The deployment of autonomous AI agents in financial services presents a material risk to systemic stability, according to analysis from Apollo Global Management and new academic research. Individually rational optimization by AI agents—such as seeking higher yields or executing trades—can aggregate into destabilizing collective behavior, including rapid capital outflows from banks and flash crashes in securities markets. This emergent risk profile requires financial institutions and regulators to shift focus from individual agent performance to the systemic integrity of AI-driven ecosystems.
1. Key Developments
Torsten Slok, chief economist at Apollo Global Management, warned on September 28, 2026, that AI agents could systematically drain low-cost bank deposits by autonomously reallocating funds to higher-yielding assets CoinDesk. This scenario, termed an ‘agentic bank run,’ could operate at a speed and scale exceeding human-driven events, creating significant liquidity challenges for traditional banking institutions Finextra.
Supporting this view, academic research published on arXiv models the mechanisms for such failures. A paper titled “Financial Fragility in Societies of LLM Agents” introduces a model demonstrating how individually protective decisions by LLM agents can lead to collective financial fragility arXiv:2609.30940. A separate study, “Agentic Limit Order Books,” examines markets populated exclusively by autonomous trading agents, identifying how their interactions can cause abrupt phase transitions in order flow and market stability, heightening the risk of flash crashes arXiv:2609.31260.
2. Strategic Implications
The primary financial exposure is concentrated in institutions reliant on substantial, low-cost retail and commercial deposits. An agent-driven exodus of this capital would directly increase funding costs and compress net interest margins. For asset managers and exchange operators, the key risk is heightened market volatility. The potential for agent-driven flash crashes could erode investor capital and undermine confidence in market price discovery mechanisms.
Conversely, this risk landscape creates a market for advanced AI governance and security solutions. Technology vendors are responding to this demand. For example, Feedzai launched its Farol agent on September 25, 2026, to improve fraud detection and reduce investigation latency for banks Feedzai. Financial institutions investing in robust internal data architecture and AI guardrails, such as Capital One, may be better positioned to mitigate these systemic risks and maintain a competitive advantage in deploying AI securely Capital One.
3. What to Watch (12–18 months)
Over the coming 12-18 months, decision-makers should monitor regulatory bodies for the issuance of frameworks governing AI agent deployment in critical financial functions. Expect formal proposals on AI safety standards, including requirements for systemic stress testing. Track the operational results of early adopters of specialized risk agents, such as those from Feedzai, to gauge their efficacy in production environments. Finally, observe research from adjacent sectors, like CGI Federal’s work on agentic AI for supply chain resilience, for methodologies that could be adapted to financial risk management CGI Federal.
Executive Summary
- Strategic Imperative: Autonomous AI agents, acting rationally on an individual basis, can create systemic financial instability, including bank runs and market crashes, when their actions aggregate.
- Primary Near-Term Risk: An ‘agentic bank run,’ where AI agents rapidly move capital from low-yield bank deposits to higher-return assets, threatening bank liquidity and funding models.
- Capital Markets Impact: In capital markets, the proliferation of autonomous trading agents could lead to abrupt shifts in liquidity and order flow, increasing the probability of flash crashes.
- Commercial Opportunity: A commercial opportunity exists for specialized AI governance and risk mitigation technology; firms providing these solutions are positioned for growth as institutions seek to manage deployment risks.
- Immediate Executive Action: Immediate executive action is required to audit all current and planned AI agent deployments, establish robust governance frameworks with real-time monitoring, and engage in shaping regulatory standards.
Projected AI Agent Penetration in Key Financial Functions (2027-2029)
5%
12%
25%
20%
35%
50%
30%
45%
60%
10%
18%
30%
15%
25%
40%
Source: Internal analysis based on market trends and academic modeling.