Die Suchmaschine für Unternehmensdaten in Europa

UK-Förderung (32.000 £): VibeCheck: Intentsbewusste Sicherheitszusicherung für Vibe Coding Ukri 01.04.2026 Forschung und Innovation im Vereinigten Königreich, Großbritannien

Auf einen Blick

Text

VibeCheck: Intentsbewusste Sicherheitszusicherung für Vibe Coding

Zusammenfassung The increasing use of AI-assisted coding tools, such as GitHub Copilot and other generative frameworks, is transforming how software is designed and developed. Developers are increasingly using natural language prompts to guide code generation, a practice often referred to as _vibe coding_. While this approach significantly improve productivity and accessbility, it also introduces hidden security risks. AI-generated code can contain hidden vulnerabilities, insecure defaults, or missing safeguards that are difficult to detect using traditional security tools. These challenges are further amplified as organisations deploy AI-driven agents that autonomously generate and modify code with limited human oversight. _VibeCheck_ addresses this emerging challenge by providing a security assurance framework specifically designed for prompt-driven and AI-assisted software development. The project will develop a system that continuously evaluates AI-generated code to identify security weaknesses, unsafe programming practices, and failure modes unique to generative development. Crucially, _VibeCheck_ considers not only the generated code itself but also the intent expressed by developers through prompts, helping to detect mismatches between expected behaviour and actual implementation._VibeCheck_ integrates seamlessly with modern development workflows, including CI/CD pipelines and automated coding agents, enabling security and integrity checks to be applied before software is deployed. The framework combines three complementary approaches: * **Zero-trust assurance,** where AI-generated code is treated as untrusted until it satisfies explicit security and integrity criteria. * **Anomaly-based detection**, which identifies unexpected or risky coding behaviour by comparing AI-generated outputs against established patterns of secure software design practices. * **Adaptive learning**, allowing the system to improve over time by learning from new vulnerabilities, developer feedback, and changes in AI coding practices, ensuring the system stays effective as AI tools evolve. The innovation lies in creating a practical, AI-native security solution that enables organisations to safely benefit from AI-assisted coding. Existing tools are largely designed for human-written code and lack awareness of prompt-driven development, generative uncertainty, and autonomous agents. _VibeCheck_ fills this gap by enabling continuous, context-aware security assurance without undermining the productivity benefits that drive AI adoption. The project outcomes will benefit software developers, security teams, and organisations adopting AI-assisted coding by reducing security risks while preserving development speed and flexibility. By supporting safer and more responsible use of generative coding technologies, VibeCheck contributes to more secure software, stronger digital infrastructure, and greater trust in AI-enabled systems.
Kategorie Collaborative R&D
Referenz 10193042
Status Active
Laufzeit von 01.04.2026
Laufzeit bis 31.07.2026
Fördersumme 32.000,00 £
Quelle https://gtr.ukri.org/projects?ref=10193042

Beteiligte Organisationen

NEWCASTLE UNIVERSITY

32.000,00 £

Die Bekanntmachung bezieht sich auf einen vergangenen Zeitpunkt, und spiegelt nicht notwendigerweise den heutigen Stand wider. Der letzte verfügbare Stand wird auf folgender Seite wiedergegeben: Newcastle University, Newcastle Upon Tyne, Großbritannien.