--- title: Netflix reveals how it babysits an AI judge, Omarchy... description: Netflix published a detailed lifecycle for keeping an LLM-as-a-judge from drifting in production, a rare look at how evaluation actually holds up at scale.... canonical: https://daily.dev/posts/netflix-reveals-how-it-babysits-an-ai-judge-omarchy-patches-a-root-escalation-bug-doe8aprv4 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Netflix reveals how it babysits an AI judge, Omarchy patches a root escalation bug | daily.dev og:description: Netflix published a detailed lifecycle for keeping an LLM-as-a-judge from drifting in production, a rare look at how evaluation actually holds up at scale.... og:url: https://daily.dev/posts/netflix-reveals-how-it-babysits-an-ai-judge-omarchy-patches-a-root-escalation-bug-doe8aprv4 og:image: https://api.daily.dev/og/posts/Doe8APrv4.png og:image:alt: Netflix reveals how it babysits an AI judge, Omarchy patches a root escalation bug 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. # Netflix reveals how it babysits an AI judge, Omarchy patches a root escalation bug **[Agentic Digest](https://daily.dev/sources/agents_digest)** · 4 min read · 1 upvotes · 0 comments ## Summary Netflix published a detailed lifecycle for keeping an LLM-as-a-judge from drifting in production, a rare look at how evaluation actually holds up at scale. Anthropic is warning that infostealer malware is hijacking authenticated Claude sessions to burn through usage without needing passwords. DHH's Omarchy Linux distro shipped a fix for a bug that silently handed root access to any process since June 2025. Meanwhile new agent-harness research keeps chipping away at long-horizon memory and context problems that plague anything running unattended for hours. ## Content **TLDR:** Netflix published a detailed lifecycle for keeping an LLM-as-a-judge from drifting in production, a rare look at how evaluation actually holds up at scale. Anthropic is warning that infostealer malware is hijacking authenticated Claude sessions to burn through usage without needing passwords. DHH's Omarchy Linux distro shipped a fix for a bug that silently handed root access to any process since June 2025. Meanwhile new agent-harness research keeps chipping away at long-horizon memory and context problems that plague anything running unattended for hours. --- ## Netflix documents the full lifecycle of running an LLM judge in production Netflix uses one model to write recommendation blurbs and a second model to judge every one before it ships. The new paper breaks the judge's maintenance into four stages, including weekly drift monitoring calibrated against how much human raters disagree with each other, not some fixed accuracy target. In a five-week test, judge-aligned explanations nudged users toward previously unwatched titles and increased successful browse-to-play sessions, a small but real effect from a system most teams would validate once and forget. [Read more](https://daily.dev/feed-by-ids?id=J7XUSD54H&id=Vg9Wh81EF) ## Anthropic warns infostealer malware is hijacking Claude sessions Malware including Vidar, LummaC2, StealC, and Atomic Stealer is lifting authenticated browser sessions from infected machines, letting attackers burn through Claude usage without ever needing a password or 2FA. Anthropic says it's unrelated to Claude itself and is responding by signing out affected accounts, pulling saved payment methods, and refunding unauthorized charges. Signing out alone doesn't fix an infected machine, so anyone hit needs to actually remove the malware and rotate credentials. [Read more](https://daily.dev/posts/Z5uCaK5u9) ## Omarchy patches a root-escalation bug that sat unfixed for over a year Every version of DHH's Omarchy Linux distro before 4.0.1 silently added users to the Docker group, meaning any process in your desktop session, including a stray npm postinstall script, had a straight path to root with no password prompt. The bug was introduced in June 2025 and fixed in August 2026 after private disclosure, while the docs kept implying Docker ran rootless. The researcher's fix recommendation is blunt: drop Docker for Podman, which doesn't need root at all. [Read more](https://daily.dev/posts/T9972AdJG) ## Agent harnesses keep beating raw model capability on long-horizon tasks Prime Agent's persistent IPython REPL harness pushed ARC-AGI-3 RHAE Best@1 from 30% to 95.5% just by fixing execution, recovery, and verification around the model rather than touching the model itself. Separately, a new paper names the 'Compaction Cliff': Claude Code's /compact prompt on Sonnet 4.6 retains only 53% of safety rules after one compaction round and drops to 10% after five, because context compression treats rigid rules the same as disposable logs. Their proposed fix, Knowledge Triage, beats single-shot compactors by 2-4x on rule preservation. [Read more](https://daily.dev/feed-by-ids?id=Ed2bthXgI&id=ULJdgXPnH) --- ## Also notable - **ChatGPT Work hits 25M users but under 1% individual adoption:** OpenAI's Thibault Sottiaux touts 25 million users, but internal Codex adoption is 98% among employees versus under 1% of individual outside subscribers and 17% of organizations. [Read more](https://daily.dev/posts/0VUihq62D) - **Atlassian's agent pipeline resolves 120+ vulnerabilities with 95% first-run merge rate:** Over three months, Atlassian's dispatcher-plus-Rovo-Dev pipeline merged 55+ agent-created pull requests fixing security vulnerabilities, with 95% merging clean on the first try. [Read more](https://daily.dev/posts/XXtgsat4y) - **Cheaper tokens are driving 13.8x more usage, not less revenue:** OpenRouter's heavy discount on GPT-5.6 Terra and Luna produced a 13.8x jump in token usage, a live demonstration of Jevons paradox in enterprise AI spend. [Read more](https://daily.dev/posts/ZPEcLRuAF) - **Uber holds AI coding costs flat despite 9.4x growth in weekly agent requests:** Uber's cost-control playbook, including CLI-resolved MCP tool calls and code-mode batching that cuts token usage by over 90% for bulk workflows, kept spend stable even as weekly active agentic users grew 7x since February. [Read more](https://daily.dev/posts/fyfellEFf) - **An AI system designed a working chip from spec to silicon in two weeks:** Redwood's AI-generated chip design, tests, and firmware hit 95% test coverage with zero bugs on the first FPGA run, and a spec change reached real hardware in under 48 hours. [Read more](https://daily.dev/posts/gwcpUU3oP) ## Similar posts on daily.dev - [Claude Mythos Preview \\ red.anthropic.com](https://daily.dev/posts/claude-mythos-preview-red-anthropic-com-wkjczmbgu) · Hacker News · 5 upvotes · 0 comments --- Tags: [#linux](https://daily.dev/tags/linux), [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#claude](https://daily.dev/tags/claude) [View this post on daily.dev](https://daily.dev/posts/netflix-reveals-how-it-babysits-an-ai-judge-omarchy-patches-a-root-escalation-bug-doe8aprv4) ```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":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/netflix-reveals-how-it-babysits-an-ai-judge-omarchy-patches-a-root-escalation-bug-doe8aprv4","headline":"Netflix reveals how it babysits an AI judge, Omarchy patches a root escalation bug","text":"Netflix published a detailed lifecycle for keeping an LLM-as-a-judge from drifting in production, a rare look at how evaluation actually holds up at scale. Anthropic is warning that infostealer malware is hijacking authenticated Claude sessions to burn through usage without needing passwords. DHH's Omarchy Linux distro shipped a fix for a bug that silently handed root access to any process since June 2025. Meanwhile new agent-harness research keeps chipping away at long-horizon memory and context problems that plague anything running unattended for hours.","url":"https://daily.dev/posts/netflix-reveals-how-it-babysits-an-ai-judge-omarchy-patches-a-root-escalation-bug-doe8aprv4","datePublished":"2026-08-31T04:17:41.097Z","dateModified":"2026-08-31T04:18:03.487Z","author":{"@type":"Organization","name":"Agentic Digest","logo":"https://media.daily.dev/image/upload/s--V91DY4ls--/f_auto,q_auto/v1772617267/logos/agents_digest","url":"https://daily.dev/sources/agents_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/agents_digest","name":"Agentic Digest"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Agentic Digest","item":"https://daily.dev/sources/agents_digest"},{"@type":"ListItem","position":3,"name":"Netflix reveals how it babysits an AI judge, Omarchy patches a root escalation bug"}]} ```