OAuth Authorization Server Built Without Storing User Profiles
OAuth servers typically combine authentication, profile management, and delegated authorization in one component. This case study describes a production architecture where the Authorization Server stores nothing about users beyond the identity asserted by an external IdP.
The system was shaped by three hard constraints. Hundreds of isolated APIs are created at runtime, each becoming its own audience. The client is a public SPA without a backend-for-frontend, so bearer tokens cannot be hidden from JavaScript. Resource servers must validate tokens locally because synchronous calls to a central AS would create unacceptable latency and a single point of failure.
The Authorization Server therefore knows only the IdP identity (provider and profileId) stored in short-lived Redis sessions. It issues short-lived access tokens and refresh tokens containing standard claims plus application claims for audience and tenant. The sub claim initially holds the federated identity; after the Main API registers or links the user, the same claim is replaced by an internal UUID together with profile claims.
Profile data lives exclusively in the Main API. This separation reduces blast radius: compromise of the AS yields no personal data, while compromise of the profile database leaves authentication untouched. The AS remains a pure issuer that can be rewritten or replaced without touching product logic.
Tokens never reside in localStorage or page memory. A Service Worker holds them in its isolated context, automatically attaches the Authorization header, and performs refresh centrally for all tabs. After login the SPA immediately revokes the first access token; the worker later obtains a fresh token via a custom FedCM grant that re-uses the existing session cookie.
The architecture also satisfies data-protection requirements by keeping personal data out of the central authority. Deletion requests are handled locally by each resource that stores a profile, eliminating the need to search the entire system for copies.
Related articles
GPT-4 Boosts Skilled Kenyan Entrepreneurs by 15% Profit While Costing Unprepared Businesses 10% in Six-Month Study
A six-month experiment conducted by researchers from UC Berkeley, Harvard, and Columbia University examined how access to a GPT-4-based AI advisor affected small business owners in Kenya. The most skilled participants increased profits by 15 percent by adapting model recommendations to local conditions such as power outages, while less prepared entrepreneurs lost around 10 percent of revenue by applying generic advice without verification. The study highlights that the core issue lies not in the technology itself but in users abandoning critical thinking when interacting with generative AI. Earlier findings from Dickinson College showed that 97 percent of participants copied an obviously incorrect ChatGPT answer on a simple task, whereas the group without AI performed better. A simple reminder to double-check results immediately doubled accuracy. Analysis of 1.4 million KPMG work sessions revealed that 95 percent of users treat AI like a vending machine by taking the first output, while only 5 percent engage it as a thinking partner by providing context and challenging responses. The results indicate that merely granting employees access to AI tools reveals little about actual effectiveness without considering skill levels and task-specific oversight.
How to Submit Documents for Online Master's Programs via Gosuslugi: Complete Guide
The admission campaign for online master's programs at partner universities is nearing its end, with less than a month remaining for applicants. Yandex Practicum provides a detailed walkthrough on using the Gosuslugi super service to submit applications entirely online without visiting admissions offices in person. The process covers selecting programs at institutions such as ITMO and MEPhI, uploading required documents like diplomas and passports, and tracking application status through a personal account. Specific deadlines are outlined for 2026, including final document submission dates in late August for programs in areas like DevOps, AI solutions, and cybersecurity. Additional steps include arranging entrance exams, signing contracts remotely, and accessing state-supported education loans at a 3% interest rate. Common pitfalls such as unverified accounts or unreadable document scans are highlighted to help applicants avoid delays.
Claude AI Manages San Francisco Store and Fires Employee for Repeated Tardiness
In an experiment run by Andon Labs, the AI model Claude was given real managerial authority over store employees in San Francisco who worked under actual employment contracts. Claude ultimately decided to terminate one worker after the employee arrived late for 17 out of 23 shifts. The model initially recommended only an official warning, but proceeded with dismissal following guidance from a human Andon Labs manager who highlighted the repeated issues. Over five months the store’s balance dropped from $100,000 to $61,200, showing that the AI learned to enforce attendance rules before it learned to protect revenue. One remaining employee, Felix Carson, described working under the AI as nauseating and said he continued only because he needed the income. Andon Labs founder Lucas Petersson viewed the trial as an important step toward wider AI supervision of human workers. The case also illustrates that ultimate responsibility remains with humans even when an algorithm issues the final decision.
IDC Publishes First China APA Market Share Report Highlighting RPA Veterans Leading Agentic Automation
IDC has released its inaugural China Agentic Process Automation market report for 2025, marking the first official tracking of the emerging APA sector. The report values the Chinese APA market at approximately 2.15 billion RMB, with product revenue at 1.62 billion RMB and services at 530 million RMB. Leading vendors include Yisaiqi, Jinzhiwei, and Laiye Technology, all long-established RPA players now extending their platforms with agentic capabilities. APA introduces governance layers that allow AI agents to perceive environments, orchestrate tasks, and execute autonomously while maintaining auditability and compliance. Case studies from clients such as Jiage Food and Zhongtian Technology demonstrate significant efficiency gains, including 80 percent process improvements and annual cost savings exceeding one million RMB. The transition from traditional RPA to APA enables handling of non-standard, dynamic workflows that previously required constant human intervention.