Low-Code AI Used to Build Surveillance Tool in Hours

The PCMag reporter began with a $20-per-month ChatGPT Plus account to access Codex, and without any prior coding skills, created the functional surveillance dashboard. The resulting tool could aggregate disparate, publicly available live camera feeds from cities globally into a single interface, demonstrating the power to assemble "scattered, individually innocuous data into a comprehensive picture." This experiment is notable as it followed a controversy where AI company Anthropic refused a US Department of Defense contract due to surveillance concerns. OpenAI subsequently accepted the contract, stating its policies protect against misuse of its technology. This highlights the ongoing debate around the dual-use nature of powerful AI development tools. The rapid development raises significant security concerns inherent in low-code platforms, often termed "shadow IT," where employees can build and deploy applications without IT oversight. This can bypass traditional security protocols like peer review and version control, potentially introducing vulnerabilities. Enterprise-grade low-code platforms aim to mitigate these risks by integrating security features like role-based access control (RBAC), multi-factor authentication (MFA), and encryption. To address security at scale, OpenAI has also developed Codex Security, an AI agent designed to audit codebases for vulnerabilities. In early tests, the tool scanned over 1.2 million code commits and identified 792 critical and over 10,500 high-severity findings, demonstrating its capacity to automate parts of the security review process that is becoming a bottleneck in fast-paced development. This trend of rapid, AI-powered tool creation is mirrored in newsrooms, where adoption is accelerating. A 2025 survey of media leaders revealed that 77% see AI-assisted content creation (with human oversight) as important, and 96% prioritize it for back-end automation like transcription and tagging. In the UK, 56% of journalists now use AI at least weekly for professional tasks. The proliferation of AI-generated video content places immense strain on technical infrastructure. A key challenge is GPU contention, where the same processors are used for both AI model inference and traditional video encoding/decoding. This bottleneck can diminish AI performance as video workloads increase. To manage this, scalable architectures often offload video transcoding to dedicated video processing units (VPUs), freeing up expensive GPUs to focus solely on AI tasks. Further strategies include horizontal scaling, where high-volume video is split into smaller chunks for parallel processing, and leveraging cloud infrastructure with auto-scaling capabilities to handle traffic spikes efficiently.