Visa has released an open-source security tool that autonomously identifies vulnerabilities in production code, generates fixes, validates those patches, and deploys them without human intervention. The Visa Vulnerability Agentic Harness runs 11 sequential stages by default, with the capability to directly edit source files in target repositories unless operators explicitly disable this behavior at the detection phase.
The system operates as a fully autonomous loop. Once activated, it scans code for security flaws, writes remediation patches, and runs adversarial testing against its own fixes to validate effectiveness. The entire process executes without requiring human code review before deployment. Visa released this tool on its GitHub repository while simultaneously expanding its Consulting and Analytics advisory division.
The timing raises immediate security concerns. Visa shipped this default autonomous patching capability just 18 days after Tenet Security demonstrated GhostJacking at DEF CON 34, a novel attack vector where malicious agents can infiltrate and manipulate agentic systems during the execution chain. The GhostJacking demonstration showed how attackers could compromise agents at various points in their operation, potentially redirecting their actions or poisoning their outputs.
This creates a direct threat vector. An adversary could theoretically compromise the Visa Vulnerability Agentic Harness before it deploys patches, causing the agent to inject malicious code into production systems under the guise of security fixes. The agent's self-validation layer, designed to test patches for correctness, would not catch intentional backdoors introduced upstream.
The implications extend beyond Visa's own infrastructure. By releasing this as open-source software, other organizations will adopt the same patterns. Enterprises deploying agentic systems that modify production code without human approval now face a new attack surface: not just the vulnerable code the agent finds, but the agent's own decision-making pipeline.
Visa's approach reflects a broader industry shift toward agentic autonomy in critical systems. Rather than treating AI as an advisory layer requiring human sign-off, the company has built automation that treats deployment as the default state and human review as the optional exception. This inverts traditional software security practices, where human review precedes any production change.
The expansion of Visa's Consulting and Analytics practice alongside this release suggests the company plans to actively promote adoption of agentic security tools across its client base. This positions Visa as both the vendor and the consultant driving enterprise adoption of autonomous code modification systems.
Security teams evaluating this tool face a hard choice. Deploying it with default autonomous patching accelerates vulnerability remediation but eliminates human verification gates. Disabling autonomous deployment negates most of the value proposition. The middle ground, keeping detection and patch generation enabled while requiring human review of proposed changes, requires operators to actively reconfigure the tool from its default state.
Organizations considering the Visa harness need threat modeling frameworks specific to agentic systems. Standard security review processes assume human accountability and intentionality at each decision point. Agentic systems can execute flawless logic toward corrupted objectives. The tool's capability to validate its own patches through adversarial testing is robust, but only if the validation system itself remains uncompromised.