Pushing towards transparency and best practice sharing within the AI community, leading AI companies (Amazon, Anthropic, DeepMind, Meta,Microsoft, OpenAI) have today unveiled their safety policies. This initiative follows a request from the Technology Secretary last month, demonstrating a collective effort to boost transparency and promote responsible AI development.
The backdrop to this unveiling is the UK Government’s release of emerging safety procedures aimed at guiding companies on ensuring the safety of AI models. These guidelines are set to shape discussions at the upcoming Bletchley Park event. The government’s paper outlines a set of practices for AI companies, which includes the implementation of responsible capability scaling. This framework represents a novel approach to managing frontier AI risk, with several companies already adopting it.
Responsible capability scaling entails AI firms preemptively outlining the risks they intend to monitor, establishing protocols for notifying relevant stakeholders if these risks manifest, and setting thresholds at which developers should slow down or halt their work until enhanced safety measures are in place. Such a forward-looking approach aims to proactively address potential risks.
The government’s suggestions also encompass AI developers collaborating with third parties to probe their systems for vulnerabilities, an attempt to identify sources of risk and potential harmful consequences. Additionally, AI companies are encouraged to provide clear information indicating whether content has been generated or modified by AI.
At the heart of these emerging safety practices is the pursuit of innovation. The UK Government underscores the necessity of understanding the risks inherent in the frontier of AI development to fully harness the economic growth and public good potential.
Prime Minister’s recent announcement revealed plans to establish the world’s first AI Safety Institute. This institute will focus on advancing AI safety knowledge and evaluating new AI models. It will also seek to share information with international partners, policymakers, private companies, academia, and civil society, fostering collaboration in AI safety research.
Moreover, recent findings from a survey indicate strong global support for a government-backed AI safety institute to evaluate powerful AI. Notably, 62% of those surveyed in the UK endorsed this idea. The survey spanned nine countries, including Canada, France, Japan, the UK, and the USA, demonstrating widespread backing for independent expert evaluations of powerful AI.
The publication of today’s paper contains processes and associated practices already being implemented by some frontier AI organizations. Others, such as responsible capability scaling, are tailor-made for frontier AI, distinct from lower capability or non-frontier AI systems.
Technology Secretary Michelle Donelan noted that these processes and practices represent the start of a vital conversation, with the understanding that they will continue to evolve as technology advances. The aim is to increase public trust in AI models and drive their adoption across society.
The paper also underscores the long-standing technical challenges in building safe AI systems, particularly regarding safety evaluations and decision-making processes. Safety best practices for frontier AI are yet to be firmly established, which is why the UK Government has introduced these emerging processes.
Rapid advancements in frontier AI pose the risk of outpacing human understanding and control. The AI Safety Summit aims to tackle these challenges by focusing on the management of risks associated with frontier AI, including misuse, loss of control, and societal harm. Frontier AI organizations are central to addressing these risks and ensuring responsible AI development and deployment.
Frontier AI Taskforce Chair Ian Hogarth emphasized the need to enhance transparency in more capable AI systems to assure responsible development. These emerging safety policies are intended to support frontier AI companies in establishing effective AI safety protocols.
Adam Leon Smith, of BCS, The Chartered Institute for IT, and Chair of its Fellows Technical Advisory Group (F-TAG), applauded these emerging processes and practices as significant advancements in the industry. He emphasized the importance of having the vision to anticipate risks and the potential for these practices to benefit the management of risks posed by AI systems already in the market.
As the UK hosts the AI Safety Summit, the government is taking a long-term perspective to make informed decisions that will shape a brighter future driven by AI advancements. This commitment underscores the UK’s dedication to fostering responsible AI innovation and ensuring its safe integration into society.
Statements from Leading AI Companies
OpenAi
OpenAI joined other leading AI labs in making a set of voluntary commitments to promote safety, security, and trust in AI. These commitments encompassed a range of risk areas, centrally including the frontier risks that are the focus of the upcoming AI Safety Summit.
In this update, we describe our progress on those voluntary commitments and further detail our evolving approach to mitigating frontier risks, including our ongoing work to develop a risk-informed development policy.
On October 3, 2023 we publicly released the system card for our text to image model DALL-E 3, the first major public release of a new frontier model within the scope of our voluntary commitments. In line with both our mission and the voluntary commitments, we did critical safety work including pre-deployment safety evaluation and red-teaming. In addition, we are working toward new methods to empower people to track the provenance of AI-generated media, and have continued to invest in responsible practices through our rollout of voice and image analysis capabilities in ChatGPT.
We have also met our voluntary commitment to “establish or join a forum or mechanism through which [we] can develop, advance, and adopt shared standards and best practices for frontier AI safety,” by co-founding the Frontier Model Forum. This new industry body, jointly established with Microsoft, Google Deepmind, and Anthropic, is a venue to advance AI safety research and promote responsible development practices for frontier AI systems.
Microsoft
Microsoft welcomes the opportunity to share information about how we are advancing responsible artificial intelligence (AI), including by implementing voluntary commitments that we and others made at the White House convening in July.[1] Visibility into our policies and how we put them into practice helps to inform and accelerate responsible technology development and deployment. It can also strengthen public-private partnerships driving progress on AI safety, security, and trust.
As a developer and deployer of AI models, API services, and applications, Microsoft works to map, measure, and manage risk and apply multi-layered governance that embeds robust checks on processes and outcomes. For frontier models specifically, Microsoft works closely with OpenAI.
Since 2019, Microsoft and OpenAI have been engaged in a long-term collaboration to develop advanced AI systems, underpinned by a shared commitment to responsible development and deployment practices. Microsoft’s efforts to deploy frontier models at scale build upon and complement OpenAI’s leading model development practices. For a comprehensive accounting of the model development and deployment practices that apply to OpenAI’s frontier models as deployed in Microsoft’s offerings, OpenAI’s and Microsoft’s responses to the UK Government’s AI Safety Policies Request should be read together.
Meta
In relation to frontier AI systems, we have signed up to the White House Commitments. We believe the commitments are an important first step in ensuring responsible guardrails are established for AI as we look towards a future with increasingly capable systems. We joined these commitments because they represent an emerging industry-wide consensus around the things that we have been building into our products for years. Our five pillars of responsible AI have been core to our own development of AI for years, and the White House AI commitments’ themes of safety, security, and trust, reflect those same values. These commitments, built with input from leading AI companies across industry, in close collaboration with the White House, strike a reasonable balance of addressing today’s concerns and the potential risks of the future.
Responsible AI work is iterative – better solutions are developed, and new challenges emerge, requiring us to continuously adapt and innovate. As demonstrated by recent discussions on frontier AI systems, the AI community is now looking towards the next generation of AI technologies. The AI Safety Summit will be an excellent opportunity for us to continue to explore the adaptations and innovations that our community may need to develop in the coming years. We are excited to be a part of that process, and look forward to more of these discussions at the AISS.
Google DeepMind
At Google DeepMind, we aim to build AI responsibly to benefit humanity. At the heart of this mission is our commitment to act as responsible pioneers in the field of AI, in service of society’s needs and expectations. We believe applying AI across all sorts of domains – including scientific disciplines, economic sectors, and to improve and develop new products and services – will unlock new levels of human progress. However, we need to develop and deploy this technology thoughtfully and responsibly — our mission is only achievable with the responsible development and deployment of AI systems.
We’ve already seen people use AI to address societal challenges, including by helping scientists better detect breast cancer, forecast floods, limit the warming effects of jet contrails, accelerate clean nuclear fusion, predict protein structures, and achieve healthcare breakthroughs. Vast potential remains to supercharge scientific research and economic productivity, tackle global challenges like climate change and co-create new approaches to perennial policy priorities like education. This is the opportunity that we must keep in mind as we work together to develop guardrails for the development and deployment of the technology.
Anthropic
We are heartened that the British Government have advocated for Responsible Capability Scaling. Anthropic agrees that such a protocol is key to managing the risks of developing increasingly capable AI systems and would like to see all frontier developers adopt such policies. This motivated the publication of our Responsible Scaling Policy (RSP) on 19 September 2023.
Anthropic’s RSP is a series of technical and organizational protocols that we are unilaterally adopting. Perhaps the most notable aspect of the RSP is our commitment to pause the scaling1 and/or delay the deployment of new models whenever our scaling capability outstrips our ability to comply with the safety procedures for the corresponding AI Safety Level.
Moreover, our RSP focuses on catastrophic risks—those where an AI model could directly cause large-scale devastation. Such risks can come from deliberate misuse of models (for example, use by terrorists or state actors to create bioweapons) or from models that cause destruction by acting autonomously in ways contrary to the intent of their designers.
Our RSP defines a framework called AI Safety Levels (ASL) for addressing catastrophic risks, modeled loosely after the US government’s biosafety level (BSL) standards for handling of dangerous biological materials. The basic idea is to require safety, security, and operational standards commensurate with a model’s potential for catastrophic risk, with higher ASL levels requiring increasingly strict demonstrations of safety.
Amazon
We thank the government of the United Kingdom for organizing the AI Safety Summit, which comes at a crucial moment in the continuing development and public understanding of AI. The pace of AI innovation in just the past year has captivated public attention. While there is recognition that AI has the potential to reshape the economy and improve modern life, recent advances have also prompted understandable questions about whether and how organisations developing “frontier” AI systems are accounting for their unique risks. Because the opportunities and the risks presented by AI are fundamentally global in nature, the AI Safety Summit is an opportune moment for industry, governments, and civil society to align around a shared vision for ensuring that frontier AI is safe, secure, and trustworthy.
Earlier this year we took an important first step towards this goal when we proudly endorsed the White House Voluntary AI Commitments. Recognizing the dynamic state-of-the-art, the White House Commitments are built on a forward-looking set of best practices that are flexible and durable enough to evolve along with the technology. The Commitments set out ambitious and concrete objectives for managing many of the unique risks of generative AI and for building trust with the public. While Amazon initially signed the Commitments in the US, our efforts to operationalize them are not territorially limited.
The AI Safety Summit is an opportunity to build on the White House Voluntary AI Commitments. To foster global confidence that the Commitments are meaningful, it is important that we provide information about how our existing AI development practices align with the underlying commitments and the additional areas of inquiry set out in the “AI Safety Policies” questionnaire. Our responses below are organised by topic and highlight the best practices and policies that guided our development and deployment of Amazon Titan, the family of large language models we recently launched in connection with Bedrock, our foundation model service.