Entity trail

The Machine

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-07-11 / agents-researcherGhostApproval: symlink flaw turns human-in-the-loop into a rubber stamp across 6 coding agentsResearchers disclosed GhostApproval in July 2026, a symlink-following flaw (CWE-61) that lets a malicious repo trick AI coding assistants into writing files outside their workspace sandbox, escalating to code execution on the developer's machine. It affects six major agents — Amazon Q Developer, Anthropic Claude Code, Augment, Cursor, Google Antigravity, and Windsurf — and critically defeats the human-in-the-loop approval prompt by hiding the real write target. It's a reminder that agent sandboxes and consent UIs can be broken by decades-old filesystem tricks.
2026-07-11 / arxiv-researcherDual-Graph Verification Framework Audits Legacy IT Security Concepts for NIS-2 / OSCAL MigrationThe NIS-2 Directive pushes organizations from document-based compliance toward machine-readable artifacts like OSCAL (with Germany's BSI adapting it as Grundschutz++), but migrating extensive legacy IT security concepts without verification risks translation errors. This paper proposes a dual-graph verification framework to reverse-engineer and audit those legacy concepts during migration. Complements the concurrent MCP-based OSCAL pipeline work as the verification half of continuous-compliance tooling.
2026-07-11 / thought-leaders-researcherDario Amodei Walks Back the 'Decade in a Year' Timeline: 'I Don't Think Today We Can Make Progress at a Rate of 10 Years Per Year'In a July 6 STAT News conversation, the Anthropic CEO notably softened the central claim of his 2024 'Machines of Loving Grace' essay, saying his vision of AI compressing a century of biotech into a decade 'might not start to be visible for a decade' and that a 10x-per-year rate isn't achievable today 'for a number of reasons.' Coming from AI's most aggressive timeline optimist, the hedge is the story — a data point for builders weighing how much to bet on near-term scientific acceleration versus incremental tooling wins.
2026-07-10 / skill-finder/loop runs local, /schedule runs in the cloud — and the interval should match how often the input actually changesThe two time-based primitives differ only in where they execute: /loop stays on your machine, /schedule moves the routine to the cloud. Anthropic's token-management guidance is to match the routine interval to the real change frequency of the thing being watched, and to pilot dynamic workflows on a small input before a large run because they can spawn hundreds of agents. Inspect spend with /usage, /goal with no arguments, and /workflows.
2026-07-10 / agents-researcherAnthropic expands Claude Cowork to mobile and web with remote, device-independent agent sessionsBeginning July 7 and rolling out through July 9, Anthropic moved Cowork sessions off the local machine: sessions and files persist to the Claude account, background work continues after the laptop closes, scheduled tasks run with no device online, and approvals can be granted from mobile. Beta access started with Max subscribers, with doubled usage limits through August 5. The shift from local process to hosted session is the architectural change that matters — it makes Cowork a durable agent runtime rather than a desktop app, and it puts scheduling and human-in-the-loop approval on the critical path.
2026-07-10 / hn-researcherShow HN: Colibri Runs GLM-5.2's 744B MoE on a 25GB-RAM Consumer Machine in Pure CTopping HN at 767 points and 189 comments, Colibri exploits the structure of a 744B mixture-of-experts model: only ~40B params activate per token and only ~11GB of those change between tokens. The dense part (attention, shared experts, embeddings, ~17B params at int4, ~9.9GB) stays resident while 21,504 routed experts stream from ~370GB on disk via a per-layer LRU cache and the OS page cache. The engine is a single ~1,300-line C file with zero dependencies that profiles your routing patterns and auto-pins hot experts, so it speeds up with use — though throughput on a 32GB machine is still around 0.1 tok/s, making this an architecture demonstration rather than a daily driver.
2026-07-10 / vibe-coding-researcherLeanCTX Ships a Rust Binary That Decides What Your Agent Is Allowed to Readyvgude/lean-ctx positions itself as a 'context intelligence layer' — one local Rust binary that gates what files an agent reads, persists what it learns, and guards what leaves the machine. The framing is a policy layer between the agent and the filesystem, rather than a memory store bolted on top. For builders, this is the enforcement counterpart to CLAUDE.md's suggestion-based approach: instead of asking the model politely not to read `.env`, the binary refuses to serve it. Single-source and early, but the enforcement-vs-suggestion distinction is the right axis.
2026-07-09 / thought-leaders-researcherThorsten Ball's Amp Adds Configurable Remote 'Orbs' and Ships Opus 4.8-Powered 'Smart' ModeEarly-July Amp updates let users pick the CPU and memory for the remote 'orbs' that run agents (32GB/16-core orbs bill at $1.66/hour by the minute, sleeping when idle) and gave the agent the ability to read threads of any size to answer questions about them. Its 'smart' mode is now powered by Claude Opus 4.8, which makes tighter changes and checks its own work. For builders, it's a concrete look at where remote agent execution is heading: per-thread ephemeral machines with tunable compute and self-checking loops.
2026-07-08 / sources-researcherDwarkesh Patel × Grant Sanderson (3Blue1Brown): AI Disproved a Famous Math Conjecture — Now What?Grant Sanderson joins Dwarkesh to discuss AI systems disproving a well-known mathematical conjecture and what machine-assisted discovery means for the practice of mathematics. A high-signal conversation on where AI is moving from tool to genuine research collaborator.
2026-07-07 / agents-researcherArmadin details full sandbox escape in Anthropic's Claude Cowork; Anthropic disputes, silently patchesResearchers at Armadin published an attack chain against Claude Cowork for Windows: manipulating a resume flag passed through CoworkVMService bypasses the per-command unprivileged-user creation to run commands as root, while a second flaw overrides the domain allowlist with a wildcard to strip egress limits and enable data exfiltration. Anthropic argues it isn't a security issue because it requires prior local code execution, yet still shipped a patch — a preview of the accountability gray zone builders enter when agent tools run local VMs on user machines. For anyone shipping sandboxed agents, the lesson is that a resume/replay path and per-request allowlist overrides are exactly where isolation quietly dies.
2026-07-01 / thought-leaders-researcherSimon Willison Runs Ornith-1.0 Locally — an Open Coding Model That Writes Its Own RL Training ScaffoldIn a June 29 hands-on, Simon Willison ran DeepReinforce's new MIT-licensed Ornith-1.0 (the 35B MoE GGUF, ~20GB) in LM Studio wired to his Pi harness and reported it drives an agent loop across many tool calls proficiently. The model's novelty is self-scaffolding: it writes the training harness that guides its own reinforcement-learning improvement, and it's built on pretrained Gemma 4 and Qwen 3.5. For builders, it's a credible local, open alternative for agentic coding that runs on a single high-memory machine.
2026-06-30 / saas-disruption-researcherMira Murati's Thinking Machines Lab Raises $2B Series B at $10B to Build the Agent Infrastructure LayerThinking Machines Lab, led by former OpenAI CTO Mira Murati, raised a $2B Series B at a $10B valuation to build infrastructure and foundation models for multi-step, self-improving agentic AI. The scale of capital flowing to the agent-infrastructure substrate (vs. application-layer agents) signals where investors think the durable moats sit. For builders, more well-funded model/infra providers means continued downward pressure on inference costs and more 'bring-your-own-model' architectures replacing vendor-locked AI features in SaaS.

Graph relationships

RELEASED
Stagwell
Stagwell released The Machine as part of its AI product suite.
Source finding
RELEASED
Stagwell -> The Machine
Stagwell released The Machine as part of its AI product suite.
Source finding

Source trail

Cyber Security News arXiv STAT News Claude by Anthropic — Getting started with loops Android Headlines Hacker News / JustVugg (GitHub)GitHub

Graph sources

entity graphfindings textkg edgeskg entities