Urgent.News

What's breaking now, across thousands of outlets.

AI

Authenticated Delegation Between Autonomous AI Agents

SGAEIA Research Series — Article 4 of 8 Aridio Silva · Independent Researcher, Brazil · ORCID Delegating a task is not the same as delegating authority. When one autonomous AI agent asks another agent to perform consequential work, the important question is not only who sent the message? It is also: under whose authority is the action being requested, within which scope, for how long, with which…

Delegating tasks to other AI agents is distinct from granting authority for consequential actions. When one autonomous agent requests another to perform work, the critical questions are who initiated the request and under whose authority, scope, constraints, and limitations. This technical article expands on the original Medium post, focusing on developer and architect perspectives without altering the core message.

The key differences between task flow and authority flow are emphasized. A mental model illustrates how an orchestration agent should not silently expand the deployment agent's authority when delegating a task. The article outlines seven properties essential for trustworthy delegation: maintaining a legitimate authority origin, identifiable actors and workloads, bounded purpose and scope, applicable context and temporal validity, explicit delegation limitations, revocation capability, and sufficient provenance for verification.

These properties are technology-agnostic and can be implemented using various existing mechanisms. Delegation must not unintentionally expand authority, which is captured by the architectural invariant: A_delegate ⊆ A_delegator. Developers should model specific threats associated with delegation, such as replay attacks, resource substitution, confused deputy behavior, and unauthorized re-delegation.

Ensuring proper revocation mechanisms is crucial, as delegated authority must be revocable to prevent dangerous asymmetries. Provenance tracking is vital for multi-hop delegation scenarios, enabling verification of authority origins and actions. While cryptography ensures message authenticity and integrity, it alone cannot determine whether an action is authorized under current conditions.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

THIS IS NOT AN ORDINARY TALK SHOW

*This film image is from L'Étrange 2025(The Stranger), Gaumont. In August 2026, several of my ChatGPT conversations became unavailable.

  • Ariel's ChatGPT data vanished on August 13-14
  • Support representatives showed sympathy but shifted focus to tracking issue
  • Head of User Ops Glen Worthington emphasized user-first perspective

Persistent Teams Are Great — But What Happens When a 50-Step Workflow Breaks at Step 37?

GitHub Trending today tells a clear story: the Agent ecosystem is moving from "single-shot execution" toward "long-cycle, persistent tasks." mvschwarz/openrig is building persistent agent teams with…

  • Persistent teams focus on team formation and methodology, not workflow stability.
  • 50-step workflows can break at any step, causing data loss or restarts.
  • Astron-agent addresses workflow resilience, handling failures without losing progress.

Meta wants your next gadget to be Muse-infused

Meta wants Muse in your TV and your toaster, so it's giving the code away for free.

  • Meta unveils Muse Gadgets, open-source project for AI agent integration.
  • Provides open-source firmware, Linux SDK, and project ideas.
  • Muse Home Link connects to home networks and smart devices.

Japan's entertainment robots diversify with AI, offer comfort

OSAKA (Kyodo) -- The range of home entertainment robots is becoming increasingly diverse in Japan, with some providing emotional comfort like pets, an

  • Japanese entertainment robots evolve with AI, offering emotional comfort.
  • Sharp's Poketomo, a meerkat-like robot, debuts in December.
  • Casio's Moflin becomes expressive with increased affection.

Inside My llama.cpp Setup: Tuning Qwen 3.8 27B for 512K Context

Understanding My llama.cpp Qwen 3.8 Configuration I've been tuning llama.cpp for local AI development, and the command line can quickly become a collection of cryptic flags.

  • User runs Qwen 3.8 27B model on MacBook Pro M5 with 128 GB RAM
  • Implements speculative decoding with 8-token MTP prediction
  • Sets 524,288-token context window for 512K capacity

More from Saturday 3 October →