Speaker
Abstract
Sandboxes are the first line of defence for agentic systems: restrict the bash commands, filter the URLs, lock down the filesystem. But sandboxes operate on the syntax of requests, not the semantics of your authorization model. They don't know whether GET /customer/id_1234 is in scope for the task at hand, and they can't tell the downstream system whether the request is coming from a human, a delegated agent, or a rogue process. The same applies to MCP tool calls — allowing read_file or create_issue tells you nothing about whether those actions make sense in context.
MCP has given the industry a common protocol for tool calls: and in doing so, made agent authorization problems impossible to ignore. As auth lead for MCP at Anthropic, these problems aren't theoretical for me. They come up in enterprise deployments, in working group discussions, and in the gap between what the spec says and what teams actually need to build. This talk isn't about MCP specifically: it's about the authorization primitives that any agentic system needs, and the standards work needed to provide them.
We'll examine four quirks that make agent authorization hard: granularity (scopes are too coarse for resource-level enforcement), attribution (is this the user, a delegated agent, or a service account?), information leakage (coarse tokens let agents see more than the task requires), and cardinality (consent flows don't scale to ephemeral agent sessions).
A key design decision sits at the heart of all of this: should your agent act as a delegated extension of the user, or as its own service account? The answer has deep implications for auditability, blast radius, and what standards you need.
Then we'll do a whistle-stop of the emerging standards landscape trying to close these gaps — Rich Authorization Requests and draft RAR metadata extensions, Cross-App Access (XAA) for frictionless token acquisition, and how patterns from SPIFFE and Workload Identity Federation might translate into agent identity. None of these fully solve the problem alone, but together they sketch the shape of what's missing.
QCon London 2026 is a three day conference for senior software engineers, architects and team leads. An international program committee of working engineers selects every session. Patterns and practices, not products and pitches.
Part of the track
Architecture in the Age of AI Hosted by Fabiane Nardon Data Expert, Java Champion & Data Platform Director @totvsFrom the same track
Wednesday 18 March
10:35 Fleming (3rd Fl.) Session agentic coding The Right 300 Tokens Beat 100k Noisy Ones: The Architecture of Context Engineering Patrick Debois, Baruch Sadogursky Your agent has 100k tokens of context. It still forgets what you told it two messages ago. 11:45 Fleming (3rd Fl.) Session AI/ML Beyond Benchmarks: How Evaluations Ensure Safety at Scale in LLM Applications Clara Matos Director of Applied AI @Sword Health, Focused on Building and Scaling Machine Learning Systems As LLM systems move from prototypes to production, the gap between benchmark performance and real-world reliability becomes impossible to ignore. Models that score well on benchmarks can still fail unpredictably when facing the complexity, ambiguity, and edge cases of real users. 13:35 Whittle (3rd Fl.) Session data platform engineering Building an AI Ready Global Scale Data Platform George Peter Hantzaras Engineering Director, Core Platforms @MongoDB, Open Source Ambassador, Published Author As organizations move from single-cloud setups to hybrid and multi-cloud strategies, they are under pressure to build data platforms that are both globally available and AI-ready. 14:45 Whittle (3rd Fl.) Session Your Agent Sandbox Doesn't Know My Authz Model: A Standard-Shaped Hole Paul Carleton Member of Technical Staff @Anthropic, Core Maintainer of MCP Sandboxes are the first line of defence for agentic systems: restrict the bash commands, filter the URLs, lock down the filesystem. But sandboxes operate on the syntax of requests, not the semantics of your authorization model. 15:55 Whittle (3rd Fl.) Session Explicit Semantics for AI Applications: Ontologies in Practice Jesús Barrasa Field CTO for AI @Neo4j Modern AI applications struggle not because of a lack of models, but because meaning is implicit, fragmented, and brittle. In this talk, we’ll explore how making semantics explicit (using ontologies and knowledge graphs) changes how we design, build, and operate AI systems.