Phoenix Security launches AI code scanning strategy and Phoenix Purple

Phoenix Security on June 29 launched Phoenix Purple, a graph-native AI scanner for AI-generated code, and a new SDLC security strategy built to stop new vulnerabilities and reduce existing backlog. The company says the approach cuts scanning costs sharply while helping teams keep pace with AI agents that now generate code and pull requests faster than human review cycles can handle.

Why it matters: - AI agents are changing how software gets written, reviewed and secured. - Phoenix Security is pitching a model that treats security as part of the development workflow, not a separate bottleneck. - The company says the approach is designed to reduce both new vulnerabilities and the backlog already in the estate.

What happened: - Phoenix Security launched Phoenix Purple, a graph-native AI scanning product for AI-generated code. - The launch also introduced a new agentic SDLC security strategy built around three controls: control the agent, close the tap and burn the backlog. - The announcement was made June 29 in London.

The details: - Phoenix Purple scans AI-generated code inside the agent's working session on the graph in seconds. - The graph shows which functions are reachable, which paths carry untrusted input and which parts of the repo a change actually touches. - Phoenix says that lets a smaller model do work that would otherwise require a frontier model across the full codebase on every change. - On a modeled fleet of 26 repositories with 250,000 lines each, scanned monthly, a file-by-file harness on a frontier model cost about $598 per month and $64.30 per confirmed vulnerability. - The same modeled fleet with Phoenix Purple cost $74.88 per month and $3.60 per confirmed vulnerability. - Phoenix says that is about eight times less across the fleet and roughly 87% lower cost for the same findings on the same model. - Phoenix says the cost gap widens as codebases grow because file-by-file cost rises with every line added while graph-native cost stays tied to real findings. - Control the agent works by reading a team's backlog and turning confirmed findings into rules the agent follows every session. - Those rules are written as plain constraints in files agents already read, including .mdc, AGENTS.md and .cursorrules. - Phoenix says classes of bugs a team has already fixed stop being regenerated, and each new finding can become a new guardrail. - Close the tap uses four controls before code reaches the main branch: architectural rules and threat modeling at design, agent-level scanning in session, package intelligence at install, and a pull-request gate that re-runs the same checks. - Burn the backlog uses Phoenix Orange to aggregate every scanner into one queue, remove what is not reachable, attribute findings to their real owner and rank the rest by risk. - Phoenix says that process can turn a list of 112,000 findings into about 300 that need human attention. - Phoenix Green then reads the ranked queue and writes the fix as a pull request with threat context in the description. - Low-consequence fixes are fast-tracked, while breaking or dangerous ones wait for a human.

Between the lines: - Phoenix is arguing that more scanners are not the answer when teams already have more findings than they can act on. - The pitch reframes security spend around burn rate, with findings closed per week expected to outpace findings created per week. - The product stack is designed to be composable, so teams can adopt aggregation, agent controls, scanning or remediation separately depending on where the bottleneck is. - CEO and CISO Francesco Cipollone said the model can lower token costs while helping agents move faster and let security become a way to open the lane rather than slow it down.

What's next: - Phoenix Purple is available now. - Teams can review the product at graph-native AI scanning for AI-generated code. - Teams can estimate costs with the scan cost calculator. - Phoenix is also directing prospects to request a demo. - The company shared social links on LinkedIn, Instagram, Facebook, YouTube and X.