Analysis
CodeWall emerged from stealth with $4 million in pre-seed funding led by Osney Capital, with InReach Ventures, Antler and angel investors participating, to build autonomous agents that probe a company's defenses from the outside, without credentials or privileged access, according to Pulse 2.0. The London-based company, founded by CEO Paul Price, runs these tests continuously and presents the results as evidence of what an attacker could actually compromise, rather than a periodic report.
"Enterprises shouldn't have to wait for a real attacker to find out whether their defenses hold," Price said. Osney Capital partner Paul Wilkes framed the bet more bluntly: "We believe there's a future where having an attacker working for you is as fundamental as having defenses in place."
CodeWall arrives in a category that's already well-funded and well ahead of it:
“- Harmonic Security -- $50M Series B (October), to govern how employees use AI tools.”
- Pluto Security -- $42M combined seed-and-Series A round (October), to police enterprise AI agents.
- Harmonic Security -- $50M Series B (October), to govern how employees use AI tools.
- Reach Security -- $50M+ Series B (October), to defend against AI-driven attacks.
CodeWall's pitch is narrower than any of theirs: not governing AI agent behavior or employee AI use, but continuously red-teaming the attack surface itself, closer to a product like Pentera or a bug-bounty program than an AI-governance platform. The one customer cited at launch, Axos Financial CISO Raghu Valipireddy, said the appeal is speed: "As AI changes what attackers are capable of, we think it's important to be adopting these technologies early."
However, that's a single reference customer at launch, not a roster, and a $4 million check buys CodeWall limited runway to prove the continuous-testing model works at scale before a larger, better-capitalized rival builds the same feature as an add-on. What happens next depends less on the technology than on whether enterprises will grant an autonomous agent the same access a real attacker would need to test -- a trust threshold none of CodeWall's better-funded peers have fully solved either.