The Newest Attack Vector: AI-Driven Bank Integration
AI-generated image · US National Wire
Anthropic and OpenAI are racing to link LLMs directly to your financial accounts, but 'read-only' access doesn't eliminate the systemic risk of expanding your digital attack surface.
Let's be clear: treating your bank account as a data source for a Large Language Model isn't a productivity win; it's a security nightmare. We are seeing a race between AI giants to integrate directly into our most sensitive financial silos, and the industry's insistence that these tools are 'read-only' is a dangerous distraction from the broader threat model.
As Engadget first reported, Anthropic recently launched a feature for Claude called Personal Finance, positioning it as a direct competitor to OpenAI's Finances feature in ChatGPT, which launched in May. Both tools aim to provide users with insights into their spending habits, investments, and monthly expenses through generated dashboards. To function, both companies have partnered with Plaid to facilitate connections with major financial institutions, including Chase, Capital One, and Bank of America.
From a defender's perspective, the 'read-only' label is a thin veil. Engadget reports that Anthropic's tool can access account names, types, the last four digits of account numbers, balances, credit limits, and two years of transaction history. Similarly, OpenAI's tool can view liabilities, investments, transactions, and balances. While the AI cannot move your money, it now possesses a comprehensive map of your financial life.
***
**Opinion: The Privacy Illusion**
When companies argue over privacy settings, they are arguing over how they manage the data they've already convinced you to hand over. Whether the data is used for model training or deleted after 30 days is irrelevant if the initial connection creates a permanent vulnerability in your security posture. By linking these accounts, you aren't just using a tool; you are creating a high-value target for any adversary who can compromise your AI account.
***
Even the differences in privacy settings highlighted by Engadget feel like splitting hairs. Anthropic claims that conversations within the Personal Finance hub are not used for model training by default and that unlinking an account removes Claude's access to previously provided data. OpenAI, conversely, ties its Finances tool to the user's general ChatGPT settings. This means users must manually opt out of the "Improve the model for everyone" setting to prevent their data from being used for training. Furthermore, OpenAI states that synced account data is deleted within 30 days of disconnection.
Beyond the data harvesting, the incentives are shifting. Engadget notes that Anthropic's tool currently avoids ads, sponsored recommendations, and the offering of its own loans or credit cards. OpenAI, however, has been testing ads for its lower-cost tiers as of January and is partnering with Intuit to expand ChatGPT's Finances into areas like tax queries and credit card applications.
As Anthropic seeks partnerships with wealth managers and banks to refine its tool, the goal is clear: total integration. But for those of us who actually threat-model our lives, the only safe way to use an AI for finance is to keep it far away from your actual bank credentials.