Financial markets are currently undergoing a structural revolution defined, not merely by the speed of execution but by the level of autonomy ceded to machine intelligence. This era, characterised by the ‘AI liquidity wars’, describes the intense, competitive pressure among self-learning systems striving to provide the deepest, most capital-efficient liquidity across every global asset class. Its transition mandates an assessment of whether these new self-learning networks will ultimately stabilise or destabilise the next generation of global financial systems. And the scale of this shift is profound - by 2025, artificial intelligence is predicted to handle almost 89% of the world’s trading volume, signifying a fundamental transition of control away from direct human instruction. These sophisticated AI agents utilise advanced techniques (e.g. machine learning and natural language processing) to analyse market trends, price movements, sentiment and liquidity in real time. They automatically execute trades, rebalance assets and optimise complex decentralised finance (DeFi) strategies without manual intervention.
Notably, Coinbase’s latest recent strategy marks a decisive attempt to blur the boundary between crypto exchanges and traditional stock markets. By adding equities, derivatives, prediction markets and a clear roadmap toward tokenised stocks, Coinbase is positioning itself less as a specialist venue and more as an always-on financial utility. For traditional exchanges, the challenge is structural rather than competitive on price: Coinbase offers continuous engagement, integrated wallets and on-chain settlement rails that operate and function 24/7. The potential for both stability and fragility arises out of the transition from automated execution to autonomous strategy generation. Traditional algorithmic trading was rule-based, meaning humans set the parameters and retained control over the strategy, delegating only 80% of the execution; conversely, an autonomous AI system does closer to 99% of the work, learning from past data and developing its own emergent strategies. This transference of control means the market’s responsiveness becomes non-linear and emergent - the risk shifts from simple error in code logic to systemic failure caused by the collective, uncoordinated behaviour of machines. This competitive necessity is driving rapid, universal integration: the global market for AI trading is projected to reach $35 billion by 2030, compelling market participants to adopt these high-performance models quickly, even as regulators monitor the emergent concentration risks.
The key technology driving this efficiency is deep reinforcement learning (DRL), which enables agents to learn optimal behaviour through interaction with complex, dynamic environments such as the limit order book. In traditional finance (TradFi) high-frequency trading (HFT), DRL frameworks have demonstrated measurable superiority over conventional machine learning models. A DRL model that integrates convolutional and recurrent neural networks to process real-time market signals has achieved enhanced prediction accuracy and market adaptability, demonstrating a verifiable 62.8% win rate and a profit factor of 2.4 in comparative analyses. This technology is inherently suited for complex financial problems such as autonomous market making, as DRL can be designed with customised reward functions specifically to control for inventory risk - the danger of accumulating an unfavourable position which is a primary challenge for human or simple algorithmic market makers. And the impact of autonomous agents is equally transformative in the DeFi landscape; DRL agents, often employing algorithms such as proximal policy optimisation, are now essential for optimising liquidity provisioning in concentrated liquidity automated market makers (CLAMMs), such as Uniswap v3. These systems dynamically adjust liquidity positions using real-time price dynamics to balance fee maximisation against the inherent risk of impermanent loss, a task too computationally intensive for typical retail liquidity providers.
This successful mitigation of impermanent loss through autonomous management is a primary catalyst for institutional capital flowing into decentralised systems. By effectively addressing this core technical risk, DRL models significantly lower the barrier to entry for large, sophisticated capital providers by means of fostering deeper and more stable liquidity pools in DeFi. Furthermore, the rapid reduction in computational cost for advanced AI models, with inference costs for systems such as GPT-3.5 dropping over 280-fold between 2022 and 2024, democratises this high-performance trading capacity. If these highly profitable DRL models become ubiquitous and affordable, the total number of autonomous, complex actors operating across markets will surge whereby increasing the difficulty of systemic regulation. The shift in capabilities can be summarised by contrasting the old algorithmic approach with the new AI paradigm:
AI Agents vs. Algorithmic Trading
Source: Forbes, Arxiv, Emanresearch
Proponents of AI-driven markets argue that autonomous liquidity management creates a more robust, unified and efficient financial ecosystem. In TradFi, investments to create blockchain-powered platform are typically motivated to improve liquidity and post-trade cost efficiencies. Treasury functions are becoming automated through solutions such as Just-in-Time Funding which will eventually render the role of funding accounts to bots, freeing human treasurers to focus on higher value-add tasks. This institutional engagement forms a powerful “liquidity flywheel”. The convergence of institutional custody, stablecoin rails and prime brokerage services enhances capital efficiency and accelerates global adoption across centralised spot markets. For example, modern custody solutions leverage specialised networks to allow institutional clients to trade on exchanges instantly while assets remain in secure custody. This infrastructure aims to bridge the reliability of traditional financial infrastructure with the revolutionary possibilities of digital assets, allowing market participants to manage both through a unified workflow. Furthermore, central stability institutions are actively leveraging AI as a defensive mechanism; the Bank for International Settlements (BIS) has confirmed that AI supports financial stability analysis by monitoring market anomalies during periods of low liquidity and market dysfunction. This ability to proactively monitor emergent vulnerabilities establishes an AI-vs-AI dynamic, indicating that the instability caused by autonomous agents may only be solvable through equally sophisticated AI surveillance tools. However, this appearance of stability masks a structural risk. The operational centralisation of liquidity, where trades are executed across multiple venues but settled through a limited number of specialised prime brokerage networks, creates a single point of concentration. If the sophisticated AI controlling asset allocation within that critical network were to fail, the entire system of interconnected liquidity could be instantaneously disrupted.
Yet, despite the gains in efficiency, the core concern remains the fragility paradox: individual optimisation by autonomous agents leads directly to collective failure when faced with stress. Historical precedents, such as the 2010 Flash Crash, serve as stark reminders of how quickly markets can spiral in the digital age. That event demonstrated how small data errors or uncoordinated reactions by high-speed algorithms can unintentionally amplify volatility, leading to a sudden and extreme market decline. Likewise, the contemporary threat is more sophisticated: the International Monetary Fund (IMF) has cautioned that algorithmic trading systems are often programmed with safety mechanisms that trigger de-risking or complete shutdowns when facing unprecedented price movements. Whilst these safeguards are intended to protect individual firms, the simultaneous activation of these ‘self-preservation’ protocols across numerous institutional agents could create destabilising feedback loops, so resulting in a devastating evaporation of market liquidity. This fragility is observed even in cornerstone markets. Liquidity in the massive US Treasury market, which underpins global finance, briefly deteriorated in April 2025 in response to geopolitical and trade policy uncertainty therefore underscoring the system’s sensitivity to external shocks. When AI agents are deployed by non-bank financial institutions (NBFIs), which already exhibit structural vulnerabilities related to excessive leverage and liquidity mismatch, these flaws risk being amplified instantaneously.
Certainly, the collective failure mechanism is engineered - since DRL agents are trained to optimise for individual profitability and instantaneous risk mitigation, their optimal strategy during a stress event is to liquidate assets or withdraw liquidity immediately. If thousands of autonomous agents execute this profit-preserving impulse simultaneously, the resulting vacuum of liquidity can cause cascading failures across systems. Furthermore, the convergence of TradFi and DeFi means that a liquidity crisis originating in a critical stablecoin pool could trigger automated margin calls in a connected institutional prime brokerage system, leading to cross-market failures and systemic contagion. The potential for autonomous systems to cause systemic instability has spurred urgent regulatory and self-governance responses globally; financial stability authorities are concentrating on structural concentration risk. The Financial Stability Board (FSB) is actively monitoring vulnerabilities arising from financial institutions’ reliance on a few critical third-party AI providers, focusing on criticality, concentration and substitutability risks in the AI supply chain. This focus is crucial because the proprietary DRL models are often opaque, meaning regulating the technology developer is sometimes more effective than regulating the complex, emergent market behaviour; regulators are responding with landmark legislation designed to establish technical resilience. The Central Bank of Ireland highlights that significant regulatory developments shaping 2025 and beyond include the EU’s Digital Operational Resilience Act (DORA), the Markets in Crypto Assets Regulation (MiCA) and the overarching European Artificial Intelligence Act. Essentially, these frameworks are intended to establish baseline standards for operational resilience and transparency across the sector.
In DeFi, oversight is being addressed through native, decentralised governance mechanisms. AI agents are being integrated into decentralised autonomous organisations (DAOs) to enhance governance by analysing complex proposals and forecasting outcomes, streamlining decision-making. These agents can even vote on behalf of users based on predefined criteria, ensuring efficient community interests are prioritised. Crucially, AI agents are also acting as security layers, monitoring smart contracts and detecting anomalies in real time that could signal malicious activity or exploits, thereby protecting user funds and the integrity of the decentralised systems. The divergence between these two approaches - i.e. mandated, centralised compliance in TradFi versus transparent, on-chain oversight in DeFi - presents a complex future harmonisation challenge. The effectiveness of global stability efforts will depend on successfully merging these fundamentally different regulatory philosophies to ensure that an autonomous agent operating seamlessly across both domains adheres to a consistent standard of systemic resilience. The AI liquidity wars have permanently redefined the operating landscape of global finance. Autonomous deep reinforcement learning agents are delivering unprecedented efficiency and enabling massive capital aggregation across increasingly integrated TradFi and DeFi structures - and this self-optimising capacity is the new baseline for market function.
However, the core challenge is not technology, but co-ordination. The inherent tendency for individual, profit-optimising agents to collectively and simultaneously withdraw liquidity during stress creates the synthetic flash crash, an engineered vacuum of capital. The stability of future markets is not an innate property of the AI, but a function of engineered resilience. Success depends on the rapid and robust implementation of international frameworks such as the EU’s DORA and AI Act, which aim to mandate transparency and operational standards at the source of the technology. Simultaneously, the efficacy of native, AI-driven security and governance mechanisms within the DeFi ecosystem will be critical to preventing the uncoordinated flight of capital that defines systemic failure. The battle is no longer for speed, but for coordinated, resilient autonomy. The future is automated, and it must be safe.
This article first appeared in Digital Bytes (13th of January, 2026), a weekly newsletter by Jonny Fry of Team Blockchain.