CodeWall Raises $4M Pre-Seed to Bring Autonomous Attack Agents to Enterprise Security

A security team can patch a vulnerability and still leave an attacker another route to the same data. The harder question is whether weaknesses across applications, identities, and infrastructure can be combined into a working intrusion.

CodeWall is building its business around answering that question continuously. The London startup has emerged from stealth with $4 million in pre-seed funding led by Osney Capital, with participation from InReach Ventures, Antler, and angel investors, according to its October 14 announcement.

The company develops autonomous attack agents that approach an enterprise from the outside, without requiring credentials, source code, or internal access. Its pitch is to give defenders an ongoing way to discover and verify exploitable paths before a malicious actor uses them.

From a list of weaknesses to evidence of an attack path

CodeWall’s technical vision centers on adapting an investigation as new evidence appears. The company argues that scarce offensive-security expertise and periodic assessments leave organizations struggling to test an environment that changes between engagements.

Its announcement describes specialist agents investigating in parallel and sharing discoveries through a persistent investigation graph. A finding in one application can inform the next agent’s examination of a connected service, identity, or data store. The intended output is an account of what can be compromised, how the path works, and the evidence supporting that conclusion.

In practical terms, a graph represents relationships between findings rather than treating each alert as an isolated item. A connection may matter more than the individual weaknesses: exposure at one boundary could make a second control reachable. This explanation describes the architectural idea; CodeWall has not disclosed enough implementation detail in the announcement to evaluate how its graph resolves conflicting findings or validates every transition.

The company website says the platform maps trust relationships across services, APIs, and infrastructure, adapts its strategy during investigations, and produces reproducible proof-of-concept evidence. Those are product claims. Their value for an enterprise depends on whether findings remain reproducible and meaningful in that enterprise’s environment.

What its McKinsey research illustrates

CodeWall has already published research describing weaknesses in platforms operated by major consulting companies. Its March 9 account of testing McKinsey’s Lilli AI platform is particularly relevant to enterprises connecting AI applications with internal knowledge.

According to CodeWall, its agent obtained read and write access to the production database in under two hours. The company attributed the entry point to an unauthenticated interface and a SQL injection issue involving how request field names were handled. The significance was the reach into the application’s underlying data and configuration, rather than a failure of the language model itself.

The researchers also discussed the implications of write access to stored system prompts: compromising application infrastructure could create a route to changing the instructions that govern trusted AI outputs. That is a potential consequence described in the report, not evidence that consultants received manipulated advice.

CodeWall says it disclosed the findings to McKinsey, limited access to verification, and confirmed remediation before publication. These details come from the vendor’s research account; they should not be read as evidence of an ongoing vulnerability or a general benchmark proving its agents outperform human testers.

Continuous testing changes the operational challenge

CodeWall says its platform revisits systems as applications change, new assets appear, or fixes are introduced, operating within an authorized scope. That creates a potentially useful feedback loop: identify a path, address the contributing weaknesses, and test whether the path still works.

Retesting matters because the closure of an individual ticket does not necessarily establish that the original business risk disappeared. A fix may block one route while leaving another available, or a later deployment may undo an earlier correction.

Unite.AI’s earlier coverage of autonomous penetration testing and remediation explored the same movement toward repeated validation. CodeWall’s emphasis is its external starting point and coordinated investigation of connected systems. Similar commercial objectives do not imply identical agent architectures or execution controls.

More frequent testing also increases the importance of operational boundaries. Enterprises evaluating such a platform should establish permitted targets and actions, evidence-handling rules, escalation procedures, and the ability to stop an investigation. These are evaluation criteria, rather than capabilities established by CodeWall’s funding announcement.

What the funding will build

Founder and CEO Paul Price positions the company as a way to make offensive testing a permanent defensive capability. CodeWall says it is already helping banks and financial institutions identify and verify exploitable attack paths, although the announcement does not provide named financial-sector customers or independent performance measurements.

The funding will support expansion of the environments its agents can investigate, together with sustained testing, verification, and remediation workflows. The company also plans to grow its offensive-security, engineering, customer-success, and go-to-market teams in London.

For buyers, an important distinction will be how far those remediation workflows extend. Producing evidence and recommending a correction differs from making a production change automatically. The announcement does not establish that CodeWall independently deploys fixes, so enterprises should clarify where approval and execution sit.

The strongest measure of progress will be the quality of the resulting security decisions: reproducible paths, clear business impact, appropriate handling of sensitive evidence, and confirmation that remediation closes the exposure. CodeWall’s pre-seed round gives it resources to pursue that operating model. Its task now is to make continuous adversarial testing dependable enough to become routine enterprise work.