A recent investigation led by security researcher Chris Nesbitt-Smith has revealed more than 100 significant vulnerabilities within publicly accessible UK government code. These vulnerabilities, which could potentially allow an attacker read-write access to a public body’s entire Microsoft 365 estate, bring to light a contentious issue within Whitehall: whether the government should continue to publish its code openly. This unsettling revelation was initially reported by WIRED and has stoked the fires of debate surrounding transparency and security in government coding practices.
Nesbitt-Smith, affiliated with ControlPlane, meticulously examined 349 codebases from 33 public-sector organizations and uncovered serious weaknesses in approximately two-thirds of them. The nature of these issues was alarmingly familiar—exposed credentials, insufficient access controls, overly powerful service accounts, and insecure defaults. What sets this investigation apart is the method used: consumer-grade AI models from Anthropic, run on a standard laptop. The researcher has emphasized that many of these vulnerabilities could have been detected through routine IT health checks. Notably, all identified issues have since been reported to the respective organizations and remedied.
Olly Maxwell, a principal security operations analyst at Huntress, expressed despair viewing the findings. Transitioning from his role in the Foreign, Commonwealth & Development Office to tracking malware threats, Maxwell noted the all-too-familiar political reaction within Whitehall following such revelations. He insinuated that the immediate response often involves restricting open repositories and retreating from transparent coding practices. “That reflex shows a fundamental misunderstanding of both the underlying causes of these exposures and the realities of the current threat landscape,” he asserted.
Among the vulnerabilities discovered, one concerning case involved the Planning Inspectorate, an agency responsible for handling planning appeals and national infrastructure projects in England. An Azure Synapse notebook was inadvertently exposed on Github, containing extensive Microsoft Entra ID application credentials. Despite being intended for a limited task—retrieving a single HR file from SharePoint—the permissions associated with these credentials extended access across all SharePoint sites and intruded into private employee OneDrive accounts, including crucial libraries under legal hold for statutory planning inquiries. Maxwell attributed this severe lapse to a crucial breakdown in governance and basic security hygiene, rather than a failure of open-source philosophy.
Further insight from Collin Hogue-Spears, a senior director of solution management at Black Duck, pointed out that the permissions model was essential in transforming a leaked secret into a possible disaster. “The weaknesses stem from a lack of implementing least privilege and implicit trust controls. The excessive permissions allowed the Planning Inspectorate leak to pose a significant risk,” he remarked.
The security implications of these vulnerabilities are further complicated by the fact that while the credentials were removed from the repository branch, they were not revoked from the identity provider. An orphaned commit lingered on GitHub for nearly 18 months, allowing the credentials to remain valid even after the repository’s history was altered. Subsequent to the researcher’s report that identified the vulnerability, the agency took swift action, revoking the credential about 20 minutes after notification.
Ronald Lewis, head of cybersecurity governance at Black Duck, highlighted that this incident illustrates a fundamental flaw in security design. He argued that the core issue is not the public nature of the code but rather an inherent reliance on obscurity for security. Lewis pointed out that while it might be tempting to blame AI for uncovering these weaknesses, the real concern lies in the architectural choices that allowed the vulnerabilities to exist in the first place. He noted, “AI simply made these weaknesses easier and faster to identify.”
The use of AI in identifying these vulnerabilities arrives at a pivotal moment when the UK government itself has leaned on advanced models to uncover flaws. Noteworthy is the Government Cyber Coordination Centre’s publication of findings from internal hackathons earlier this year, which identified 407 vulnerabilities across various departments. Maxwell suggests that the revelations from Nesbitt-Smith’s investigation are a wake-up call regarding the evolving threat landscape.
He emphasized that the ease with which an independent researcher could expose over 100 flaws using consumer-grade AI underscores a significant threat asymmetry. “If an independent researcher can surface these gaps with minimal effort, one must assume that adversaries are actively mirroring these repositories and harvesting exposed credentials,” he warned.
Additionally, on a broader scale, Hogue-Spears pointed to instances where attackers are actively utilizing AI to search public repositories for leaked application secrets. This situation emphasizes the need for vigilance from organizations. Attacks may not happen immediately; skilled perpetrators can store valid access keys and quietly maintain a foothold within sensitive environments until they choose to exploit their access.
The response from security experts is clear: an urgent shift in mindset is needed within organizations. While there is instinctive pressure to restrict open-source repositories in light of such findings, experts argue that this approach does not address the underlying flaws. A concerted effort should focus on implementing robust automated security measures throughout the software supply chain, ensuring that security best practices are embedded into development processes.
Maxwell concludes with a pertinent reminder about the continuous threat landscape. “In a climate where the UK faces sustained hostile cyber activities, the government should remain transparent while also mandating stringent automated security protocols,” he urged.
As the debate over the future of open code continues, one truth remains certain: merely closing repositories won’t eliminate these vulnerabilities. What is essential is that governments and organizations alike adopt a proactive approach in their cybersecurity strategies to safeguard against future threats.