25% of Y Combinator Winter 2025 startups had codebases that were 95%+ AI-generated. After 40 iterative AI edits, those codebases carried 37% more vulnerabilities — with no error thrown, no test failing, and no developer aware. Speed without quality assurance is a liability, not an advantage.
45%
of AI-generated code introduces a security vulnerability
Tested across 100+ large language models on 80 real-world coding tasks — OWASP Top 10 flaws including injection, broken access control and hardcoded credentials.
Veracode GenAI Code Security Report, 2025
62%
of AI-generated solutions contain security flaws or architectural defects
Even when using the latest foundation models. Authorization failures, missing access controls and design-level vulnerabilities dominate the findings.
Cloud Security Alliance, July 2025
10×
more security findings introduced by AI-assisted developers
AI-assisted teams wrote 3–4× more code but produced ten times as many security issues. The velocity gain is real; so is the risk multiplication.
Apiiro, June 2025
37%
more vulnerabilities after 40 iterative AI modifications
Iterative AI-assisted development — the normal startup workflow — degrades codebase integrity progressively. Each round of changes compounds the risk.
Iterative Degradation Study, 2025
32%
of teams cite quality as their #1 blocker in production AI agents
Quality tops the list above latency, security and cost. Hallucinations and inconsistent outputs are the most-cited specific failure modes.
LangChain State of AI Agents, 2026
88%
of enterprise AI agent pilots never reach production
Blocked by governance, compliance, data quality and integration gaps — not by model capability. The model works; the system around it doesn't.
Enterprise AI Deployment Analysis, 2026
40%+
of agentic AI projects will be cancelled
Quality failures, hallucinations and uncontrolled agent behaviour are the leading causes. Organisations that invest in validation and oversight survive the cull.
Gartner, 2026
21%
of organisations have a mature governance model for agentic AI
The other 79% are running autonomous systems with inadequate oversight. Only 35% of executives say they could immediately stop a rogue agent.
Deloitte State of AI in the Enterprise, 2026