Apple Pledges to the White House’s AI Safety Initiative

Apple has pledged its commitment to the White House’s initiative for the development of AI that is ethical, secure, and reliable, according to a news release published on Friday. This commitment comes as the tech giant prepares to integrate its new generative AI service, Apple Intelligence, into its flagship products, thus bringing generative AI to its 2 billion customers.

Entering into this commitment alongside 15 other tech powerhouses including Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI, Apple adopted the Biden administration’s principles for the development of generative AI in July 2023. Until then, Apple had kept under wraps the extent to which it intended to incorporate AI into iOS. However, the company made its intentions unmistakably clear at the WWDC event in June, announcing an aggressive move towards integrating generative AI, beginning with incorporating ChatGPT into the iPhone. In doing so, amidst frequent scrutiny by federal oversight bodies, Apple aims to demonstrate its readiness to conform to the guiding principles set by the White House on AI, potentially to gain favor in anticipation of imminent legislative challenges surrounding AI.

But what is the actual impact of Apple’s voluntary pledges to the White House? They represent a modest beginning at best. The White House has described these as a “first step” for Apple, along with 15 other AI entities, to pursue the creation of AI technologies that are dependable, secure, and ethical. This initiative was followed by President Biden’s AI-focused executive order in October, with various bills aimed at enhancing AI regulation now making their way through legislative bodies at both the federal and state levels.

According to the pledge, AI enterprises commit to conducting red-team exercises (simulating cyber attacks to test an organization’s defenses) on AI models prior to public deployment and disclosing those findings openly. Moreover, the White House’s voluntary initiative requires AI firms to handle unpublicized AI model data with confidentiality. Apple, along with others, agrees to manage AI model data within highly secure settings and to restrict access to such data to a minimum number of staff. Furthermore, these companies are committing to the development of mechanisms for content authentication, such as digital watermarks, to enable users to identify content generated by AI.

In a related note, the Department of Commerce has announced its plan to release a study on the possible advantages, hazards, and ramifications of using open-source foundational models in AI. The discourse surrounding open-source AI is becoming a focal point of regulatory debate, with opinions divided over the accessibility of AI model data. Restricting access could potentially hamper the growth of AI startups and research. The position the White House takes on this issue could profoundly influence the broader AI sector.

Additionally, the White House has highlighted notable advancements by federal agencies in fulfilling the goals of the AI executive order from October. To date, these agencies have amassed over 200 AI-related appointments, granted more than 80 research teams access to essential computational resources, and unveiled numerous AI development frameworks (a favored tool of the government).