Is Traditional Banking Prepared to Defend Against AI?

• AI changes the speed and frequency with which a vulnerability can be explored.
• The risk also extends to the technology providers of financial institutions.

The hacking of Coldcard hardware wallets has once again put security at the center of the conversation about bitcoin (BTC). However, the episode left an important distinction: the Bitcoin protocol was not compromised. The attack targeted a tool used by users to safeguard their funds.

This nuance serves to broaden the perspective. Artificial intelligence (AI) is increasing the capacity to detect and exploit vulnerabilities, and that risk does not belong solely to the cryptocurrency ecosystem. Traditional banking also depends on applications, software, technology providers, and connected systems that can become entry points.

The difference lies in the scale of the consequences. A vulnerability in a device can affect its users. In the financial system, where numerous banks share providers, cloud services, and technologies, a single flaw can simultaneously impact several institutions and turn into a financial stability issue. This is the thesis that various organizations have begun to warn about in recent months.

In May, U.S. Treasury Secretary Scott Bessent brought the concern directly to Wall Street. After meeting with Federal Reserve (Fed) Chairman Jerome Powell and executives from major U.S. banks to discuss the capabilities of advanced artificial intelligence models, he was asked whether citizens should be worried about the possibility of these tools being used to hack bank accounts. His response was brief: "They should be."
U.S. Treasury Secretary Scott Bessent. Source: @NBCNews -- YouTube.

Bessent also explained that there has been a significant leap in the capabilities of large language models and raised the need to find a balance between maintaining the technological leadership of the United States and protecting its financial system.

The problem is no longer finding a flaw, but doing so before {#h-the-problem-is-no-longer-finding-a-flaw-but-doing-so-before}

A report published in June by researchers from the International Monetary Fund (IMF) helps to understand why the warning goes beyond an individual bank account.

According to the organization, the main change introduced by AI does not necessarily consist of inventing new modes of attack. What changes is the speed, frequency, and breadth with which vulnerabilities can be discovered and potentially exploited.

This factor reduces an advantage that defenders have had for years: time.

The IMF cites data from CrowdStrike indicating that the activity of AI-assisted attackers increased by 89% between 2024 and 2025, while the average time they needed to move within a network once access was gained fell to 29 minutes, a reduction of 65%. In the most extreme cases, that movement occurred in seconds.

Here lies one of the greatest difficulties for banking. Their systems may be closed to the public, but they are not isolated. Banks, markets, and payment systems depend on cloud platforms, operating systems, shared software, and technology providers.

In this regard, the IMF warns that this architecture creates a "correlated exposure." If several institutions use the same vulnerable component, AI can facilitate the identification and attack of that weakness across multiple institutions within a very short period.

That is to say, the risk is no longer just that an attacker manages to breach a single bank. The real problem arises when the same door exists in many banks at the same time.

The graph shows that vulnerabilities are being exploited faster and before there is room to react (zero-day). Since 2018, the points have shifted towards fewer days and a greater proportion of attacks: as of 2026, the rate exceeds 70%, while the average exploitation time approaches zero.
The horizontal axis indicates the average time, in days, that it takes for a flaw to be exploited. Source: International Monetary Fund (IMF).

Wall Street is already looking for cracks {#h-wall-street-ya-esta-buscando-las-grietas}

The warnings are not purely theoretical. Major U.S. banks have begun using advanced AI tools to review their own systems.

Reuters reported in May that entities such as JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley gained access to Mythos, an Anthropic model specialized in cybersecurity, as explained by CriptoNoticias. Tests began to reveal hundreds or even thousands of low and moderate risk vulnerabilities that the banks needed to review more quickly than usual.

The problem was not necessarily finding a single critical flaw. Mythos showed the ability to connect several minor vulnerabilities to turn them into a more dangerous attack vector, something that could take much longer for a human team.

🤖 Anthropic's AI Claude Mythos discovered vulnerabilities in HAWK, a post-quantum digital signature algorithm currently being evaluated by NIST. Although it does not compromise current systems, it anticipates the role of AI in the sector. pic.twitter.com/XNYCx75tqf --- CriptoNoticias (@CriptoNoticias) July 29, 2026

The IMF's own report uses the model as an example of the new scenario. In tests, Mythos identified thousands of high-severity zero-day vulnerabilities and autonomously completed a 32-step corporate attack simulation, according to documentation cited by the organization.

AI is thus showing its two faces. The same capability that would allow finding a vulnerability before a bank can also enable the bank to find it before the attacker. And that’s where the race begins.

The banking sector is also building its defense with AI {#h-la-banca-tambien-esta-armando-su-defensa-con-ia}

The Financial Stability Board (FSB) reaches a similar conclusion. In its report on the responsible adoption of artificial intelligence in financial institutions, the organization notes that these technologies can increase "the ease, frequency, and impact of cyberattacks," while also enhancing institutions' ability to defend themselves.

Therefore, the FSB proposes twelve best practices for incorporating AI within the financial system, including incorporating scenarios of AI-assisted attacks in security tests, strengthening information sharing, and using AI itself to manage cyber risks.

Another sensitive point emerges: the providers.
The use of AI enables hackers to accelerate their attacks. Source: Image generated with Grok.

Many institutions rely on the same companies for artificial intelligence services, cloud infrastructure, or critical software. The IMF warns that this concentration can create single points of failure. A problem with a widely used provider could quickly spread among various entities.

The FSB, for its part, recommends that banks especially assess the quality of data, business continuity, supply chain, and concentration when contracting third-party AI services.

Bitcoin and banking have distinct vulnerabilities {#h-bitcoin-and-banking-have-distinct-vulnerabilities}

Coldcard allows for a more precise return to the starting point. The attack did not reveal a weakness in the Bitcoin protocol.

In banking, something comparable occurs, although with a completely different architecture. An attacker also does not need to compromise the heart of the entire financial system to cause harm. They can find a flaw in a provider, an application, a credential, or a component used by numerous entities.

This difference is that the strong interconnection of traditional finance can amplify the consequences. The IMF warns that AI is elevating cyber risk from an operational issue to a central question for financial stability, precisely due to the combination of shared infrastructure, common providers, and propagation capacity.

The question, therefore, is not whether a closed system is automatically protected against AI. Nor whether AI inevitably favors the attacker.

The race is on to see who can find and fix the next vulnerability first. And when the money of millions of people depends on systems that increasingly share technology, being a few minutes late can start to be too much.

Tags: Bitcoin (BTC) Artificial Intelligence (AI) Latest Prices and Trading