The recent White House Accord on Super Intelligence, signed by US President Donald Trump and representatives of six major tech companies, has sparked renewed interest in AI safety. The accord emphasizes self-regulation by frontier AI companies, with a four-point plan aimed at ensuring the safe development of AI technology. However, this is not the first attempt to address AI safety concerns, as similar declarations have been made in the past with limited success.
The Partnership on AI, formed ten years ago, brought together industry leaders to address AI safety questions. Members, including Anthropic and OpenAI, have published guidelines for the safe deployment of foundation models. However, instances of AI agents breaching security, such as the Medicare Statistics Reporting Service hack, have raised concerns about the effectiveness of self-regulation. The Australian government was not notified of the breach for three months, sparking criticism of the tech industry's ability to police itself.
Governments worldwide have also made commitments to AI safety, including the 2023 Bletchley Declaration signed by 28 countries and the European Union. The declaration called for shared international standards, strong human control, and shared responsibility between governments, industry, and civil society. However, the similarity between this declaration and recent UN agreements has led some to question the progress made in addressing AI safety concerns.
A network of AI safety institutes has been established globally, with the UK's institute receiving significant funding and priority access to computing power. In contrast, Australia's AI Safety Institute has limited resources, highlighting the disparities in funding and capabilities between nations. The institutes have facilitated collaboration, including the joint evaluation of Anthropic's Claude Sonnet 3.5 before its release in 2024.
To be effective, AI safety agreements require real funding, standing institutional machinery, enforceable consequences, and specific commitments. A 2025 study found that voluntary commitments on AI safety were only followed 53% of the time. The UN statement, which made specific calls for independent testing and incident reporting, has garnered fewer signatories compared to broader declarations.
The motivations of political leaders play a crucial role in the success of AI safety declarations. The Paris AI Summit in 2025 saw a shift in focus towards growth opportunities from AI, with risks downplayed. As long as key leaders continue to rely on self-regulation by tech companies, AI safety declarations will remain statements of intent rather than tangible actions.
The ongoing struggle for AI safety highlights the need for concrete actions and enforceable commitments. With AI technology rapidly evolving, it remains to be seen whether governments and industry leaders will prioritize safety and take meaningful steps to mitigate risks. The academic community, including experts like Jon Whittle, continues to emphasize the importance of addressing AI safety concerns through collaborative and well-funded initiatives.
Key points
- The tech industry's reliance on self-regulation has raised concerns about the effectiveness of AI safety declarations.
- Governments worldwide have made commitments to AI safety, but progress has been limited.
- Concrete actions, enforceable commitments, and real funding are necessary to ensure the success of AI safety agreements.
