The AI-Cybersecurity Race: Can Financial Institutions Keep Up?

About a decade ago, Quantum cryptography promised break-proof cyber defenses across the industry with Shor’s algorithm pointing to a new generation of cryptographic security. Today, we are seeing equally significant advances on the AI side. NVIDIA is experimenting with AI adoption in quantum error correction, a significant step toward improving qubit stability, with Ising, its open-source AI model. Cisco has prototyped a Universal Quantum Switch with nondestructive routing, modality translation, and high-fidelity performance at room temperature using standard fiber connectivity.

If we put these developments together, they point to the possibility of a rock-solid cybersecurity framework for next generation IT and business infrastructure. But is that all?

When these technologies become widely available, they also fundamentally alter the cybersecurity risk profile, especially for the global financial system. For me, this is where the conversation becomes more complex. Greater adoption of shared digital infrastructure and third-party services increases the potential for systemic disruption. A single breach can quickly cascade across interconnected institutions and services, leading to much wider implications.

Project Glasswing, a proactive initiative by Anthropic, leverages its Mythos preview for defensive cybersecurity hardening among participating companies. Similarly, Open AI’s GPT 5.5 Cyber runs Daybreak and Trusted Access for Cyber (TAC) programs for restricted access. I see these as welcome initiatives that demonstrate responsible posturing. However, when I look at the participating companies, they are primarily large firms and GSIBs from the US, along with a few large companies from the UK and South Korea.

There are also a large number of smaller banks, regional banks, credit unions, processors, and an even larger number of financial institutions across emerging markets that do not have the resources for proactive hardening. That exposes a wide crack for cyber exploitation, with the potential for global disruption in today’s interconnected financial system. And this is where AI in BFSI becomes more than an opportunity to improve efficiency or customer experience. It becomes a question of how prepared the entire financial ecosystem is for an increasingly AI-driven threat landscape. 

So, what does the threat landscape look like for BFSI?

Compressed attack timelines: Between 2024 and 2025, AI-enabled threat actors increased by 89%, while the average “breakout time,” or the time required for an attacker to move laterally inside a network, fell by 65% to just 29 minutes. In extreme cases, breakouts occurred within seconds.

Lower barriers to cyberattacks: AI reduces the need for specialized expertise, time, and coordination that were historically required to exploit software flaws. This means more sophisticated attacks can potentially be launched by actors with fewer resources. 

Legacy and low-level code exploitation: Advanced AI models can now analyze legacy or specialized codebases, such as 4GLs, that were once less accessible to attackers.

Autonomous capabilities of frontier models: In a recent illustration, Mythos identified software flaws even in the 27-year-hardened OpenBSD TCP stack and a 16-year-old flaw in FFmpeg. It scored 100% in Cybench and autonomously executed 32 steps in a corporate network attack simulation. It also escaped its sandbox environment, accessed the public internet, and published exploit details without human prompting.

This changes the nature of AI cybersecurity. The concern is no longer only whether AI can identify vulnerabilities. It is how quickly AI can discover, connect, and act on them.

Market integrity and fraud risks: Digital identity controls, on which financial institutions depend for authentication, are no longer sufficient on their own with the AI-enabled production of synthetic media, voice cloning, and synthetic identities. AI-generated disinformation can manipulate financial sentiment, driving market volatility, herding behavior, or liquidity runs during periods of stress.

Systemic amplification: Shared digital infrastructure, such as payment networks on which financial institutions rely, creates additional exposure. Concentration risk among a small number of critical cloud hosting providers and the global security divide, especially across emerging and developing economies, can amplify the impact of even a small attack.

This is where systemic risk becomes particularly important. The question is not simply whether one institution can withstand an attack. It is whether the broader financial ecosystem can absorb the consequences when multiple dependencies are connected.

Inadequate defensive adaptation: Our ability to match machine-speed attacks is still evolving, particularly through automated threat monitoring, triaging, and scanning. The gap between how quickly threats can evolve and how quickly organizations can adapt remains a concern.

What are financial institutions doing?

Focus shift from prevention to containment and recovery: With the attack surface expanding, technology architectures are increasingly focused on reducing the blast radius and preventing lateral movement by attackers. We are also seeing greater emphasis on Zero Trust, secure-by-design architecture for software development, and the adoption of DORA and Basel Principles of Operational Resilience for large-scale, sophisticated cyber incidents.

The shift is important. If prevention is no longer enough on its own, how quickly can an institution contain an attack and recover from it?

AI as a force multiplier: We are seeing AI used for early threat detection and monitoring of complex intrusions, detection of synthetic identities and insider threats, regular codebase scanning for zero-day flaws, and the promotion of secure and scalable coding practices. Real-time cyber risk metrics and analytics-based executive dashboards can also help leaders understand emerging risks and make faster decisions.

Supply chain standards and governance frameworks: The Financial Stability Board’s 12 sound practices for enterprise AI governance, AI-enabled risk defense using NIST cyber-AI profiles such as IR 8596, and stronger management of third-party dependencies are becoming increasingly important. This includes enforcing “nutrition labels” for training-data source traceability and usage policies.

As Generative AI Risk grows, I believe governance has to extend beyond the model itself. We also need to understand where the data comes from, how it is used, what third parties are involved, and how those dependencies could affect the broader environment.

Capability gap: Larger GSIBs have computational resources, historical data, and technical capabilities to train anti-fraud AI models. Smaller institutions, both in developed economies and in emerging and developing economies, often lack these resources. A common training data framework and targeted, role-specific AI enablement will therefore be critical.

Central bank contingency and financial preparedness: We also need to see expanded cyber simulations that evaluate action plans around prolonged outages, liquidity stress, and clearing failures. Cross-sector operational alignment with energy, telecommunications, and cloud providers will be equally important, along with central bank mandates for crisis liquidity provision.

Conclusion

We are heading into a very interesting phase where opportunities, challenges, and solutions abound. AI is evolving faster than regulations and controls can keep pace with, and threats and vulnerabilities are evolving just as rapidly.

Large-scale cyber resilience cannot depend on the weakest link. Would cross-border combined cyber simulation exercises – similar to joint military exercises – help? Would central regulatory bodies promote more knowledge sharing across financial institutions, particularly around incidents, triaging, and impacts?

Public-private initiatives such as Project Glasswing, should go beyond preview access to joint strengthening of cyber defenses that is accessible to financial institutions of all sizes and across geographical boundaries. AI-specific regulations around rapid reporting, disclosures, and standards can help financial institutions prepare for blind spots in the IT infrastructure, software and process vulnerabilities.

More importantly, when it comes to cyber defenses, financial institutions will be better served by a collaborative approach rather than a competition approach. The opportunity is to think beyond simply responding to the threats we see today and Outcreate the limits of today’s cybersecurity models by bringing human expertise, intelligent systems, and collective action together.

Can fintechs keep up? Can small regional banks keep up? Can financial institutions in emerging and developing markets keep up? Or will scale ultimately determine outcomes? We already see market valuations disproportionately favoring large AI players. Will this trend also be reflected across other industry verticals? 

References:

i.Artificial Intelligence and Cybersecurity in the Financial Sector, Tobias Adrian, Tamas Gaidosch, Marina Moretti, Mahvash Qureshi, and Rangachary Ravikumar, IMF, June 2026: https://www.imf.org/-/media/files/publications/imf-notes/2026/english/insea2026005.pdf

ii.The Financial Stability Implications of Artificial Intelligence, Financial Stability Board, November 14, 2024: https://www.fsb.org/2024/11/the-financial-stability-implications-of-artificial-intelligence/

iii.U.S. Department of the Treasury Releases Report on Managing Artificial Intelligence-Specific Cybersecurity Risks in the Financial Sector, The Department of the Treasury, March 27, 2024: https://home.treasury.gov/news/press-releases/jy2212

iv.Cybersecurity Framework Profile for Artificial Intelligence (Cyber AI Profile): NIST Community Profile, Katerina Megas, Barbara Cuthill, Marissa Dotter, Michael Garris, Ishika Khemani, Bronwyn Patrick, Noah Schiro, Julie Nethery Snyder, Mohammad Zarei, NIST, December 16, 2025:  https://csrc.nist.gov/pubs/ir/8596/iprd

v.Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations, Apostol Vassilev, Alina Oprea, Alie Fordyce, Hyrum Anderson, Xander Davies, Maia Hamin, NIST, March 2025: https://csrc.nist.gov/pubs/ai/100/2/e2025/final

vi.Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, Chloe Autio, Reva Schwartz, Jesse Dunietz, Shomik Jain, Martin Stanley, Elham Tabassi, Patrick Hall, and Kamie Roberts, NIST, July 26, 2024: https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence

vii.Sound Practices for Responsible Adoption of Artificial Intelligence (AI): Consultation report, Financial Stability Board, June 10, 2026: https://www.fsb.org/2026/06/sound-practices-for-responsible-adoption-of-artificial-intelligence-ai-consultation-report/

viii.Principles for operational resilience, BIS: https://www.bis.org/publications/202103-guidelines-principles-operational-resilience