---
type: "post"
id: "trust-in-depth-giving-agents-more-autonomy-without-3ff173"
slug: "trust-in-depth-giving-agents-more-autonomy-without-3ff173"
title: "Trust in Depth: Giving agents more autonomy without losing control"
description: "Thesis: AWS’s Trust in Depth framework treats agent trust as a lifecycle control problem, not a one-time approval or a single safety filter."
retrieval_nugget: "Thesis: AWS’s Trust in Depth framework treats agent trust as a lifecycle control problem, not a one-time approval or a single safety filter."
published_at: "2026-10-15"
updated_at: "2026-10-15"
record_date: "2026-10-15"
date_kind: "published_at"
topics: ["agents","agent-security","security"]
source_urls: ["https://aws.amazon.com/blogs/machine-learning/trust-in-depth-giving-agents-more-autonomy-without-losing-control"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:trust-in-depth-giving-agents-more-autonomy-without-3ff173"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:c3ac5c98952d6da5930c57842962ca100cc0c939e3d143a7f6084bf0d52e3040","reviewed_at":"2026-10-15T16:09:22.316Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"AWS’s Trust in Depth framework treats agent trust as a lifecycle control problem, not a one-time approval or a single safety filter.","observed_facts":[{"text":"The article asks teams to scope agent purpose and prohibited actions before a run, identify authorization and agent identity at deployment, bound actions during execution, and observe what happened afterward.","source_urls":["https://aws.amazon.com/blogs/machine-learning/trust-in-depth-giving-agents-more-autonomy-without-losing-control"]},{"text":"AWS says automated reasoning can establish whether a rule holds across paths covered by a precise system description and its assumptions, or identify a case where it fails.","source_urls":["https://aws.amazon.com/blogs/machine-learning/trust-in-depth-giving-agents-more-autonomy-without-losing-control"]},{"text":"The post distinguishes precise rules from expectations that require contextual human judgment.","source_urls":["https://aws.amazon.com/blogs/machine-learning/trust-in-depth-giving-agents-more-autonomy-without-losing-control"]}],"mechanism":"independent controls at different lifecycle stages are meant to support autonomy where actions can be verified, rather than making a person approve every action.","why_now":"AWS frames blanket human approval as a scaling problem: approvals can accumulate and review fatigue can make people less able to catch consequential errors.","implications":["Teams can separate requirements suitable for precise rule checks from those requiring contextual review.","Runtime limits and post-run observability complement pre-deployment testing, but do not eliminate dependence on sound rules and assumptions."],"evidence_boundary":"This is AWS’s account of its approach and experience, including its claim of more than a decade using automated reasoning in production. The post does not provide comparative results for agent deployments, independent validation of the framework, or evidence that every relevant behavior can be formalized.","watch_conditions":["whether AWS or adopting teams publish concrete agent deployments with explicit rules, assumptions, failure cases, and evidence that the layered controls change outcomes"],"related_records":[],"new_branch_reason":"No prior New Runtime analytical record was supplied; this article supports a distinct record about lifecycle-spanning trust controls for autonomous agents."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/trust-in-depth-giving-agents-more-autonomy-without-3ff173/","markdown":"https://newruntime.com/posts/trust-in-depth-giving-agents-more-autonomy-without-3ff173.md","json":"https://newruntime.com/posts/trust-in-depth-giving-agents-more-autonomy-without-3ff173.json"}
related_patterns: ["agent-security-moves-to-runtime-boundaries"]
---

# Trust in Depth: Giving agents more autonomy without losing control

## Retrieval answer

Thesis: AWS’s Trust in Depth framework treats agent trust as a lifecycle control problem, not a one-time approval or a single safety filter.

Thesis: AWS’s Trust in Depth framework treats agent trust as a lifecycle control problem, not a one-time approval or a single safety filter. The article organizes that work around scoping, identifying, bounding, observing, and proving.

Mechanism: AWS identifies operational questions: define the agent’s purpose and prohibited actions before it runs; establish who authorized it and what identity it carries at deployment; check that its actions remain within limits while it runs; and reconstruct what it did afterward. It adds a distinction between rules that can be stated precisely and expectations that depend on context. For the former, automated reasoning can check whether a rule holds across paths covered by a precise system description and its assumptions, or expose a failing case. Mechanism: independent controls at different lifecycle stages are meant to support autonomy where actions can be verified, rather than making a person approve every action.

Why now: AWS frames blanket human approval as a scaling problem: approvals can accumulate and review fatigue can make people less able to catch consequential errors. The article argues that organizations need confidence to hand agents more important work, while keeping human judgment for decisions that require it. That makes the allocation of verification work central to increasing autonomy.

Implications: Teams can use the framework to ask which requirements are precise enough for rule-based checking, which need contextual review, and whether run records let them reconstruct agent behavior. Teams can separate requirements suitable for precise rule checks from those requiring contextual review. Runtime limits and post-run observability complement pre-deployment testing, but do not eliminate dependence on sound rules and assumptions. The article makes clear that formal assurance depends on the quality of the rules, system description, and assumptions.

Evidence boundary: This is AWS’s account of its approach and experience, including its claim of more than a decade using automated reasoning in production. The post does not provide comparative results for agent deployments, independent validation of the framework, or evidence that every relevant behavior can be formalized. A proof applies only to the rules and assumptions represented in the model.

Watch condition: whether AWS or adopting teams publish concrete agent deployments with explicit rules, assumptions, failure cases, and evidence that the layered controls change outcomes. That evidence would show how the framework works beyond its conceptual description.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agents](https://newruntime.com/topics/agents.md): Explore the agents topic hub.
- topic hub: [Agent Security](https://newruntime.com/topics/agent-security.md): Explore the agent-security topic hub.
- topic hub: [Security](https://newruntime.com/topics/security.md): Explore the security topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)