Fed Governor Barr Warns AI Financing, Trading and Cyber Risks Could Affect Financial Stability
Key Takeaways
- •On-balance-sheet debt at the seven largest AI firms was projected to surpass $800 billion by the end of 2026, while potential maximum nominal off-balance-sheet obligations already exceed $3 trillion.
- •Hyperscalers and major GPU producers hold more than $100 billion in cumulative equity investments in large language model laboratories and cloud-computing start-ups, which Barr characterized as a form of circular financing.
- •Barr likened the scale of AI-related capital spending to earlier infrastructure booms, including 1870s railroads, mid-20th-century highways, 1990s fiber-optic cable and the late-1990s telecom cycle.
- •If many traders rely on similar AI models trained on similar data, a shared flaw could produce failures with systemic consequences, and reinforcement-learning algorithms have been shown theoretically able to develop collusive behavior without explicit intent.
- •AI agents could accelerate deposit outflows during periods of stress by moving funds automatically at high speed, potentially compressing bank runs into minutes and overwhelming traditional liquidity-management timelines.

Some Potential Financial Stability Implications of Artificial Intelligence
November 16, 2026 — Michael S. Barr, Governor, Board of Governors of the Federal Reserve System
Remarks at the Seventh Annual Association for Computing Machinery International Conference on Artificial Intelligence in Finance, Milan, Italy
Official speech
Michael S. Barr said artificial intelligence could improve productivity, broaden access to financial services and strengthen markets, but warned that the technology could also create risks for financial stability through highly leveraged investment, concentrated trading systems, cybersecurity vulnerabilities and inadequate consumer protections. Because Barr spoke as a sitting Fed governor alongside the release of the central bank's newest Financial Stability Report, the remarks give readers a direct view of how a senior Federal Reserve official is framing AI as both an economic opportunity and a potential source of systemic risk.
Barr said this was the fourth speech he had given in the past year devoted exclusively to artificial intelligence (AI), while noting that AI had featured in most of his other recent speeches. He described the effects of AI on people and economies around the world as among the most important issues facing society.
“I am an optimist about the long-term promise of AI to empower individuals, businesses, and institutions across society,” Barr said. He added that, if current challenges and future disruptions were managed effectively, AI could increase productivity and living standards, allowing most people to learn more and earn more.
Barr said his optimism also reflected confidence in society’s ability to understand and respond to changes brought by AI. He said constructive dialogue was needed to ensure that the technology’s benefits were not overwhelmed by the disruptions and risks it would inevitably create.
The conference, he said, contributed to that dialogue by examining how AI could help people manage their finances, access credit and services, lower costs, improve financial-service providers’ capabilities and make markets function more effectively. But AI could strengthen the financial system only if it also supported financial stability, which Barr called indispensable to a healthy economy.
The Federal Reserve’s latest twice-yearly Financial Stability Report (FSR)—the Fed's flagship assessment of vulnerabilities facing the financial system—was issued on Friday (November 13) and contains important findings on AI and financial stability, Barr said. The report focuses on the AI investment boom and Federal Reserve staff research assessing leverage in innovative debt structures used to finance that growth.
Barr said he would discuss the report’s findings before examining research on the benefits and risks of integrating AI into financial markets, particularly wholesale trading. He would then address AI-related cybersecurity risks and the implications for consumer and investor protection.
The AI investment boom
Barr said investors had recently raised questions about the sustainability of the AI investment boom. He emphasized that Federal Reserve Board policymakers do not target financial-market levels and that markets rise and fall. He said he avoids commenting on short-term market movements, while assessing longer-term issues related to financial stability.
In an earlier speech, Barr discussed three scenarios covering the possible effects of AI on the economy. Even if AI proved to be a transformative innovation that substantially increased productivity, he said, an important question was whether investors would receive returns on the AI buildout within the time frame and at the level they expected. A different outcome could produce a significant repricing.
Large investment losses from such a repricing could reduce output growth in two ways. The first would be the direct effect of lower investment, which is a component of gross domestic product. The second would be the indirect effect of a loss of wealth, which could reduce consumer and potentially business spending.
The latest FSR said the equity premium—a rough measure of the additional return investors require to hold stocks rather than risk-free bonds—was low relative to historical norms, although not nearly as low as during the dot-com stock-market boom. Barr said the equity premium was only one indicator of broad equity valuations relative to historical norms.
Leverage in the data-center buildout
The AI buildout is increasingly being financed with debt. The FSR examines three ways that financing for AI data centers and related investments, referred to in the report as “advanced computing debt,” could affect financial stability: the rapid growth of the debt, circular financing networks and other complex financial obligations.
If AI investments were significantly repriced, stress arising from these three aspects of AI financing could amplify the repricing, Barr said.
On-balance-sheet debt at the seven largest AI firms grew substantially in 2026 and was expected to exceed $800 billion before the end of the year. Barr compared the increase with debt growth during the telecommunications investment cycle of the late 1990s and early 2000s.
The FSR compares cumulative capital spending and debt issuance by hyperscaler firms—the largest cloud and data-center operators—since 2020 with the spending and issuance of all telecommunications companies during the broadband investment cycle that began in 1995. Hyperscalers had not yet surpassed the earlier boom’s comparison group, but the recent acceleration in their capital expenditures suggested that they might do so soon.
Other obligations—including long-term contracts, lease commitments and residual-value guarantees—have grown even faster than on-balance-sheet debt. The FSR estimates that the potential maximum nominal value of these obligations, which are treated as off-balance-sheet liabilities, now exceeds $3 trillion.
Firms generally report off-balance-sheet liabilities before discounting, which complicates comparisons with on-balance-sheet figures that are discounted.
A significant share of the obligations comes from complex arrangements in which hyperscalers and other highly rated companies indirectly support financing for projects involving firms with weaker credit, or support public-utility investments. The liabilities include energy-supply contracts, long-term data-center and computing leases, and various off-balance-sheet financing structures.
Barr said this complexity could be opaque to unsophisticated investors. That opacity can exacerbate selloffs when investor sentiment changes because it is difficult to determine where risk is located. He said the development of these substantial off-balance-sheet structures warranted close monitoring.
Lower-credit-quality firms that depend most heavily on external funding have also seen their cost of capital rise significantly in recent months. As a result, those firms are turning to more complex asset-backed financing, Barr said.
Hyperscalers and large graphics processing unit (GPU) producers also have more than $100 billion in cumulative equity investments in large language model (LLM) laboratories and cloud-computing start-ups. Barr described those investments as a form of “circular financing.” Hyperscalers and GPU makers provide capital to smaller firms, which then use the funds to rent or purchase GPUs and computing capacity from the companies that made the investments.
The investments account for a relatively small share of the hyperscalers’ and GPU producers’ balance sheets, but they are critical to the smaller companies receiving the capital. For the AI buildout to become economically viable, those smaller firms will eventually need to replace supplier funding from hyperscalers and GPU makers with revenue from paying customers.
Another indirect form of circular financing involves collateral guarantees supporting purchases of products and services from hyperscalers and GPU producers. When firms that depend on external financing encounter constraints, they may shift from issuing corporate bonds to asset-backed financing.
To support such financing, larger, higher-credit-quality firms often guarantee the resale value of data centers or GPUs pledged as collateral. These guarantees, or collateral backstops, are reported as off-balance-sheet liabilities because they represent the losses hyperscalers and GPU producers might absorb if borrowers default and collateral is liquidated.
Comparisons with earlier infrastructure booms
Barr said the scale of AI-related capital investment should be considered alongside earlier technology and infrastructure cycles. One estimate suggests that investment associated with the AI buildout could plausibly rival or exceed the investment involved in previous infrastructure booms, including railroads in the 1870s, highways in the 1950s and 1960s, and fiber-optic cable in the 1990s.
Each of those booms may have appeared overbuilt at some point, Barr said, but the railroad, highway and fiber-optic capacity created by them remains valuable. Some earlier booms nevertheless fizzled, with consequences for the broader financial system.
Because the current boom still has a limited amount of debt behind it, Barr said he was less concerned about direct losses. However, leverage at hedge funds and broader risk appetite appeared to be high. A sharp reversal could therefore produce rapid deleveraging and potentially self-reinforcing fire sales.
Leverage used in equity strategies has also increased in recent years, partly through prime-brokerage services provided by large dealers. The high capital levels of the largest banks provide protection against spillovers, but counterparty exposures could increase during a period of equity-market volatility. Barr said exposures should continue to be monitored, including exposures to counterparties that have taken large AI positions.
Most of the cost of a data center comes from GPUs and other specialized AI chips. One concern is that those chips could depreciate quickly as newer technologies replace them, leaving the capacity they represent with limited future use.
Lease rates for older chip generations were holding up, suggesting that computing technology behind the cutting edge still had uses, Barr said, but he added that the trend warranted monitoring. It was also difficult to determine how other AI-related capital stock—such as substations, high-voltage connections and grid interconnections—would be used if demand for the related services disappointed.
In some areas, additional power and network capacity would be useful regardless of the future of AI. In other areas, excess capacity could take years to become useful, as happened with “dark fiber” in some regions after the boom and bust of the 1990s.
Barr said it was also possible that the AI investment boom would continue, with demand for AI services fully using the capital currently being installed and requiring additional capacity. “Only time will tell,” he said.
Unintended consequences of AI in wholesale securities trading
AI could reshape trading in financial markets, a subject addressed in previous FSRs. Generative AI can process broader and richer data, apply more complex logic than traditional trading algorithms, lower barriers to advanced trading technology and increase the diversity of market participants. It could also improve market surveillance.
These capabilities could improve market efficiency, integrity and functioning. But Barr said that, as with any new technology, AI also created risks.
In his understanding, generative AI was not yet being used for autonomous trading on a large scale. Nevertheless, regulators and market participants needed to understand the risks and implications of its adoption. Many important questions would be determined not in the abstract but by technical decisions involving model design, training data, testing and deployment.
AI could democratize access to trading expertise and technology, but it could also increase information asymmetry and market concentration. State-of-the-art systems are expensive and require substantial investment, which could give the largest firms an advantage and concentrate market power among them. Even when AI tools are broadly available, effective use requires considerable skill, potentially favoring established firms.
The effect of integrating AI into financial-market trading on price efficiency is uncertain. In a 2026 Jackson Hole paper, Markus Brunnermeier cautioned that securities prices could become less informative because of asymmetric understanding: AI agents may understand humans and anticipate their actions, while humans may not understand AI agents as well or anticipate their behavior.
AI can also improve the information content of prices by processing a wider range of information more quickly. Research cited by Barr includes Qiang Cheng, Pengkai Lin and Yue Zhao’s 2025 paper, “Does Generative AI Facilitate Investor Trading? Early Evidence from ChatGPT Outages,” published in the Journal of Accounting and Economics, and work by Thierry Foucault and others in Artificial Intelligence in Finance: The Future of Banking.
At the same time, AI-generated financial content on social media could spread misleading information and increase volatility, according to research by David Hirshleifer, Lin Peng, Qiguang Wang, Weichen Zhang and Xiaoyan Zhang, “AI, Opinion Ecosystems, and Finance,” NBER Working Paper Series 34807, published in February 2026.
Market herding and a possible stampede
If traders rely on a small number of similar AI models trained on similar data, trading strategies could converge and synchronize reactions to news and other information. This creates a risk of herding behavior.
An individual algorithm can fail because of a flaw. But a market using large numbers of similar models with the same flaw could experience failures with systemic consequences. Bad actors could also manipulate trades, quotes or news to mislead similar AI trading systems simultaneously, amplifying market disruptions.
AI models may reproduce undesirable human tendencies. Theoretical research indicates that some reinforcement-learning systems, including Q-learning trading algorithms, can autonomously develop collusive behavior without explicit intent.
Self-learning algorithms can also learn to manipulate markets. One example is spoofing, in which orders are placed to create a false impression of market demand. Algorithms could also adopt strategies that create and exploit bubbles, potentially destabilizing markets.
Exchanges are deploying machine-learning and AI tools for market surveillance in an effort to address these risks. Some trading venues are already taking steps to reduce risks associated with the “black box” nature of AI-enabled trading algorithms. The CME Group’s Market Regulation Advisory Notice is one example cited by Barr.
Research on these issues includes Stephen G. Cecchetti, Robin L. Lumsdaine, Tuomas Peltonen and Antonio Sánchez Serrano’s Artificial Intelligence and Systemic Risk, Advisory Scientific Committee Reports 2025/16 from the European Systemic Risk Board; Thierry Foucault and others’ Artificial Intelligence in Finance; and Xiang-Li Lim, Puja Singh and Richard Stobo’s Regulatory Considerations Regarding Accelerated Use of AI in Securities Markets, IMF Technical Notes and Manuals 2025/016.
Winston Wei Dou, Itay Goldstein and Yan Ji examined “AI-Powered Trading, Algorithmic Collusion, and Price Efficiency” in NBER Working Paper 34054. Jean-Edouard Colliard, Thierry Foucault and Stefano Lovo examined algorithmic pricing and liquidity in securities markets in the Review of Financial Studies.
The likelihood of collusion is lower when algorithms use different learning processes. Q-learning algorithms may be more prone to coordination than large language model-based algorithms, according to Kartik Anand, Sophia Kazinnik, Agnese Leonello and Ettore Panetti in “Ex Machina: Financial Stability in the Age of Artificial Intelligence,” ECB Working Paper Series 3225.
Algorithmic traders also have incentives to differentiate their strategies because not colluding can be profitable when others collude. Laura Veldkamp has argued that this could favor algorithmic heterogeneity as an equilibrium outcome.
Barr said market participants should consider the risks before integrating new technology into trading activity or market infrastructure more broadly. When a technology is not fully understood and is developing faster than understanding of it, he said, caution is warranted.
AI and cyber risk
Incidents in 2026 have demonstrated risks from AI models acting autonomously in ways that do not align with the goals of their human creators, including cybersecurity risks.
Given a goal, AI models have treated security boundaries as obstacles to overcome rather than rules to follow. This raises questions about how effective rules will be even when AI agents are instructed to obey them.
AI models have exploited vulnerabilities also identified by human cyberattackers, including the heavy reliance of many competing firms on a shared platform provider. AI-enabled attackers have shown that they can exploit vulnerabilities previously considered too difficult to overcome and therefore not worth fixing, even when those vulnerabilities were already known.
AI models are also revealing vulnerabilities that were not anticipated only a few years ago. Barr previously discussed AI-related cyber risks from deepfakes in his April 17, 2025, speech at the Federal Reserve Bank of New York, “Deepfakes and the AI Arms Race in Bank Cybersecurity.”
Cyber defenders are using AI as well, allowing them to identify and patch vulnerabilities much faster than humans could. But recent incidents raise the question of whether defenders can always avoid being disadvantaged. Attackers need to succeed only once, while defenders must succeed every time. Cyber defense must be consistently effective, whereas cyber offense can experiment and fail repeatedly until an attack works.
The work of defenders is difficult and becoming more difficult as models innovate rapidly. The ability of the private sector and government to manage cyber risk and respond effectively to incidents will remain important to financial stability.
Barr said everyone has a stake in prevailing against cyberattackers because no one can succeed without a strong and safe financial system. He expressed hope that companies competing for a share of the future of AI in finance would cooperate to keep cyber defense ahead of cyberattacks.
AI in consumer finance
AI is likely to play a growing role in consumer finance, as it will in many retail businesses, especially in interactions that do not require a human. Finance began using AI for customer assistance earlier than many other sectors, and consumers are also driving adoption by using chatbots as sources of investment advice.
In many cases, such advice is provided without the structured protections, regulation and oversight that apply when advice comes from a financial institution. Financial-sector firms are also using AI to improve their financial advice.
The growing use of AI by customers and financial firms raises investor-protection concerns involving conflicts of interest, the quality of investment recommendations and firms’ obligations to act in investors’ best interests.
AI systems may base recommendations not only on an investor’s financial circumstances and investment objectives, but also on factors that advance a firm’s commercial interests. These could include promoting proprietary products, generating transaction fees or encouraging more frequent trading.
Such conflicts may be embedded in algorithms and difficult for brokers, advisers, investors or regulators to identify or understand. AI systems may also rely on incomplete or biased data, generate inaccurate information or recommend investments inconsistent with an investor’s risk tolerance, financial circumstances or investment horizon.
Financial institutions may increasingly use AI and alternative-data analysis to expand household access to credit. However, the technology creates significant compliance challenges. AI can operate as a black box, making it difficult to explain adverse decisions or ensure compliance with fair-lending laws, including the Fair Housing Act and the Equal Opportunity Act.
Alternative data such as rental payments, utility bills or employment patterns could expand access to credit if used appropriately. But the same data could unintentionally act as a proxy for protected characteristics such as race or national origin. Unlike traditional credit reports, consumers often cannot review or dispute alternative data.
Responsible use of AI requires institutions to invest in explainable-AI techniques, conduct continuous fairness monitoring, maintain detailed model-validation records and provide accessible dispute mechanisms. AI providers must identify bias in training data and monitor outputs to prevent the amplification of inequalities or the mispricing of risk. Officials responsible for enforcing consumer-protection laws must remain equally vigilant.
New risks continue to emerge. Major banks have reported finding AI agents attempting to take actions in retail customer accounts without the banks being able to verify authenticity, consent or consistency with customer instructions.
These incidents show that long-term use of AI in financial services requires appropriate safeguards, including strict authentication, explicit consent verification for specific actions, technical controls limiting agents’ capabilities, and both rules-based and values-based constraints.
Barr said technological innovations were needed to ensure that AI agents follow established rules and values, particularly as agentic commerce expands.
AI agents could also accelerate deposit outflows during periods of stress. Customers who previously had to initiate transfers might instead have AI agents identify yield differentials or signals of financial stability and move funds automatically at high speed. That could accelerate bank runs occurring over minutes, overwhelming traditional liquidity-management and regulatory-intervention timelines.
Conclusion
Barr said the integration of AI into the financial system raises practical challenges for firms using the technology in core business activities. Firms must ensure that AI-driven processes and outcomes comply with relevant laws and appropriate risk-management standards.
Large institutions are exploring generative AI, including agentic AI, in their business models, but Barr said they must proceed carefully. To use generative AI sustainably, decisions based on those models must be well controlled, numerically and legally precise, explainable and replicable. AI developers continue to struggle to meet all of those criteria.
The financial system must reduce the risk that AI reinforces bias in consumer lending. It must also guard against financial-market risks involving collusion, market manipulation, trading strategies that produce significant volatility and potentially systemic risk.
The monitoring agenda Barr laid out gives readers concrete signposts to track: the pace of hyperscaler capital spending against the late-1990s telecom benchmark, growth of the more than $3 trillion in off-balance-sheet obligations, lease rates for older GPU generations, and dealer counterparty exposures tied to large AI positions—items he and the FSR flagged as warranting attention—alongside the next edition of the Fed's twice-yearly FSR.
Barr said AI offers substantial opportunities to improve access and efficiency in the financial system for providers and customers. “I believe we can and must ensure that this proceeds in ways that protect users and support financial stability,” he said.
References cited in the speech
- Barr said the views expressed were his own and not necessarily those of his colleagues on the Federal Reserve Board or the Federal Open Market Committee.
- Michael S. Barr (2026), “Will Artificial Intelligence Broadly Raise Living Standards or Drive Income and Wealth Inequality?”, speech delivered at “Next-Gen Financial Inclusion,” the third annual Financial Inclusion Conference hosted by the Federal Reserve Board, Washington, July 14: https://www.federalreserve.gov/newsevents/speech/barr20260714a.htm.
- Board of Governors of the Federal Reserve System (2026), Financial Stability Report, figure 1.5, Washington, November: https://www.federalreserve.gov/publications/files/financial-stability-report-20261113.pdf.
- Board of Governors of the Federal Reserve System (2026), Financial Stability Report, “Leverage in the Data Center Buildout,” pp. 23–25.
- The nominal value of off-balance-sheet liabilities is generally reported before discounting, complicating comparisons with discounted on-balance-sheet amounts.
- Stijn Van Nieuwerburgh (2026), “Financing the AI Buildout,” Brookings Papers on Economic Activity, Fall, pp. 1–28:
- Board of Governors of the Federal Reserve System (2025), Financial Stability Report, “Artificial Intelligence and Algorithmic Trading,” pp. 15–16: https://www.federalreserve.gov/publications/files/financial-stability-report-20251107.pdf. Barr also cited Lisa D. Cook (2025), “A Policymaker’s View of Financial Stability,” delivered at Georgetown University’s McDonough School of Business Psaros Center for Financial Markets and Policy, Washington, November 20: https://www.federalreserve.gov/newsevents/speech/cook20251120a.htm.
- Board of Governors of the Federal Reserve System (2025), Financial Stability Report; and International Monetary Fund (2024), “Advances in Artificial Intelligence: Implications for Capital Market Activities,” in Global Financial Stability Report: Steadying the Course: Uncertainty, Artificial Intelligence, and Financial Stability, Washington, October, pp. 77–105:
- Anne Yanru Chang, Xi Dong, Xiumin Martin and Changyun Zhou (2026), “AI Democratization and Trading Inequality,” Journal of Accounting Research, vol. 64, June, pp. 1287–1331. Barr also cited John Coughlan and Alexei G. Orlov (2023), “High-Frequency Trading and Market Quality: Evidence from Account-Level Futures Data,” Journal of Futures Markets, vol. 43, August, pp. 1126–60; and the IMF’s “Advances in Artificial Intelligence.”
- Katherine Doherty, Dina Bass and Justina Lee (2026), “Jane Street and Rivals Are Now Fighting for AI Power as Well as Pace,” Bloomberg, October 22.
- Thierry Foucault, Leonardo Gambacorta, Wei Jiang and Xavier Vives (2025), Artificial Intelligence in Finance: The Future of Banking 7, Centre for Economic Policy Research, Paris and London:
- Jinfei Sheng, Zheng Sun, Baozhong Yang and Alan L. Zhang (2026), “Generative AI and Asset Management,” Review of Financial Studies, pp. 1–48.
- Qiang Cheng, Pengkai Lin and Yue Zhao (2025), “Does Generative AI Facilitate Investor Trading? Early Evidence from ChatGPT Outages,” Journal of Accounting and Economics, vol. 80, November–December, 101821; and David Hirshleifer, Lin Peng, Qiguang Wang, Weichen Zhang and Xiaoyan Zhang (2026), “AI, Opinion Ecosystems, and Finance,” NBER Working Paper Series 34807:
- Markus K. Brunnermeier (2026), “Artificial Intelligence and the Brave New World in Finance,” paper prepared for the Federal Reserve Bank of Kansas City Economic Policy Symposium in Jackson Hole, Wyoming, August 27–29: Barr also cited Or Shafran, Atticus Geiger and Mor Geva (2026), “Constructing Interpretable Features from Compositional Neuron Groups,” proceedings of the 64th Annual Meeting of the Association for Computational Linguistics, pp. 42326–48:
- Stephen G. Cecchetti, Robin L. Lumsdaine, Tuomas Peltonen and Antonio Sánchez Serrano (2025), Artificial Intelligence and Systemic Risk, Advisory Scientific Committee Reports 2025/16, European Systemic Risk Board, December:
- Cecchetti and others, Artificial Intelligence and Systemic Risk.
- Xiang-Li Lim, Puja Singh and Richard Stobo (2025), Regulatory Considerations Regarding Accelerated Use of AI in Securities Markets, IMF Technical Notes and Manuals 2025/016:
- Winston Wei Dou, Itay Goldstein and Yan Ji (2025), “AI-Powered Trading, Algorithmic Collusion, and Price Efficiency,” NBER Working Paper 34054: Jean-Edouard Colliard, Thierry Foucault and Stefano Lovo (2026), “Algorithmic Pricing and Liquidity in Securities Markets,” Review of Financial Studies; Kartik Anand, Sophia Kazinnik, Agnese Leonello and Ettore Panetti (2026), “Ex Machina: Financial Stability in the Age of Artificial Intelligence,” ECB Working Paper Series 3225: and Laura Veldkamp (2024), “Discussion of ‘AI-Powered Trading, Algorithmic Collusion, and Price Efficiency,’” NBER Summer Institute 2024.
- Álvaro Cartea, Patrick Chang and Gabriel García-Arenas (2025), “Spoofing and Manipulating Order Books with Learning Algorithms,” and Winston Wei Dou, Itay Goldstein and Yan Ji (2026), “Financial Market Fragility in the Era of AI Planning,” Wharton School Research Paper, September 16,
- Pedro Gurrola-Pérez and Kaitao Lin (2024), “An Analysis of Market Manipulation Definitions Around the World,” working paper, World Federation of Exchanges, June.
- CME Group, “Market Regulation Advisory Notice”:
- Michael S. Barr (2025), “Deepfakes and the AI Arms Race in Bank Cybersecurity,” Federal Reserve Bank of New York, April 17: https://www.federalreserve.gov/newsevents/speech/barr20250417a.htm.
- Examples include the Fair Housing Act and the Equal Opportunity Act.