitment to ensuring safety in the field of artificial intelligence.
The “secure by design” approach highlights the importance of integrating security from the early stages of AI developmenthighlighting the need to thoughtfully consider threats, protect supply chains and respond promptly to vulnerabilities.
The establishment of a bug bounty system (literally a “bounty hunt” for bugs) highlights the recognition of the importance of collaboration and of information sharing in the timely detection and resolution of potential risks and these guidelines, promoted through international collaboration, represent a significant step towards a more uniform and global approach to AI safety, fundamental to mitigating emerging challenges and building a solid foundation for the safe adoption of this technology advanced.
#Artificial #Intelligence #countries #release #development #guidance