Thought Machine and AWS launch Vault Forge to pull banks off mainframe cores

Thought Machine said Tuesday it partnered with AWS to launch an AI-powered migration path that uses AWS Transform and Vault Forge on Amazon Bedrock to extract legacy mainframe logic and synthesize simulation-tested Python financial products for banks.

Banks have spent years and fortunes trying to crawl off cores still written in COBOL. Thought Machine and AWS say the job for AI is to extract the business rules, not rewrite every line, then put tested Python products in a Vault sandbox inside the bank’s own AWS environment.

On Tuesday, 29 September 2026, Thought Machine announced an AI-powered migration solution with Amazon Web Services (AWS) Transform. Thought Machine calls itself a cloud-native core banking and payments company. Core banking is the system that keeps the accounts, the interest, the fees, and the payments. The Thought Machine press page is dated September 29, 2026, and does not print an hour. The AWS US Press Center carries the same announcement the same day, datelined London and New York, 29 September 2026, and does not print an hour either. Thought Machine’s headline says the companies partnered to launch an AI-powered solution to accelerate legacy core banking migrations. The AWS headline says they partnered on an AI-powered solution to accelerate legacy banking migrations. Those lines are the companies’.

What the pages say the partnership is for. Vault Forge, powered by Amazon Bedrock, is meant to cut the time and the cost of a migration programme. A migration programme is the project of moving a bank off its old core. Amazon Bedrock is Amazon’s service for running AI models. The pages say Vault Forge uses frontier AI models to turn mainframe logic that has already been extracted into Python financial products that have been tested in a simulation. Frontier, in that phrase, is the companies’ word for the most advanced models they are using on this job. A mainframe is the large central computer many bank cores still run on. Python is the language Vault uses to describe a product. A simulation, here, is a test run before anyone treats the product as the live bank. The pages also say the solution lets banks modernise old systems faster, safely, and at a fraction of traditional costs. They do not print that fraction, and they do not print a price.

How the pipeline is supposed to work, as both pages state it. It combines AWS Transform’s tooling for mainframe reverse engineering with Vault Forge, which Thought Machine calls an AI tooling suite. The AWS page names that AWS side as tooling for mainframe reverse-engineering agents. Reverse engineering, here, means reading an old system to recover what it actually does. The pages say banks can turn millions of lines of mainframe code, such as COBOL, into tested financial products in a live sandbox, compressing multi-year processes into days. COBOL is the programming language still used on many of those cores. A sandbox is a closed copy of the platform where the new products can be tried. A live sandbox, in that sentence, is a running test copy. It is not the bank’s production core. Those lines are the companies’. The pages do not name a bank that has finished that compression.

The method, set against a line-by-line rewrite. Rather than translating code one line at a time, the joint pipeline extracts pure business intent inside the customer’s secure cloud environment. Business intent, here, is what the product is supposed to do: the interest, the fee, the repayment, the life of the account. Both pages say Thought Machine is the first core banking technology company to partner with AWS Transform’s modernisation agents, so the work shifts from risky line-by-line code translation to automated extraction of business rules. “First” is the companies’ claim. The pages do not name another core-banking company they measured themselves against.

Why they say the old core is hard to move, and why Vault is shaped the way it is. Traditional cores hardcode financial products into the database underneath, which the pages say makes a migration slow and risky. Hardcode means the product rules are buried in that database, not written as a separate program. Vault defines financial products entirely as Python. Product logic such as interest calculations, fee schedules, repayment workflows, and account lifecycles is decoupled from the core infrastructure. Decoupled means those rules sit apart from the machines that store the accounts. The pages say AI agents can then skip the database tangle and translate the old rules into that high-level code. An agent, here, is software that carries out a step, not only a chat reply. Those lines are Thought Machine’s and AWS’s.

Four stages, which both pages list, and which they say run entirely inside the bank’s secure AWS environment. Extraction: AWS Transform discovers and reverse-engineers the old applications and writes the business logic as structured specifications in Easy Approach to Requirements Syntax, shortened to EARS. EARS is a set way of writing what a system is supposed to do. Consolidation: the pipeline folds hundreds of redundant old product variations into lean, modern specifications. Redundant, here, means many versions of what is really the same product. Synthesis: Vault Forge orchestrates specialised AI agents through Amazon Bedrock, meaning it lines those agents up, to produce Python financial products, test suites, and platform configurations. The pages call those products SDK-compliant. An SDK, a software development kit, is the set of rules the code has to follow to run on Vault. Validation: a person still has to sign off. Thought Machine’s page says those review gates give the bank explicit sign-off authority. The AWS page says they give bank engineering and risk teams that authority. Once the products pass, they can be deployed to a Vault sandbox for further evaluation. The pages also say banks receive fully validated migration proof early in the process. Those lines are the companies’.

The programme that comes with the launch. Thought Machine is launching a Core Modernisation Accelerator. The AWS page calls it a joint programme. Technical architects, product specialists, and forward-deployed engineers work alongside the bank’s own team. A forward-deployed engineer is an engineer who sits with the customer’s people. Thought Machine’s page says they work with bank teams on delivery, governance, and how the system is set up. The AWS page says Thought Machine’s forward-deployed engineers work with bank engineering teams. Governance, here, is who may approve a change. Those lines are the companies’. The pages do not name a bank enrolled in the programme, and they do not print a fee.

Paul Taylor, chief executive of Thought Machine, said the company was built “to properly and permanently free the world’s banks of legacy technology.” He said the AWS partnership is “a massive leap forward” in that promise, and that combining AWS Transform with Vault Forge means “rewriting the economics of core modernisation.” He said that “for the first time, banks can deploy AI to rapidly convert complex legacy mainframe specifications into simulation-tested, cloud-native financial products.” Cloud-native means the software is built to run in a cloud, not on the old mainframe. He said that gives large institutions “the execution speed, audit rigor, and confidence required to eliminate technical debt and complete their cloud transition.” Technical debt, in that sentence, is the cost of systems that are old and hard to change. Audit rigor means a record a reviewer can check. That quotation is his, on both pages. The pages do not print a before-and-after cost.

Charith Mendis, head of worldwide banking for financial services industries at Amazon Web Services, said financial institutions are “at an inflection point.” He said the cost of keeping old infrastructure keeps rising while customer expectations and regulatory demands move quickly. He said Vault Forge and AWS Transform let banks “safely extract business logic from legacy systems,” and that this is “a new continuous, automated path from mainframe code to cloud-native core banking.” He called that one of the hardest modernisation problems in banking, and said AWS believes it “will fundamentally change how the industry approaches core transformation.” That quotation is his, on both pages. “We believe” is the wording on the page.

The about box on the AWS page describes the company, not a customer list for this launch. It says Vault is trusted by over 65 leading banks and financial institutions worldwide, and that several of those also count as strategic investors. Over 65 is that company figure. It is not a count of banks running Vault Forge on this announcement. The same box says Thought Machine has raised more than £500 million. The page prints that figure as £500m. It says the team works from London, New York, Singapore, Sydney, and Lisbon. Those lines are the about box. The pages do not name a bank customer for the new pipeline, and they do not print annual revenue.

In plain terms, Thought Machine and AWS said on Tuesday that a bank can point AWS Transform at a COBOL core, pull the business rules out as structured specifications, and have Vault Forge write tested Python products for a Vault sandbox, with a person still signing off before anything is deployed. The work is supposed to stay inside the bank’s own AWS environment. The “days instead of years” line, the “fraction of traditional costs” line, and the claim to be first are the companies’. The pages do not name a bank that has completed a migration this way, and they do not print a price.

The picture is Thought Machine’s partnership graphic for the release. The Thought Machine wordmark sits above the line “powered by,” then the AWS mark. The card names Vault Forge and AWS Transform. It is the announcement image from the press page. It is not a photograph of a bank’s machine room, and it does not print a date.