- United Kingdom AI The Safety Institute is looking to expand across the Atlantic to expand its technical expertise and cement its position as a global authority AI Security.
- Extension presented as AI The Safety Institute is publishing for the first time ever AI Results of safety tests on publicly available models and agrees to new collaboration with Canada.
- comes before the jointly organized one AI Seoul summit, UK demonstration AI The Safety Institute’s continued leadership on a global scale AI Security.
The pioneering work of the British government AI The Safety Institute will expand its international horizons by opening its first overseas office in San Francisco this summer, Technology Minister Michelle Donelan announced today (Monday, May 20).
The expansion represents a critical step that will enable the UK to tap into the Bay Area’s wealth of technical talent and collaborate with the world’s greatest AI Laboratories headquartered in London and San Francisco and strengthening relations with the United States to move forward AI Security in the public interest.
The office is expected to open in the summer and hire the first team of technical staff led by a research director.
It will be a complementary branch of the London Institute headquarterswhich is getting stronger and already has a team of over 30 technical employees. The London office will continue to grow and acquire the necessary expertise to assess the risks of border areas. AI systems.
By expanding its presence in the United States, the Institute will build close cooperation with the United States and support the country’s strategic partnership and approach to AI security while sharing research and conducting joint assessments AI Models that can inform AI Security policy around the world.
Michelle Donelan, Secretary of State for Science and Technology, said:
This expansion represents the British leadership role AI in action. It is a crucial moment for the UK’s ability to examine both the risks and potential of AI From a global perspective, we are strengthening our partnership with the United States and paving the way for other countries to leverage our expertise as we continue to lead the world AI Security.
Since the Prime Minister and I the AI Safety Institute has gone from strength to strength and in just over a year we have built the world’s leading government here in London AI Research team attracting top talent from the UK and beyond.
Opening our doors abroad and expanding our alliance with the United States are central parts of my plan to set new international standards AI security, which we will discuss at the summit in Seoul this week.
The expansion takes place in the United Kingdom AI The Safety Institute publishes a selection of current safety test results from five publicly available advanced tests AI Models: The first government-funded organization in the world has published the results of its assessments.
While the results represent only a small part of the Institute’s broader approach, they demonstrate the significant progress the Institute has made since November AI Safety Summit as it builds its state-of-the-art safety testing capabilities.
The institute evaluated AI Models in four key risk areas, including how effective the security measures installed by developers actually are in practice. The institute’s tests have shown, among other things, that:
- Several models have overcome cybersecurity challenges while struggling to overcome more demanding challenges.
- Several models show similar ones PhDKnowledge of chemistry and biology at a level.
- All models tested remain highly vulnerable to basic jailbreaks, and some produce malicious output even without targeted attempts to bypass security measures.
- Tested models were unable to perform more complex, time-consuming tasks without human supervision.
AI Ian Hogarth, Chairman of the Safety Institute, said:
The results of these tests mark the first time that we have been able to share some details of our model evaluation work with the public. Our evaluations will contribute to an empirical assessment of the model capabilities and the lack of robustness of existing protective measures.
AI Security is still a very young and emerging field. These results represent only a small part of the evaluation approach AISI develops. Our goal is to further expand the boundaries of this area by developing state-of-the-art assessments with a focus on national security risks.
AI Safety remains the UK’s top priority as it continues to drive the global conversation around the safe development of technology.
This effort was launched in November AI security summit at Bletchley Park, and momentum continues to build as the UK and the Republic of Korea prepare to co-host the summit AI Seoul summit this week.
As the world prepares to gather in Seoul this week, the UK is committed to working with Canada, including through their respective countries AI Safety Institutes to advance their goal and create a growing network of government-supported organizations that the organization focuses on AI Security and governance. Confirmed by British Technology Minister Michelle Donelan and Canadian Science and Innovation Minister François-Philippe Champagne, this partnership will serve to deepen existing connections between the two nations and stimulate joint work in the area of systemic security research.
Under this agreement, countries want to share their expertise to strengthen existing testing and evaluation work. The partnership will also enable deployment routes between the two countries and work to jointly identify areas for research collaboration.
Notes for editors
The institute’s security testing this year was conducted on five publicly available large language models (LLMs), which are trained on large amounts of data. The tested models were anonymized.
The results represent only a snapshot of the model functions and do not label systems as “safe” or “unsafe”. The tests performed represent a small part of the evaluation techniques AISI developed and used, as outlined in the Institute’s evaluation approach published earlier this year.
Today’s publication can be found on the AI Security Institute website.
Today also sees the latest progress update from Institute Chair Ian Hogarth, which can be found on the website here AI Security Institute website.