CommBank Opens Its Internal AI Security Tool to All Australian Businesses
CommBank has released CodeCheck, a free tool that steers AI coding agents like OpenAI's Codex and Anthropic's Claude Code through vulnerability analysis, built on the bank's internal security system.

Why it matters
- CommBank launched CodeCheck, a free cybersecurity tool for Australian businesses, on 18 November 2026.
- CodeCheck works with AI coding agents including OpenAI's Codex and Anthropic's Claude Code.
- The tool is based on a system CommBank uses internally across roughly 2.5 billion lines of code and thousands of engineers.
- Human engineers remain responsible for reviewing findings and deciding what action to take.
- The tool targets businesses that build software with AI agents, not buyers of off-the-shelf software.
Commonwealth Bank of Australia has released CodeCheck, a free cybersecurity tool that uses AI coding agents to find software vulnerabilities before code ships. The bank built the system for its own engineers, who maintain roughly 2.5 billion lines of code, and it made the tool available to Australian businesses on November 18, 2026.
The release matters because it marks a major financial institution converting an internal defensive capability into a public good. Large organisations are increasingly deploying AI not just to write code but to check it for weaknesses, and CommBank is now pushing that approach out to any Australian company willing to use it.
What does CodeCheck actually do?
CodeCheck works with AI coding agents — software programs that autonomously write and modify code — including OpenAI's Codex and Anthropic's Claude Code. The tool guides the AI through a defined process: it instructs the agent to analyse code, investigate possible vulnerabilities, and return findings for human engineers to review.
The tool accepts added context. A business can describe what its software does and how it is intended to work, and CommBank says this helps the AI make more relevant assessments of potential vulnerabilities. Context, in other words, is a lever the user controls.
Two design choices deserve attention. First, engineers remain responsible for reviewing every finding and deciding what action to take — the tool assists, it does not decide. Second, businesses choose what information they share with the system, a concession to the sensitivity of source code and internal documentation at companies that may be reluctant to hand material to a bank-run tool.
The takeaway: CodeCheck automates the scanning step of security review, but the judgement step stays human.
Who is it for — and who is left out?
CommBank has drawn a clear boundary around the target audience. CodeCheck is intended for organisations that already use AI coding agents in their development workflows — companies with in-house engineering teams that have adopted generative AI for coding tasks.
It is not for businesses that buy off-the-shelf software. If a company does not build its own software, CodeCheck has nothing to analyse. That exclusion covers a large share of the Australian small-business market, and it is worth noting the headline — "for Australian businesses" — applies to a narrower slice than it may first suggest.
The focus reflects a real and growing segment. More companies now use generative AI for engineering work while facing unresolved questions about code quality, security and oversight. CommBank is betting that segment will keep growing, and that it will want tooling.
The takeaway: the tool serves AI-assisted developers, not the software market as a whole.
Why did a bank build this in the first place?
The backstory is scale. CommBank's systems are supported by about 2.5 billion lines of code and thousands of engineers, according to the bank. At that size, software security and code review become significant operational problems — and AI tools are now common across the industry's development pipelines, multiplying both the volume of code produced and the volume that needs checking.
CodeCheck is based on a system already used inside CommBank's own software development process, built by the bank's security engineers. The public release simply extends an internal working model outward.
Nicola Nicol, CommBank's Chief Security Officer, described how that internal experience shaped the tool. In the bank's statement, she said:
"With thousands of engineers working across systems powered by around 2.5 billion lines of code, we've seen first-hand how AI can help identify potential vulnerabilities earlier."
She added that the bank built the capability for itself first: "We've built this capability to help strengthen our own defences. By making CodeCheck freely available, we want to share what we've learned and help more Australian businesses use AI to strengthen theirs."
The takeaway: this is a bank exporting a practice that already works at enormous scale — at least internally.
What does the release say about the threat picture?
CommBank framed CodeCheck as part of a response to adversaries, not just inefficiency. The bank's argument is that cyber criminals increasingly use AI, and that defensive tools must evolve in parallel. Nicol was explicit on this point:
"Cyber security is a shared challenge, and strengthening Australia's cyber resilience requires us all to play a part. As cyber criminals increasingly use AI, we also need to put AI to work on the defensive side. Making CodeCheck freely available is one practical way we can help Australian businesses do that."
The stated context is a business environment where attacks increasingly affect supply chains, software vendors and internal systems, and where cyber risk is a growing concern for companies of all sizes. CommBank is positioning itself as a contributor to national cyber defence efforts beyond banking.
The claim that criminals are weaponising AI at scale is presented here as motivation, not as a finding the bank documents in the release. Readers should note the evidence base sits outside this announcement.
The takeaway: the bank's pitch is symmetry — if attackers use AI, defenders should too.
What are the limits and caveats?
Several boundaries deserve stress-testing, and some questions the announcement leaves unanswered.
- Automation is partial. CodeCheck automates parts of the review process but is designed to assist engineers rather than replace them. Human review remains central, particularly when a potential vulnerability must be assessed against a company's own systems and risk standards.
- The tool's effectiveness figures are not disclosed. CommBank offers no published benchmarks, detection rates or false-positive data in the release. "Works for us" is the evidence provided; independent verification is not.
- Trust questions are deflected, not resolved. Businesses choose what to share, but the release does not detail how submitted code or context are handled, stored or protected.
- The free price carries no stated conditions in the announcement. What the bank gains — reputation, ecosystem goodwill, data — is not spelled out.
None of this makes CodeCheck unattractive. It does mean adopters are relying on CommBank's word that an internal tool generalises to their environments.
Why does this fit a wider pattern?
The release highlights a broader shift among major financial institutions: internally developed cyber tools are increasingly treated as assets that can be adapted for the wider market. CommBank is extending a system shaped by its own security needs to other Australian businesses seeking earlier warning of software weaknesses.
That pattern has precedent in the logic of shared infrastructure. Banks lose if their vendors, suppliers and customers are compromised, so raising the floor of national cyber resilience serves their interests as well as the public's. A free tool from a bank is generosity with an aligned incentive — a caveat worth attaching to every promise of free security.
The practical question ahead is adoption and proof. Whether CodeCheck measurably reduces vulnerabilities in Australian businesses will depend on uptake among AI-agent-using developers and on whether CommBank, or independent researchers, eventually publish data showing it works beyond the bank's own 2.5-billion-line estate.
By James Calloway
Source: itbrief.com.au
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
722 articles
Related articles
- 87% of Enterprise Developers Use Low-Code. That Is the Agent Security Problem
- OpenAI Launches Aardvark, an Agentic AI Security Researcher
- 83% of Firms Run Unapproved AI Coding Agents, BlueFlag Research Finds
- OpenAI Launches GPT-5.2-Codex, Its Most Advanced Coding Model
- GitHub and Microsoft's AI bug scanner produced 99 verified reports across six projects