Which AI models fail 1 in 3 security tasks?

Four years of data. Over 100 AI models total. Zero signs of a security breakthrough.

Report   2026 GenAI Code Security Report   Veracode tested 11 new models across 80 tasks. See every ranking and what it means for your team.   Download the Report     100+ AI models. Four years of data. The average security pass rate for AI-generated code is 56% and hasn't moved. GPT-5.5 leads the Summer 2026 dataset at a 68% security pass rate. Six of eleven models cluster between 50-53%. Even the best model still fails nearly 1 in 3 security tasks. The worst model fails 1 in 2. AI-generated code volume has gone up sharply while the failure rate hasn't come down. That gap is yours to manage. The 2026 GenAI Code Security Report breaks it down model by model, language by language. See what the data means for your software and what you can do about it.     Download the Report       36 Queen Street London, EC4R 1BN United Kingdom     emea@veracode.com      +44 (0)20 3761 5501 © 2026 Veracode Inc. All rights reserved. Unsubscribe  |  Manage preference