Tech • IA • Crypto
ChatGPT Can See Your Money Now (This Feels Dangerous)
8/10TL;DR
Major AI platforms are rapidly expanding into personal finance, device control, and data recovery, signaling a shift toward deeply integrated, real-world AI assistance.
KEY POINTS
OpenAI introduces financial account integration
OpenAI has launched a feature allowing US-based ChatGPT Pro users to connect bank accounts, credit cards, loans, and investment portfolios directly within the platform. Using Plaid, which supports over 12,000 financial institutions, the system aggregates balances, transactions, and liabilities into a unified dashboard. Users can then query their real financial data conversationally, moving beyond generic advice to personalized financial insights.
AI-driven financial guidance and planning
The system enables detailed queries such as affordability analysis, subscription management, and investment evaluation based on actual spending behavior. Powered by GPT-5.5 Thinking, the tool scored 79/100 on internal finance benchmarks, with a higher-tier version reaching 82.5, reflecting strong performance on complex financial reasoning tasks. This positions AI as a potential everyday financial advisor rather than a passive assistant.
Privacy safeguards and limitations
OpenAI states that the system cannot execute transactions or access full account numbers, limiting it to read-only insights. Users can disconnect accounts at any time, with data deleted within 30 days. Features like “financial memory” store contextual user goals separately, while respecting existing data training preferences. Temporary chats are excluded from accessing financial data, addressing some privacy concerns.
Expansion toward financial actions
The roadmap includes deeper integrations through partners such as Intuit, aiming to enable actions like applying for credit cards or consulting tax professionals directly within ChatGPT. This signals a shift from advisory capabilities to execution, potentially transforming the platform into a full-service financial interface.
Tencent develops OS-level AI assistant “Marvis”
Chinese tech giant Tencent is building Marvis, an AI assistant that operates at the operating system level across Windows and Android, with plans for macOS and iOS. Unlike typical chat-based tools, Marvis can control applications, manage files, and execute multi-step workflows using coordinated AI agents, effectively acting as a digital operator for devices.
Multi-agent system with real task execution
Marvis uses a multi-agent architecture where specialized agents collaborate under a supervisor to complete tasks. It can generate documents, analyze system performance, and even build software systems autonomously. In one test, it created a functional knowledge management platform, including backend structure and interface, without human coding input.
Performance strengths and current limitations
While capable, Marvis remains inefficient in some scenarios. A simple image search task required analysis of over 100 files and consumed 2 million tokens, highlighting computational overhead. It can identify system settings but may fail to execute final steps without user intervention. Pricing inaccuracies and language inconsistencies also indicate early-stage development.
AI assists in recovering lost Bitcoin
In a notable case, Anthropic’s Claude helped recover 5 Bitcoin, worth nearly $400,000, lost for over a decade. The system did not break encryption but analyzed old files and backups, identifying a pre-password-change wallet.dat file and pairing it with an existing recovery phrase. This enabled access through legitimate means.
Implications for crypto and data recovery
Millions of Bitcoin are believed to be permanently inaccessible due to lost credentials. AI tools are emerging as powerful assistants in scanning historical data, identifying patterns, and surfacing recovery clues. While not bypassing security, they significantly lower the technical barrier for recovery efforts.
CONCLUSION
AI is rapidly evolving from a passive tool into an active agent capable of managing finances, operating devices, and uncovering lost assets, raising both opportunities and new concerns around trust, privacy, and control.