US and UK agree to jointly test the security of artificial intelligence models

The US and UK governments have signed a memorandum of understanding to create a common approach to independently assess the security of advanced artificial intelligence models like ChatGPT and DALL-E.

Here's What We Know

Under the partnership, the UK's AI Safety Institute and its US counterpart, previously announced by Vice President Kamala Harris, will jointly develop test suites for risk assessment and security of cutting-edge AI models.

The parties plan to share technical expertise, information and personnel. One of the initial goals will be to conduct joint tests on a publicly available AI model.

UK Science Minister Michelle Donelan emphasised the need to act quickly, as a new generation of AI models is expected to emerge next year whose capabilities are not yet known.

The partnership is the world's first bilateral agreement on AI safety, although both the US and UK intend to engage other countries in the future.

US Commerce Secretary Gina Raimondo said the partnership will accelerate the work of institutions to assess the risks of AI to national security and society at large, providing a deeper understanding of AI systems and issuing rigorous recommendations.

Flashback

In addition to testing and evaluation, governments around the world are also developing rules to regulate AI technologies. The White House signed an executive order to ensure that federal agencies only use safe AI tools, and the European Parliament has approved sweeping legislation banning manipulative and discriminatory AI systems, as well as requiring the labelling of AI-generated content.

Source: Engadget