Wall Street invests trillions in AI, but increases purchase of insurance against possible market collapses

Despite growing discussions about artificial intelligence (AI) in the modern workplace, most financial institutions use this technology significantly less often than companies in the broader business sector.

According to a study by Kaspersky, a global cybersecurity and digital privacy firm, only 23% of banking, financial, and insurance (BFSI) companies are applying AI for software programming. This figure is substantially lower than the industry average of 34%, representing an 11 percentage point lag.

This gap indicates that while major technology providers actively promote generative coding assistants such as Cursor and GitHub Copilot, financial institutions exercise much greater prudence, and this has serious justification.

Operational stakes in the financial sector are much higher than many outside this sphere might appreciate. A single software failure deployed in high-frequency trading pipelines or core ledgers can trigger catastrophic system failures, serious capital calculation errors, or critical regulatory breaches.

Technical concerns are compounded by a widespread management issue: 37% of BFSI executives surveyed indicated that employees use unauthorized tools—including public models like ChatGPT or DeepSeek—as a frequent and persistent risk. This percentage significantly exceeds the global industry average of 30%.

In practice, this means an engineer or analyst might turn to an external, unauthorized AI platform outside the enterprise's controlled architecture, instantly creating serious risks for confidentiality and disclosure of sensitive financial or proprietary client data. Such unauthorized use effectively bypasses enhanced defensive barriers, rendering formal perimeter controls and compliance protocols obsolete.

Meanwhile, while industry discourse remains heavily focused on automating legacy coding tasks and reducing routine costs, Paul Banting, co-founder of growthguru.ai, warns that operational efficiency alone is insufficient to serve as the primary corporate objective.

Banting explained: 'Approximately 79% of executives expect AI to drive significant revenue growth by 2030, up from the current 40%. However, only 24% have a clear idea of where this future profit will originate. The true goal should be implementing AI to identify urgent commercial gaps and capture high-value market opportunities before competitors do.'

Anthropic filed for an Initial Public Offering (IPO) on Monday, which could become one of the largest debuts in financial market history. As part of this filing, the company detailed its rapid business growth and presented an extensive list of potential risks.

The document submitted to the Securities and Exchange Commission (SEC) paves the way for the company's stock market debut and trading, which could happen as soon as possible, no later than November 16th. Reports suggest the IPO could value the company at up to $2 trillion, making it one of the world's largest public companies. The company's shares will trade under the ticker ANT on the Nasdaq exchange.

The San Francisco-based company did not disclose in the filing how many shares it intends to sell or at what price. This information will be determined in the coming weeks before the market debut.

In the filing, the founders wrote a letter to potential shareholders outlining their vision for the value of artificial intelligence while simultaneously disclosing key business risks. These risks range from competition with other AI labs to 'catastrophic or existential threats to humanity.'

The IPO document provides insight into how Anthropic sees the future of its business despite existing concerns and uncertainties. In the first six months of 2026, Anthropic's revenue grew by 1300 percent compared to the previous year, reaching $16.3 billion. The company's operations are largely dependent on the use of its Claude AI system by corporate clients. Anthropic reported having over 14,000 corporate clients, with about 250 of them generating $10 million or more in annual revenue.

However, according to the SEC report, the company recorded a total net loss exceeding $80 billion. Less than $9 billion of this loss was related to infrastructure and AI computing costs, while over $72 billion was due to 'changes in the fair value of convertible bonds, derivative obligations, and SAFEs.' This means that only a small portion of the losses was tied to actual expenses in the first half of the year.

The document does not present results for the most recent fiscal quarter, which ended on September 30th. Anthropic's IPO will be a test of Wall Street's appetite for AI, as the public, policymakers, and industry technologists raise serious questions about AI-related risks and whether the rapid development of research and the construction of data centers should be slowed down.

The filing also touches upon many topics that have raised concerns about an 'AI bubble,' including the possibility that demand for AI will grow slower than expected. Previously, Anthropic was valued at $965 billion after raising $65 billion from investors in May.

Jacob Born, an Emarketer analyst, noted that Anthropic's decision to move forward with the IPO signals a 'readiness to face increased public scrutiny.' He added that Anthropic's IPO occurs at a turning point for the AI industry, as the expected wave of public offerings has stalled amid growing concerns over debt-funded infrastructure investments, uncertain returns, and AI safety risks, which reportedly led competing OpenAI to postpone its own IPO until next year.

Anthropic's filing describes four main categories of AI-related risks: misuse, such as using it to develop and deploy bioweapons or cyberattacks; loss of human control; authoritarian abuse, such as potential use for surveillance systems; and economic shocks.

The document contains 93 pages of Anthropic's business risk factors. This is significantly more than other tech companies that went public in the last decade, such as Snap, which had about 30 pages of risk factors, and Uber, which had about 50.

The filing states that if Anthropic's models cause extreme harm, the company could face liability exceeding its insurance coverage. The company also stated that its AI models may behave unintentionally or 'inconsistently' and may commit 'malicious acts that were not intended by a user or third party.'

Anthropic attempted to reassure potential investors by stating that it believes 'frontier AI companies must invest in safety and alignment to help ensure the responsible deployment of this technology.' However, after going public, the company may face conflicting demands balancing safety and investor returns.

Anthropic also sees several other potential risks to its business, related to the ability to secure sufficient computational resources, as well as uncertain regulatory and political conditions. Furthermore, the company noted that a large part of its revenue depends on a limited number of products and clients.

Nevertheless, according to Dan Ives, partner and senior managing director at Yorkville & Ives Co., Anthropic's growth is likely to overshadow concerns about losses, problems, and potential business risks.

Tech giants, including Meta, Google, OpenAI, and Anthropic, are investing billions of dollars in the AI race, hoping to become dominant players in an industry poised to change society. According to a Gartner market study, global spending on AI in services, software, infrastructure, and other areas is expected to reach approximately $3.6 trillion in 2027, up from $1.8 trillion in 2025.

OpenAI gained an early advantage after launching ChatGPT in November 2022, but competition has intensified over the past four years. Anthropic's Claude Code tool quickly became a favorite among software developers after its launch in early 2025, and the company aimed to replicate this success for other office tasks earlier this year with its Cowork agent.

Claude Cowork contributed to the shift from AI chatbots to agents, causing concern among software stocks due to fears that it might render other companies' software tools obsolete. Fintech company Ramp reported that in September, 43.8 percent of US businesses paid for Anthropic's services, compared to OpenAI's 40.6 percent.

Claude App also demonstrated sharp growth this year. The analytics firm Sensor Tower estimates that in the third quarter of 2026, it will have over 67 million monthly active users compared to 9.4 million the previous year.

However, OpenAI also strengthened its agents this year, releasing the Codex coding application in February and the ChatGPT Work agent in July for multi-step tasks such as writing summaries or analyses. Anthropic stated in its S-1 that its own employees achieved huge efficiency gains using Claude, citing revenue per employee growth from $1.4 million in September 2025 to over $6.6 million in September 2026, reflecting sharp overall revenue growth during this period.

Now, tech companies are competing to create AI agents that will become gateways to our entire digital lives, handling everything from purchases to restaurant reservations and communicating with friends. Anthropic challenges Meta and OpenAI in this area with its new Duet agent, designed to perform multi-step tasks across multiple applications.

At the same time, Anthropic has been at the center of ongoing concerns about the rapid pace of AI development. Last month, CEO Dario Amodei called for an industry-wide slowdown to allow time for catching up on safety research, industry standards, and regulation, a call echoed by some other leading AI figures. His essay followed an employee departure due to concerns that AI might pose a threat to humanity, and follows reports from Anthropic and its competitors about AI models being hacked by other companies during testing.

Anthropic also had a difficult relationship with the White House. The company temporarily revoked access to its most powerful models when the Trump administration demanded it to prevent foreign nationals from using the technology in June. However, Anthropic CEO Dario Amodei became one of several AI leaders who met with President Donald Trump in September and signed a voluntary commitment to self-regulate AI models.

Large financial institutions in the United States are implementing strict restrictions on external artificial intelligence (AI) agents that gain access to customer bank accounts. An Axios report indicates that some of these systems demonstrate the ability to perform actions that exceed the original instructions received.

Institutions such as Morgan Stanley, Bank of America, JPMorgan, Wells Fargo, and Citi are adopting measures to mitigate associated risks, mainly due to the complexity of differentiating authorized automated behaviors from potentially suspicious activities.

Often, customers themselves consent to AI agent access to their accounts by providing login credentials. However, this practice frequently violates the guidelines of various banks, which stipulate that account holders must access their accounts directly.

The concern intensifies when these tools are authorized to execute financial transactions, such as purchases or money transfers. Sources consulted by Axios indicated that some agents also contact call centers or take actions without explicit user request.

Morgan Stanley has begun blocking third-party agents on its platforms, following the example of JPMorgan, Bank of America, and Citi, who are also limiting access to these technologies.

A Chase spokesperson advised customers never to share their usernames and passwords with third parties. In turn, a Morgan Stanley representative assured that the bank had not identified cases of 'out-of-control AI agents accessing our platforms.'

The central challenge lies in determining whether an agent is performing a legitimate task or acting autonomously. This uncertainty causes apprehension among industry executives, especially when the tools have access to financial data and permission to operate.

According to the report, most banks already offer, or are in the process of offering, read-only access. This model is also used by third-party applications, such as Quicken, which assists in budget management and other financial activities.

Institutions recognize the need for consumers to use AI agents to manage finances. Therefore, completely prohibiting this access may not be the ideal solution. The current focus is on developing methods that allow data consultation and the use of these tools without putting customers at risk of unwanted operations.

Banks are investing in prevention mechanisms and preparing response protocols for potential incidents. Axios sources informed that these plans include measures to protect accounts and ensure quick recovery of access by users.

The rapid evolution of artificial intelligence further complicates this work, as new vulnerabilities can emerge in real time, forcing institutions to adapt their defenses while keeping pace with the expansion of these tools.