--- title: The future of governing AI agents | daily.dev description: Autonomous AI security agents are outpacing governance frameworks. A four-layer architecture separating skills, reasoning, models, and context is proposed to... canonical: https://daily.dev/posts/the-future-of-governing-ai-agents-rhnih4nos twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: The future of governing AI agents | daily.dev og:description: Autonomous AI security agents are outpacing governance frameworks. A four-layer architecture separating skills, reasoning, models, and context is proposed to... og:url: https://daily.dev/posts/the-future-of-governing-ai-agents-rhnih4nos og:image: https://api.daily.dev/og/posts/rhniH4nOs.png og:image:alt: The future of governing AI agents og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # The future of governing AI agents **[elastic](https://daily.dev/sources/elastic)** · 16 min read · 1 upvotes · 0 comments ## Summary Autonomous AI security agents are outpacing governance frameworks. A four-layer architecture separating skills, reasoning, models, and context is proposed to make agent behavior predictable, auditable, and compliant with regulations like ISO 42001, DORA, NIS2, and the EU AI Act. Extracting reasoning into an explicit, versioned layer decouples trust evidence from specific LLMs, enabling progressive trust models where oversight tightens or relaxes based on measured performance. Continuous evaluation across classification accuracy, planning quality, retrieval quality, and grounding quality provides the evidence base. Human-in-the-loop oversight is shown to degrade under volume, requiring metrics like approval latency, override rates, and evidence engagement to detect rubber-stamping. Uber's ADR system is cited as a real-world example of enterprise-scale agentic telemetry. Elastic Security is positioned as the platform to ingest, store, and query agent reasoning chains at scale. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What percentage of Claude Code permission prompts do users approve, and does that suggest human oversight is working? Users approved 93% of permission prompts in Claude Code, with experienced users auto-approving at roughly twice the rate of new users. Anthropic concluded this pattern showed the oversight mechanism was degrading rather than protecting, since high approval volume led to reduced scrutiny per approval, and the company redesigned its approval model as a result. _Teams designing human-in-the-loop controls for AI agents can follow this kind of oversight research on daily.dev._ ### How does Uber's Agentic Detection and Response (ADR) system monitor AI agent behavior for security purposes? Uber's ADR system, deployed across more than 7000 hosts, captures traditional security signals like file writes and API calls alongside agent-specific telemetry including reasoning chains, prompts, and chains of intent leading to execution. A Sensor component forwards reconstructed agent sessions to a backend for two-tier detection analysis, using Elasticsearch as the underlying telemetry store. _Engineers building agent observability tooling can track real-world deployments like this one on daily.dev._ ### Why should reasoning logic be separated from the underlying LLM in an autonomous security agent architecture? Separating reasoning into its own explicit, versioned layer means trust evidence attaches to the methodology rather than to a specific model, so swapping LLMs (due to pricing, deprecation, or regulatory concerns) doesn't reset accumulated trust or require re-evaluation from scratch. It also lets a smaller, cheaper model execute well-defined methodology and outperform a frontier model guessing at methodology. _Architects weighing model portability against vendor lock-in can follow this governance approach on daily.dev._ ## Similar posts on daily.dev - [The missing layer in enterprise agentic AI](https://daily.dev/posts/the-missing-layer-in-enterprise-agentic-ai-3eduo8jqv) · InfoWorld · 0 upvotes · 0 comments - [Building trusted agentic AI in financial services: From data to autonomous action](https://daily.dev/posts/building-trusted-agentic-ai-in-financial-services-from-data-to-autonomous-action-pc9tqxd43) · elastic · 1 upvotes · 0 comments - [What It Takes To Scale AI Agents in Production](https://daily.dev/posts/what-it-takes-to-scale-ai-agents-in-production-rombpft2h) · The New Stack · 1 upvotes · 0 comments - [Medium](https://daily.dev/posts/medium-t4qlwxuej) · Medium · 0 upvotes · 0 comments --- Tags: [#security](https://daily.dev/tags/security), [#ai-agents](https://daily.dev/tags/ai-agents), [#observability](https://daily.dev/tags/observability), [#elk](https://daily.dev/tags/elk), [#ai-governance](https://daily.dev/tags/ai-governance) [View this post on daily.dev](https://daily.dev/posts/the-future-of-governing-ai-agents-rhnih4nos) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"The future of governing AI agents","url":"https://daily.dev/posts/the-future-of-governing-ai-agents-rhnih4nos","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/the-future-of-governing-ai-agents-rhnih4nos"},"datePublished":"2026-07-08T16:46:22.225Z","dateModified":"2026-09-13T20:52:15.330Z","description":"Autonomous AI security agents are outpacing governance frameworks. A four-layer architecture separating skills, reasoning, models, and context is proposed to...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/46f6095619b8e966671af6531d88811c?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/46f6095619b8e966671af6531d88811c?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"elastic","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"elastic","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/b9f4b44c864c42f68aa7f164715da8b4","url":"https://daily.dev/sources/elastic"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/the-future-of-governing-ai-agents-rhnih4nos","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"security,ai-agents,observability,elk,ai-governance","timeRequired":"PT16M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"elastic","item":"https://daily.dev/sources/elastic"},{"@type":"ListItem","position":3,"name":"The future of governing AI agents"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/the-future-of-governing-ai-agents-rhnih4nos#faq","mainEntity":[{"@type":"Question","name":"What percentage of Claude Code permission prompts do users approve, and does that suggest human oversight is working?","acceptedAnswer":{"@type":"Answer","text":"Users approved 93% of permission prompts in Claude Code, with experienced users auto-approving at roughly twice the rate of new users. Anthropic concluded this pattern showed the oversight mechanism was degrading rather than protecting, since high approval volume led to reduced scrutiny per approval, and the company redesigned its approval model as a result. Teams designing human-in-the-loop controls for AI agents can follow this kind of oversight research on daily.dev."}},{"@type":"Question","name":"How does Uber's Agentic Detection and Response (ADR) system monitor AI agent behavior for security purposes?","acceptedAnswer":{"@type":"Answer","text":"Uber's ADR system, deployed across more than 7000 hosts, captures traditional security signals like file writes and API calls alongside agent-specific telemetry including reasoning chains, prompts, and chains of intent leading to execution. A Sensor component forwards reconstructed agent sessions to a backend for two-tier detection analysis, using Elasticsearch as the underlying telemetry store. Engineers building agent observability tooling can track real-world deployments like this one on daily.dev."}},{"@type":"Question","name":"Why should reasoning logic be separated from the underlying LLM in an autonomous security agent architecture?","acceptedAnswer":{"@type":"Answer","text":"Separating reasoning into its own explicit, versioned layer means trust evidence attaches to the methodology rather than to a specific model, so swapping LLMs (due to pricing, deprecation, or regulatory concerns) doesn't reset accumulated trust or require re-evaluation from scratch. It also lets a smaller, cheaper model execute well-defined methodology and outperform a frontier model guessing at methodology. Architects weighing model portability against vendor lock-in can follow this governance approach on daily.dev."}}]} ```