In response to Holden Karnofsky’s call for case studies on standards, I’ve spent the last six weeks or so researching the Basel Accords—the premier set of standards for international banks that have become deeply embedded into banking regulation across the globe. My work culminated in this case study, written primarily to inform potential artificial intelligence and biosecurity standards. I’ve laid out the key takeaways below.
Key Takeaways
1. What are the Basel Accords?
The Basel Accords are a set of banking standards (Basel I, II, and III) set by the Basel Committee on Bank Supervision (BCBS). The current set of standards, a continuation of Basel III and subsequent updates, is known as the Basel Framework.
They are aimed at dealing with credit, market, capital, and operational risks that affect the banking sector. To deal with these risks, the Basel Accords set conformance standards targeted at ‘internationally active banks’.
In particular, Basel I (1988-2004) laid out a framework for risk-weighted capital requirements for banks; Basel II (2004-2010) expanded to market and operational risk, and Basel III (2010 - present) kicked off an expansion to Basel II culminating in a complete regulatory framework that includes different measures and requirements of capital risk for different contexts; guidance on managing operational risk at banks, and detailed rules on bank supervision.
2. How are the Basel Accords implemented today?
The Basel Committee on Banking Supervision (BCBS) is one of several committees of the Bank for International Settlements (formed of the central banks of (now) 63 central banks and monetary authorities). Its mandate is concerned with developing global regulatory standards for banks. It was formed in 1930 and comprises 45 voluntary members from 28 jurisdictions.
The structure of BCBS comprises a chair that governs the activity of the Basel Committee, standing groups with specialised functions, and a Secretariat that provides administrative and analytical support to the Committee and its numerous working groups. The Secretariat also organises the drafting of standards. The BCBS reports to the Group of Central Bank Governors and Heads of Supervision (GHOS), which oversees and endorses major BCBS decisions.
Its policy development is a messy process where many members of the BCBS itself consult widely with stakeholders for years to come to a consensus on priorities and standards. This consultation is extremely wide-ranging: involving regular meetings; policy proposals sought from non-members; working groups and task forces that take on specific research questions and aspects of standard development; regular communication with governments and central banks, and comments from the public. Consensus-forming is the central function of the Basel Committee: it maintains a flexible structure and process, ascertaining priorities and next steps as it continues to consult with its member states and its wider set of stakeholders, culminating in the publication of reports, updates to standards, and new standards when needed.
Basel members agree to adopt the Basel Accords, which they typically do, given all members are closely involved in developing standards. However, the Basel Accords have been adopted in over 100 jurisdictions (albeit to varying degrees of compliance) and even often applied to domestic banks, given it is now the norm in international financial regulation.
Basel members are responsible for implementing the Basel Accords, usually done via national legislation. States are responsible for enforcement (e.g. whether audits are held).
3. Why and how did the Basel Accords emerge?
Efforts to increase globalisation after World War II, yet consciousness about the economic risks involved (particularly given several banking crises had already occurred) led to increased international trade and several international organisations that attempted to ensure international stability—namely the IMF G10 and the European Economic Community.
International trade between Europe and the US led to large amounts of US dollars held in foreign accounts (the ‘Eurodollar’), and this market caused concern amongst central bankers and the aforementioned institutions. The IMF initially created the Euro-currency Standing Committee in April 1971, and a collection of central bankers at EEC organisations formed the informal Groupe de Contact in 1972 to discuss international regulation.
In 1973, the Yom Kippur War increased oil prices and spurred central banks to figure out how to recycle the flow of funds into their economies. In 1974, Bankhaus Herstatt was a German bank that collapsed due to making bad bets on the US dollar, and this led to large US losses due to regulatory inconsistencies resulting in US banks not receiving back dollars for Deutschmark they had sold to Bankhaus Herstatt. The collapse of Bakhaus Herstatt triggered awareness of a need for greater international regulation. Through the Bank for International Settlements, individuals at the Euro-currency Standing Committee, Groupe de Contact, and central banks formed the Basel Committee on Banking Standards, consisting of the G10 countries.
The Basel Committee began with principles for bank supervision. But increased global interest in capital adequacy and pressure from the US (and to a lesser extent the UK) for regulatory standardisation, particularly due to fears of Japanese banks remaining more profitable by having more lax capital adequacy standards, led to the emergence of Basel I—a capital adequacy standard adopted by the member states of the Basel Committee.
Basel II emerged in 2004 after a call for a new standard in 1999, given Basel I was insufficient. Basel III emerged in 2010 after the financial crisis, also resulting in an expansion in membership and more widespread adoption.
4. What can we learn from the Basel Accords?
My takeaways (and expert opinion) on lessons from the Basel Accords for implementing a Basel-like international standard include:
Consensus is key: continuous consensus-forming, relationship-building, and widespread consultation is the cornerstone of how the Basel Accords became an international norm of financial regulation, even given the contestability of financial regulation.
Consensus takes time: achieving the level of consensus of the Basel Accords takes many years.
Consensus is a continuous process: the Basel Committee maintain regular lines of communication to stay highly sensitive to shifting incentives, thinking, and priorities.
A clearly-defined problem space was important: consensus was possible because there was agreement about the importance of avoiding a banking crisis (and clarity on what one looked like).
Earlier market failures were pivotal for creating this consensus and spurring action.
Alignment with commercial incentives was important but can be (and was) shaped.
Market pressure to comply was not merely implicit but was widely perceived.
A flexible organisational structure and dynamic processes are important.
Institutional backing mattered: the Basel Accords were built on the BIS, IMF, and EEC.
A dedicated operational unit (the Secretariat) was vital for structuring a messy process.
Clear and concerted leadership is important: initial adoption was driven by a few individuals who owned the process.
Power matters: US interest in capital adequacy standards was important for creating pressure that ultimately led to the emergence and adoption of Basel I.
Incessant and indirect communication is key: the BCBS also runs events, consultations, and support outside of its standard-setting to ensure continued relationship-forming.
Maximising transparency was valuable for trust.
The structure of the Basel Committee means little work is done after the publication of a standard to ensure compliance, given members already commit and are involved in the consultative process.
Factoring scaling and compliance considerations into the design of the standard itself is important, and necessarily occurs given the extensive consultation that goes into standard design.
Complexity brings complex trade-offs: it’s not straightforward that complexity and compliance are inversely proportional. Basel I was too simple to mitigate financial risk effectively, and Basel III commands compliance given the ability of states to pick which aspects are most relevant to it.
Standards don’t have to be perfect when developed in an iterative and high-feedback way.
Targeting the most important players made a difference: the G10 was a small group that represented most of the world’s banking capital.
Involving everyone is a core part of The Basel Process: anyone who might be affected or has a stake in the Basel Accords was consulted or had avenues to communicate with the Committee.
Relationship-forming was essential: leveraging a wide set of stakeholders required a wide network and continuous relationship forming with central banks and organisations worldwide.
Iterative development was important: Basel I started simple and gradually built up until today.
Key causes of resistance are worries about economic impact and restricting sovereignty.
Basel Committee communications routinely affirm the importance of financial risk mitigation.
Providing widespread support: for implementation helped ensure widespread adoption of a complex standard, albeit much more resource-intensive.
Some specific levers for organisational competence include being very well-prepped, well-connected, and well-read.
Domestic consensus is key for national adoption and influencing the international agenda.
Some speculative and lower-confidence takes relevant to AI and biosecurity standards include:
I’m not excited about prioritising near-term efforts to develop Basel-like international standards without much more background work done to establish consensus about AI and biosecurity issues.
I wouldn’t be excited about AI/bio standards that aren’t aligned with commercial incentives, but I think there are opportunities to shape these incentives as the Basel Accords did.
Given how contestable AI and biosecurity regulation will likely be, a consensus-driven approach could be important.
A consensus-driven approach may have limited scope for AI and bio, however, because there are additional risks from malicious agents not found in financial risk mitigation.
Standards can take quite a lot of complexity, although the best-case scenario may look like partial compliance. This could affect how we think about AI/bio standards.
In general, I would expect an effort to create Basel-like standards to be less successful in expectation unless there are numerous (non-existential) early warnings, given how important this was for Basel.
Part of the theory of change for AI and bio x-risk research should be academic consensus-forming.
Potential standard-setters and key stakeholders could probably be doing a lot more relationship-forming now if the ambition is a Basel-like international standard.
We shouldn’t rely on standards to delay AI timelines, given how long consensus-forming takes.
I think there’s an exciting opportunity right now to develop better cooperation between the AI labs.
I think voluntary standards right now could be very useful for AI, primarily for just establishing the personnel, agenda, and infrastructure for future standards.
For biosecurity, I think IBBIS is in a similar position to the pre-1988 BCBS, and could potentially learn from the details of the history of the Basel Accords.
The full report remains fairly scrappy, and I’d encourage further feedback and thoughts—particularly corrections of any factual errors. I’ll likely factor these into some minor updates in the future, and I’m also keen to continue adding lessons. However, I didn’t want perfection to be the enemy of the good and felt this was ready enough for public reading. More generally, feel free to reach out for further questions and/or comments.
See the appendices for some conceptual background to standards themselves; reasons I was excited about the Basel Accords as a case study for AI and biosecurity; an overview of some of the successes and criticisms of the Basel Accords; an overview of why the Basel Accords have achieved widespread adoption amongst non-members; an overview of what I consider the key strategic questions for a standard-setter to consider, and some further explanation of my tentative takeaways for AI and bio.
Acknowledgements: thank you to Open Philanthropy for sponsoring this project; Holden Karnofsky for providing some early feedback and direction; all the individuals I interviewed for this project for their invaluable input, and Rachel Edwards and Kit Harris for their assistance with connections and thoughts!
At first scan, this is looks like a lot of really impressive work. I’m looking forward to digging in because I think this is a clever analogy that I wish I would have thought of! I’m new today here at LW, sorry if this is dumb question: I don’t see your bio or link to an individual social page (eg, LinkedIn), is there a reason there isn’t an “About the author” section, for readers who are curious who you are? Thank you!