What proprietary AI has Commonwealth Bank built in-house?
Commonwealth Bank builds its own AI applications and agents but does not disclose a foundation model of its own; its large language models come from partners. The oldest in-house system is the Customer Engagement Engine, a machine learning system launched in 2015 that ran over 2,000 real-time models at June 30, 2026. Since 2023 the bank has added CommBank Gen.ai Studio (an H2O.ai-powered environment that produced 50-plus generative AI use cases between June and November 2023), an AI factory on AWS EC2 P5 instances for fine-tuning LLMs (September 2024), and an AI Guardrails-as-a-Service layer. Named systems the bank says its own teams built include CommBank Companion, the in-app assistant 'designed and built by CommBank's own engineering and data science teams' and announced on May 26, 2026; an agentic fraud system built in three months that has helped write three quarters of card fraud rules (April 2026); two cyber threat-hunting agents that cut threat assessment from two days to 30 minutes (March 2026); and Project Coral for tech-debt fixes. Companion is built with Amazon Bedrock, the CommBiz messaging service uses Claude 3 and Cohere models, and staff use ChatGPT Enterprise. Chief AI Officer Ranil Boteju said in September 2026 the bank is looking at 'scaling small language models or using open source to manage costs'.[1][2][3][4][5][6][7][8][9][10][11][12]
Data as of Oct 22, 2026 · Commonwealth Bank profile updated Oct 13, 2026 · 26 sources cited on this page · 140 banks profiled on this site · Markdown
Which AI systems has Commonwealth Bank built itself, and on what?
Systems the bank describes as built or developed in-house. 'Built on' names the outside model or platform where the bank or its vendor has said so; blank model details mean none were disclosed.
| Platform or model | What it does | Built on | Since | Status |
|---|---|---|---|---|
| Customer Engagement Engine (CEE) | Machine learning system for next best conversations across banking channels; enabled Benefits finder, which has supported more than 400 rebates since 2019 | In-house; generative AI gave a '30x increase in experimentation capability' against the CEE A/B testing framework (February 2024) | 2015 | In production; over 2,000 real-time models at June 30, 2026[1][5][2][3] |
| Bill Sense | Machine learning that learns a customer's regular payments and predicts when bills will arrive and how much they will be, with the customer confirming each step | Not disclosed | 2020 | In production[5] |
| Abuse in Transaction Descriptions model | Detects harassing or threatening payment messages; trained on anonymised high-risk cases from CBA transaction data, retrained on cases the Next Chapter team confirms; flags over 3,600 cases a year | Machine learning, natural language processing and pre-trained LLMs; shared with other banks with H2O.ai | Shared 2023 | In production[5] |
| CommBank Gen.ai Studio | Safe environment for experimenting with LLMs; produced 50-plus generative AI use cases between June and November 2023 | H2O.ai | May 2023 | Reported February 2024[3] |
| AI factory | Compute for testing and developing AI solutions, 'including fine-tuning and training LLMs' | Amazon EC2 P5 instances (AWS) | September 2024 | In use[5][4] |
| AI Guardrails-as-a-Service (GaaS) | Shared safety layer for generative AI; in Ceba a groundedness guardrail checks each LLM answer against the retrieved context before it reaches the customer | In-house | Described December 2025 | In production[5] |
| Ceba with generative answers | The bank's chatbot, introduced in 2018, handles over 200 tasks and can answer some questions from the public CommBank website by retrieval-augmented generation | LLM provider not named in the report | 2018 (Ceba) | In production[13][5] |
| CommBiz messaging service | Answers business customers from more than 80 CommBiz user guides and FAQs in natural language | Amazon Bedrock Knowledge Bases, Claude 3 and Cohere LLMs, Amazon OpenSearch | Late 2024 | In production[4] |
| Compass AI | Generative AI assistant for business bankers; answers knowledge-base questions over three times faster than before | Not disclosed | July 2024 | More than 500,000 questions answered[5][1] |
| CommBank Companion | In-app assistant for budgeting, home buying and business cash flow; multi-agent, with domain teams building sub-agents and purpose-built guardrails for gambling, financial abuse and hardship | Designed and built by the bank's own teams; built with Amazon Bedrock; model not disclosed | May 26, 2026 | Rolling out; one third of customers by August 2026[6][14][10][15] |
| Agentic fraud-rule system | Finds emerging fraud and scam patterns, assesses severity and proposes detection rules that analysts approve; contributed to three quarters of card fraud rules | Built in-house in three months on Snowflake's data cloud and the cloud core banking platform | April 24, 2026 | In production[7] |
| Cyber threat-hunting agents | One agent tests new threat research against the bank's own data and estate and reports in 30 minutes instead of two days; a second hunts indicators of compromise | Built by cyber staff with the bank's data scientists | Reported March 17, 2026 | In production[8] |
| Project Coral and Lumos | Coral raises pull requests to fix tech debt, designed by graduate and distinguished engineers and runnable on 'any AI development tool'; Lumos is a multi-agent workflow that prepares applications for the cloud | Coral: tool-agnostic; Lumos: built with AWS ProServe, using OpenRewrite and Amazon Q Developer | August 2025; September 2025 | In production[9][16] |
Which teams build Commonwealth Bank's AI?
Chief AI Officer Ranil Boteju, who joined from Lloyds early in 2026, says his remit is a single enterprise AI platform for engineers and data scientists, standard controls, guardrails, evaluation and observability, token-cost tracking, and an AI science research lab 'looking at innovations a couple of horizons away'. He told Retail Banker International that over 80% of the bank's 7,800 engineers use 'a combination of bought and built AI-powered coding agents'.[17][12]
The research lab is led by Chief AI Scientist Professor Mary-Anne Williams, who started on June 15, 2026 and heads a team of Distinguished AI Scientists specialising in machine learning, responsible AI, AI security and generative AI. A Human-Centred AI unit set up in November 2024 tests AI experiences with technologists, psychologists, economists, designers and ethnographers.[18][5]
Product systems come from business-line teams. Companion uses a federated model: domain teams across the bank build primary or sub-agents and a central team tests each one before launch, executive general manager Michael Baumann said in August 2026. The fraud agent came from the bank's in-house data science and engineering teams, and the AI Powered Engineering team, formed in July 2024, built Project Coral.[14][7][19][9]
When does Commonwealth Bank build AI rather than buy it?
The bank buys models and builds the layer around them. Its 2026 annual report names AWS, Anthropic, Microsoft and OpenAI as strategic partners; the AI factory exists so engineers can fine-tune LLMs on AWS hardware; and AWS's case study says Companion is built with Amazon Bedrock. The contact-centre orchestration platform is built on Microsoft's Copilot Studio, Dynamics 365 and Foundry.[2][4][10][20]
Where it builds, it says it does so for speed or control. General Manager of Cyber Defence Operations Andrew Pade said the bank built its own threat-hunting agents because security vendors 'can't keep up with emerging threats and the bank can't wait for a product'. For Companion the bank stresses that the assistant was designed and built by its own engineering and data science teams inside the CommBank app's secure environment.[8][6]
What has Commonwealth Bank not disclosed about its own AI?
The bank has not named a proprietary foundation model, said which LLMs it has fine-tuned in the AI factory, or named the model behind Companion or Ceba's generative answers. It has not published how many of the 2,000-plus CEE models are machine learning versus rules, or how many of the 50-plus Gen.ai Studio use cases went into production.[2][5][3]
Spend on in-house AI is not broken out: the bank reports about A$200 million of measured gross AI benefits for FY26 but no cost for building its own systems. The two published patent applications assigned to the bank, on vector embeddings and on translating non-natural language data, are not tied by the bank to a named system.[21][22][23]
What does the Commonwealth Bank profile record on this topic?
The rows below are the entries of the CommBank AI strategy profile that bear on this page, as of Oct 13, 2026; the profile holds the full record of 36 dated moves, 20 use cases and 54 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Agentic fraud-pattern detection and rule generation | An agent that assesses emerging patterns, proposes new detection rules and has contributed to developing or updating three quarters of card fraud rules; analysts review rules before implementation. It sits within a system monitoring more than 80 million signals a day. | In production | [7][24] |
| Detecting abuse in transaction descriptions | A model that learns to block abusive words and phrases in payment descriptions; the annual report says over 265,000 transactions were blocked in FY26 and the AI model identified over 1,700 cases for human review. | In production | [2] |
| CommBank Companion in the app | A conversational AI experience for budgeting, home buying and, for businesses, cash-flow questions; one third of customers had been given access by August 2026 and half of their queries related to managing spending and saving. | Rolling out | [6][15] |
| Compass AI for frontline teams | A generative AI solution that delivers business-bank knowledge-base answers three times faster than traditional methods and had answered more than 500,000 questions since July 2024. | In production | [1] |
| In-house agentic threat-hunting agents | One agent ingests new threat research, tests it against the bank's own estate and prepares a report in 30 minutes instead of two days; a second finds indicators of compromise. The cyber defence general manager said the bank built them because vendors could not keep pace. | In production | [8] |
| ChatIT employee IT-support agent | A generative AI assistant in Microsoft Teams, built on Azure and Copilot Studio, that answers IT questions and runs automated fixes; over 2.3 million messages and more than 12,000 automated fixes in the first six months of automation, with a staff satisfaction score of 4.63 out of 5. | In production | [25] |
| Project Coral agentic tech-debt remediation | An agent framework built by graduate and distinguished engineers that connects to monitoring tools, analyses code and raises pull requests; engineers must review and approve every change. The Chief AI Officer reports over 80% of the 7,800 engineers use AI coding agents. | In production | [9][12] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Aug 12, 2026 | FY26 results: AI benefits and Companion | The bank reports about A$200 million of measured gross AI benefits for FY26, expects them to double in FY27 and exceed investment, says one third of customers have been given Companion, and discloses a A$1,527 million Anthropic holding in its annual report. | [21][15][2] |
| May 26, 2026 | CommBank Companion enters testing | An AI-powered conversational experience in the CommBank app is tested with employees and select business customers, with a consumer test planned in the following weeks. | [6] |
| Apr 24, 2026 | Agentic fraud system deployed | CommBank deploys an agentic AI system that detects emerging fraud and scam patterns and proposes detection rules, reviewed by analysts before use; built in three months on Snowflake and the bank's cloud core banking platform. | [7][24] |
| Mar 17, 2026 | In-house agentic threat-hunting tools described | At Gartner's Security and Risk Management Summit in Sydney, Andrew Pade said CBA built two agents because vendors could not keep pace: one assesses emerging threats in 30 minutes instead of two days, the other looks for indicators of compromise; weekly threat signals had risen from 80 million to over 400 billion. | [8] |
| Feb 4, 2026 | Australian-first AI adoption report | CBA publishes Our Approach to Adopting AI: six responsible AI principles, over 27,600 employees engaged with its AI learning series, scam-detection AI over more than 20 million daily payments, and Compass AI having answered more than 500,000 questions since July 2024. | [1][26] |
| Aug 15, 2025 | AI Powered Engineering and Project Coral | CommBank describes an AI Powered Engineering team created in 2024 and Project Coral, an agentic tool that reads monitoring data, analyses code and raises pull requests with fixes for tech debt; engineers must approve every change. | [9] |
| Jul 1, 2025 | ChatIT IT-support agent in Microsoft's AI Innovators list | Microsoft reports ChatIT, built on Azure and Copilot Studio and launched in 2024, had handled over 2.3 million messages and run more than 12,000 automated IT fixes since its automation function started six months earlier, at about two minutes per fix against 17 minutes for a service-desk call. | [25] |
| Feb 5, 2025 | AWS collaboration expanded | CommBank and AWS expand their collaboration on cloud and AI; the release says AI adoption had doubled to over 2,000 AI models in 2024, describes a CommBiz GenAI messaging service using Amazon Bedrock Knowledge Bases, Claude 3 and Cohere models, and recalls the AI Factory launched in September 2024. | [4] |
Does Commonwealth Bank build its own AI models?
It builds its own machine learning models and AI applications, but has not disclosed a foundation model of its own. The Customer Engagement Engine has run in-house machine learning since 2015 and held over 2,000 real-time models at June 30, 2026, and the bank trained its own abuse-detection model on its transaction data. For large language models it relies on partners: Claude 3 and Cohere models in CommBiz messaging, Amazon Bedrock for Companion and ChatGPT Enterprise for staff, with an AWS AI factory used to fine-tune LLMs.
Who built CommBank Companion?
The bank says Companion was 'designed and built by CommBank's own engineering and data science teams', and AWS's case study says it is built with Amazon Bedrock. It is multi-agent: domain teams across the bank build sub-agents and a central team tests each before launch. It was announced on May 26, 2026 and had reached one third of customers by August 2026. The underlying model has not been named.
What is the Customer Engagement Engine?
A machine learning system CommBank launched in 2015 to choose the next best conversation with each customer across its channels. It enabled Benefits finder, which has supported more than 400 rebates since 2019, and it was used for personalised pricing offers to home-loan customers coming off fixed rates. The 2026 annual report says it runs over 2,000 real-time machine learning models.
What is CommBank Gen.ai Studio?
An environment powered by H2O.ai that the bank set up in May 2023 for safe experimentation with large language models. CBA's half-year results material in February 2024 said 50-plus generative AI use cases came out of it between June and November 2023 and that over 500 staff had been upskilled on AI tools. CBA led H2O.ai's $100 million Series E in 2021.
Which AI agents has Commonwealth Bank built in-house?
Named examples: an agentic fraud system built in three months that proposes card fraud rules for analyst approval (April 2026); two cyber threat-hunting agents, one cutting threat assessment from two days to 30 minutes (March 2026); Project Coral, which raises pull requests to fix tech debt (August 2025); Lumos, built with AWS ProServe, for cloud migration; and the sub-agents inside Companion.
- CommBank releases Australian-first report outlining how it is adopting AI
- 2026 Annual Report
- CBA racks up "50-plus" GenAI use cases across the bank
- CommBank and AWS expand collaboration to deliver global best cloud and AI capabilities, enabling idea to production in six weeks
- Our approach to adopting AI (report, December 2025)
- Reimagining banking with CommBank Companion
- CommBank develops AI agent that spots new fraud and helps build defences
- Bank built its own AI threat hunter because vendors can't
- CommBank enhances engineering experience with Agentic AI
- Commonwealth Bank modernizes core banking, improving response times by 30% (case study, retrieved 2026-10-15)
- Commonwealth Bank of Australia builds AI fluency at scale
- Interview: CBA AI chief on scaling A$400m in AI value with humans in loop
- Commonwealth Bank's AI customer bot slammed as savings rates slashed: 'Pathetic'
- CBA's AI Companion is expected to face some hard questions
- CBA 2026 Full Year Results Transcript
- CBA builds Lumos, an AI-driven accelerator to prepare applications for cloud
- Commonwealth Bank appoints Ranil Boteju as Chief AI Officer
- Commonwealth Bank appoints its first Chief AI Scientist
- CBA plans to use AI across 'entire software delivery'
- How Commonwealth Bank and Microsoft are reimagining the future of customer service
- CBA 2026 Full Year Results Presentation
- US20250307709A1 - Vector embedder for non-natural language data
- AU2025204862A1 - Systems and methods for translating non-natural language data into natural language data
- CommBank builds AI agent that spots fraud and helps build defences
- Commonwealth Bank's ChatIT helps employees fix IT issues in seconds using generative AI
- CommBank bids to ease customer fears over AI adoption
When Commonwealth Bank moves on AI, you'll read it here first.
when one of the banks we cover moves on AI, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning