AI-Assisted Review Finds 100-Plus Serious Flaws in UK Government Code
Breach & Incident Intelligence · 4 sources · Updated 4d ago · First seen Nov 10, 2026

ControlPlane researcher Chris Nesbitt-Smith identified more than 100 serious vulnerabilities while reviewing 349 publicly available codebases across 33 UK public-sector organisations using consumer versions of Anthropic’s Claude Opus and Sonnet models. The most serious finding was an exposed Planning Inspectorate application credential that could provide tenant-wide read-write access to SharePoint and access to employees’ private OneDrive accounts, although the integration needed only a single HR file. The credential remained usable for approximately 18 months after its removal from a repository branch. The agency said it revoked the credential within 20 minutes of receiving the report and found no evidence of a data breach.
Other findings included authentication and authorisation weaknesses in Department for Education services and an exposed administrator token on a Scottish Government portal. All highlighted issues were disclosed and addressed. Nesbitt-Smith said routine IT health checks could have detected the weaknesses and warned that withdrawing public source code would hinder researchers without protecting against adversaries who had already copied it. Security experts recommended least-privilege access, stronger identity controls, secret scanning and secure development practices, rather than closing public repositories; exposed credentials must be revoked, not merely removed from source code.


