authorization

Bilingual

Authorize, Don't Authenticate
2 months ago
- The author argues for an 'authorize, don't authenticate' model where users grant applications access to their own personal databases rather than logging into centralized apps.
- Applications should store minimal data outside the user's database, allowing users to revoke access at any time and maintain control over their data.
- Creating personal databases should be as simple as creating a document, with tools like ayb making it easy for non-technical users.
- Separating applications from databases offers user choice in data hosting and reduces reliance on single application providers, though trust is transferred to database hosts.
- Collaborative and social data require more complex models, but the author envisions decentralized aggregation without centralized controllers.
- The author's projects (ayb, Todos) demonstrate practical implementation using OAuth2 and static frontends, enabling user data agency.
MCP 2026-07-28 Specification: transport going stateless
2 months ago
- The 2026-07-28 Model Context Protocol (MCP) specification introduces a stateless protocol core, transforming MCP from a bidirectional stateful protocol into a request/response stateless one for improved reliability and scalability.
- Every request is self-describing with optional discovery; methods and tool names travel in HTTP headers for routing and authorization.
- Multi Round-Trip Requests (MRTR) replace server-initiated bidirectional streams, enabling stateless interactions like user input requests mid-call.
- List responses now include cache hints (ttlMs, cacheScope) for better caching of tool catalogs and prompt lists.
- Authorization hardening includes RFC 9207 issuer validation, client metadata documents (CIMD) replacing Dynamic Client Registration (DCR), and client credentials bound to issuers.
- Tasks move out of experimental core into a formal extension; Roots, Sampling, Logging, and legacy HTTP+SSE transport are deprecated with a 12-month window.
- Updated TypeScript, Python, Go, C# SDKs are available with migration notes; Rust SDK supports the new spec in beta.
I scanned my AI agent framework for destructive/consequential actions, and wow
2 months ago
- A static scan of 25 AI agent frameworks found 30 instances where model-controlled parameters could reach consequential actions without authorization checks.
- The scanner distinguishes between validation (checking form) and authorization (checking permission for the specific action), catching patterns like network egress, file writes, shell execution, and SQL injection.
- The study had 81% precision on its first run; 12 false positives were found and fixed, with regression tests added to prevent recurrence.
- The scanner covers frameworks like MCP, LangChain, and OpenAI but not custom agent loops or action/observation dispatchers.
- The scanner is publicly available with no dependencies or network calls, and can be run on local code with a simple command.
AI Authentication and Authorization
2 months ago
- AI security relies on existing identity and authorization patterns from the API boom.
- Three AI use cases discussed: RAG, tool use (MCP/APIs), and agentic systems.
- RAG requires authorization to filter documents before they reach the LLM.
- Tool use involves controlling AI access to APIs and services via authentication methods like OAuth 2.1.
- Agentic systems need a chain of identity to track human authorization through workflows.
- Deterministic identity enforcement is crucial for probabilistic AI systems.
- FusionAuth examples illustrate implementation with fine-grained authorization and audit logging.

authorization

Authorize, Don't Authenticate

MCP 2026-07-28 Specification: transport going stateless

I scanned my AI agent framework for destructive/consequential actions, and wow

AI Authentication and Authorization