On October 19, the 2026 Bund Summit opened in Huangpu District, Shanghai.
The annual event is co-hosted by the China Finance 40 Forum (CF40) and Tsinghua University, with the theme "Technology, Geopolitics and the Global Economy," and runs through October 22.
Committed to its "international" and "professional" positioning, this year's summit features 19 Bund Roundtables and 12 closed-door meetings.
So far, 126 international guests from 24 countries and regions have confirmed they will attend and speak, including 9 central bank governors, 4 deputy governors and 6 finance ministers and deputy finance ministers.
On the first day, artificial intelligence (AI) became the focus of discussion.
Attendees' concerns ranged from AI's impact on productivity, employment and income distribution to AI investment and AI governance.
Several guests said during the event that some risks brought by AI have not yet been fully seen — they are not on balance sheets, not within the traditional regulatory horizon, and not even within humanity's existing understanding of technology.
AI and the macroeconomy: productivity dividends and income distribution pressure
Yang Bin, vice chairman of the Tsinghua University Council, said in his opening address that rather than viewing AI with the additive thinking of "AI+," it is better to understand it as "AI to the power of x": what AI brings is not an incremental tool within a single job, but a systemic reconstruction of an entire industry and even society as a whole.
He compared the current state of AI technology to a rapidly iterating "technological adolescence," while many organizations still carry the traits of "organizational middle age," pursuing short-term returns and accustomed to linear improvement.
The collision between the two will prompt deep reflection on business logic and organizational models.
Robert Rubin, a member of the Bund Summit International Advisory Committee, the 70th U.S. Secretary of the Treasury and honorary co-chair of the Council on Foreign Relations, and Huang Yiping, dean of the National School of Development at Peking University, were the first to open a dialogue on AI's impact on the labor market.
This was also the seventh consecutive year the two have held a dialogue at the Bund Summit.
Rubin pointed out that AI may significantly boost productivity, but on the other side, knowledge workers may lose jobs on a large scale, and these people may not have the skills to shift to other work.
The complexity of the transition should not be underestimated.
Citing economic historian Robert Allen's research on the First Industrial Revolution, he noted that the "Engels pause," in which output rose substantially while real wages stagnated for a long time, is also highly relevant today.
If the technological dividends brought by AI tilt more toward capital and insufficiently toward labor, it will not only worsen income distribution imbalances but also hit the fiscal and tax structure, because under the current system labor income is taxed far more heavily than capital income.
Huang Yiping argued that if productivity gains push up aggregate supply while employment and income distribution constrain aggregate demand, more proactive public policy may be needed.
He asked Rubin about the feasibility of a universal basic income (UBI) plan.
Rubin said bluntly that UBI is not a good approach, arguing that people want to work and that work is also about identity and opportunity.
At the CF40-G30 Bund Roundtable "Central Banks in Transition: Mandates, Markets and Macroprudential Risks," AI's macroeconomic impact also became a focus of discussion.
Donald Kohn, former vice chairman of the Federal Reserve, said the AI investment boom is pulling demand before it affects supply, and is increasing inflationary pressure.
He also noted that the U.S. labor market is performing steadily and wage growth is not high, and that workers' concerns about AI may have suppressed demands for pay raises.
Jacob Frenkel, former governor of the Bank of Israel, said AI's impact on supply could be large but will not appear in the short term; at the same time, asset valuations based on future expectations are already driving demand.
Whether AI ultimately leans toward pushing inflation up or holding it down needs to be assessed within this framework.
He cautioned that the global economy has performed better over the past two years than previously feared, and people often attribute this to AI, but before clarifying its mechanism of action, it is not advisable to rely too heavily on this explanation.
Philip Lane, chief economist of the European Central Bank, said AI's global supply chain gave a boost to euro area trade this year and also provided some stimulus to the euro area economy, though to a lesser extent than in the United States.
He said euro area wages are relatively moderate and services inflation has not risen significantly, which is related to two concerns: workers fear being replaced by AI, while companies fear that competition from China will squeeze their business.
Masaaki Shirakawa, former governor of the Bank of Japan, offered another perspective: AI's impact on democracy.
He recalled that when the internet emerged 30 years ago, outsiders optimistically believed it would improve democracy, but various negative effects later appeared.
From this, he raised the question of how AI affects democracy, which he said relates to how monetary policy and financial stability measures operate.
AI and finance: debt financing and "circular commitments"
In his dialogue with Rubin, Huang Yiping raised the risk of "circular commitments" in AI investment: large technology companies make huge procurement commitments, upstream suppliers expand sharply as a result, and once procurement commitments cannot be fulfilled for some reason, it could trigger a sharp market adjustment.
Rubin said the U.S. banking system is healthy overall, large banks lending to AI-related businesses are well capitalized, and regional banks have basically not entered this field; nonbank lenders have some problems that need to be addressed.
He believes individual companies may run into trouble, but the probability of triggering a systemic financial crisis is "extremely low."
He stressed that everything related to AI comes with enormous uncertainty, and that "our economy has always had ups and downs — this is the essence of a market economy."
But Klaas Knot, former governor of the Dutch central bank, warned that according to the data he has, about one-third of the AI investment boom comes from debt financing, and off-balance-sheet commitments often mean this ratio is only a lower bound.
He said the possibility of a major market adjustment should not be underestimated, especially since U.S. consumption is quite sensitive to the wealth effect, and an adjustment could quickly have broader effects.
AI safety: Rubin calls for China-U.S. cooperation
Rubin said the issue "much bigger" than investment and growth risks is AI safety, and he had previously co-authored an article on the issue with former U.S. Treasury Secretary Henry Paulson.
He pointed out that open source makes AI technology accessible to more people, and the recursive self-improvement of artificial general intelligence (AGI) and even artificial superintelligence (ASI) is advancing rapidly.
This could lead to serious negative consequences, including various types of cyberattacks, with a fairly high probability, and there is currently no complete response.
Rubin stressed that both China and the United States should attach great importance to this and act together.
He also said that in addition to containing risks, such cooperation may help the two countries better realize AI's upside potential, but there is currently no specific mechanism design.
When discussing the White House political consensus behind U.S. technology restrictions on China, Rubin stressed that the United States and China need to move beyond this state to address AI safety issues, otherwise both countries will pay a huge price.
He believes the midterm election results are highly significant for the U.S. political environment, but their actual impact on policy direction is limited, because Congress is unlikely to act effectively in this area and related measures need to come from the executive branch.
In his view, the actions the government has taken so far are "far from enough" compared with the severity of the problem.
AI governance: how financial regulation can be implemented
At the Bund Roundtable "AI Governance in Finance: Key Issues and Feasible Paths," Zhang Wencai, managing director and chief administrative officer of the World Bank, Liu Jun, president of Industrial and Commercial Bank of China, Arthur Yuen, deputy chief executive of the Hong Kong Monetary Authority, and Lorenzo Bini Smaghi, chairman of the Group of 30, discussed responsibility attribution, regulatory capacity, inclusion and cross-border cooperation in AI governance.
Marshall Mills, IMF resident representative in China, moderated.
Zhang Wencai said a World Bank survey shows that only 11% of financial regulators in emerging markets are using AI regulatory tools, and about half say they do not have enough personnel and tools for effective supervision.
Regulatory capacity across economies is moving at two speeds: some are already at the frontier, for example using AI to prevent and detect fraud, while others have no tools at all, and there is also excessive reliance on a small number of cloud service providers and model providers.
He proposed three points: first, solidify the foundation and learn from rather than reinvent, with countries of limited capacity able to draw on existing international practices and guidelines, incorporate digital identity into data protection and payment systems, and direct limited resources to the highest-risk areas; second, help regulators use AI to supervise AI, which requires more interconnected data and staff who know how to use it; third, financial institutions must remain responsible for AI decisions even if the technology comes from elsewhere, and institutions handling cybersecurity and data protection should sit in the same room to coordinate and share regulatory tools, test results and early warnings across borders.
He also introduced the World Bank's newly released World Development Report, which proposes that AI could enable developing countries to achieve in a decade the growth that might otherwise take a century, and recommends using AI in the order of adoption, adaptation and advancement, with adaptation offering the greatest benefit to most countries: tools from high-income countries may not apply and need to be adjusted with local data and languages.
He said the World Bank advocates promoting the use of lower-cost small AI tools in developing countries and calls on developing countries to actively shape AI's development rather than passively receive it, to act quickly, build trust and leave no country behind.
Turning to financial inclusion, Zhang Wencai said AI can expand inclusion, but it needs to be designed from the start; inclusion will not happen automatically.
He cautioned against conflating the two: some people get digital loans but are still worse off than before; if models are biased, rules are unclear, or users cannot challenge decisions, and the data cannot reflect the real lives of women and informal workers, these groups will lose out.
In the World Bank survey, 59% of financial regulators in emerging markets listed fraud and scams as high risks facing consumers.
He proposed three safeguards: solidify the foundations of identity, interoperable payments and data governance; look at outcomes rather than only models, with regulators asking clearly who is approved, who is rejected and who pays higher costs, and explaining decisions in plain language with a simple appeals channel; and protect choice and competition to prevent a few platforms from controlling data, distribution and tools.
Liu Jun said AI governance faces an "identity dilemma" — is AI an actor with quasi-legal personhood, or purely a technical tool?
And how should responsibility be divided among algorithm developers, deployers, users and end users?
Under the algorithm black box, accountability and verification are both very difficult.
He also pointed to several mismatches: regulatory requirements call for full disclosure, while algorithm and data parameters are hard to disclose; data needs local storage, while data and algorithms themselves have no borders; countries find it hard to hand control of AI to other countries, giving rise to sovereign AI; and the banking industry is based on credit, yet it is difficult to give AI algorithms a credit rating.
He raised an idea: could AI be treated as a quasi-human legal entity and brought into a unified regulatory mechanism with existing market participants?
On inclusion, Zhang Wencai noted that "availability" does not equal "inclusion": if models are biased, terms are unclear and there is no appeals channel, borrowers may end up worse off.
He suggested regulators focus on outcomes rather than only models, such as who is approved, who is rejected and who pays higher costs, and provide customers with a human appeals channel.
Bini Smaghi offered a European perspective.
Asked whether Europe over-regulates AI, he said he interviewed Mistral, Europe's largest AI company, which believes current European regulation can be improved but is not the biggest problem; the real difficulty is scale, namely how to obtain sufficient capital while staying in Europe, and how to enter customer markets including public procurement.
He relayed the company's view that what they want is not deregulation but reciprocal market access.
He also warned that when both financial institutions and customers rely on agents to make decisions, market adjustments may be faster and more destructive.
He compared it to a game of musical chairs: if the agents are all smart enough and all want to judge before others when the music stops, they will all rush toward the same chair.
He said people know how to design pay and accountability incentives for humans, but do not know how to design incentives for agents smarter than people so that they value crisis lessons from years ago.
He admitted he has no answer, saying, "Perhaps we will only know after the next crisis."
Comments