{
  "id": 9187415,
  "title": "OpenAI Hands Banks a Shortcut Around Billion-Dollar Core Rewrites",
  "url": "https://urgent.news/2026/09/22/openai-hands-banks-a-shortcut-around-billion-dollar-core-rewrites",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-22T19:46:42.000Z",
  "source": {
    "name": "PYMNTS",
    "slug": "pymnts",
    "url": "https://www.pymnts.com/news/banking/2026/openai-hands-banks-a-shortcut-around-billion-dollar-core-rewrites/"
  },
  "original_language": "en",
  "account": "In modern banking, an employee at a bank's back office would typically pull up a customer's account on an outdated terminal, read the balance and input it into a newer web form to finalize the transaction. This interaction is necessary due to the age and structure of bank technology. Most systems managing balances, loan processing, and payment movement are ancient, written before modern software standards for data exchange. Banks aiming to incorporate AI into these processes have two options: replace the outdated systems— a long and costly undertaking, or create bespoke connections for each system, which is also time-consuming and expensive.\n\nA third approach is gaining traction. OpenAI has introduced its new GPT-6 Astra model, capable of independently operating a computer. This AI model can read the screen, fill out forms, update records, and navigate between programs, much like a human employee would. In internal testing, Astra achieved a 72.6% completion rate on a suite of tasks within approximately 40 minutes, outperforming its predecessor GPT-5.6 Sol, which scored 65.7% in about 75 minutes.\n\nAccording to PYMNTS, legacy COBOL systems run an estimated 95% of U.S. ATM transactions and are present in hundreds of billions of lines of code across finance, airlines, and government. Estimates from DXC, cited by Amazon Web Services, indicate that over 40% of banks still operate COBOL-based core systems, and 45 of the world's 50 largest banks rely on mainframe applications for critical operations. Replacing these systems is often slow, costly, and risky. For instance, Commonwealth Bank of Australia's core replacement took five years and over $1 billion AUD, while U.K.-based bank TSB's switch to a new platform in April 2018 left 5.2 million customers temporarily unable to access banking services. As a result, the U.K.'s Financial Conduct Authority and Prudential Regulation Authority imposed a fine of £48.65 million on TSB in 2022, with customer access only restored eight months later.\n\nTraditionally, banks have tried to automate old screens using software robots, known as robotic process automation (RPA), which follow predefined scripts. However, when a button moves or a screen changes, these robots become ineffective. In contrast, computer-use AI, such as the GPT-6 Astra model from OpenAI, operates differently. It observes the screen, determines the necessary actions, and adapts when changes occur. Major tech companies, including OpenAI, Anthropic, Google, and Amazon, now offer this type of agent. Financial institutions, such as Goldman Sachs, are testing these agents for labor-intensive tasks like trade and transaction accounting, client vetting, and onboarding, referring to them as \"digital coworkers\" for complex, process-heavy jobs. Goldman Sachs has reported that these agents have performed well in accounting and compliance work, potentially reducing the need for outside vendors as the technology matures. However, a test score on a simulation does not equate to fully functional bank operations, and expert opinions suggest that trustworthy computer-use agents remain an open research problem.",
  "summary": "In a bank’s back office, an employee pulls up a customer’s account on a decades-old terminal, reads the balance and types it into a newer web form to finish the file. The two systems don’t talk to each other. The person in the chair is the connection. That job exists because of how bank technology […] The post OpenAI Hands Banks a Shortcut Around Billion-Dollar Core Rewrites appeared first on…",
  "key_points": [
    "OpenAI's GPT-6 Astra model can independently operate computers.",
    "Model outperformed predecessor GPT-5.6 Sol in internal tests.",
    "Banks face costly, risky challenges in replacing legacy COBOL systems."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}