> ## Documentation Index
> Fetch the complete documentation index at: https://anthropic-training.neuroscale.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Research and source reading

> Papers, essays, and product research behind the session claims.

These are the sources behind the mental models we used in class. Open them when you need receipts, not vibes.

## Anthropic research and product writing

<a className="res-doc" href="https://www.anthropic.com/research/trustworthy-agents" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <path d="M6 4.5h12v15H6z" />

      <path d="M9 8h6M9 11.5h6M9 15h4" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>Trustworthy agents in practice</strong>
    <span>How Anthropic thinks about agents that need oversight, evidence, and recoverable actions.</span>
    <em>Anthropic · Research</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

<a className="res-doc" href="https://www.anthropic.com/news/our-framework-for-developing-safe-and-trustworthy-agents" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <path d="M4 7h16v10H4z" />

      <path d="M8 11h8M8 14h5" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>Framework for safe and trustworthy agents</strong>
    <span>The product-and-policy framing behind the decision airlock we practiced.</span>
    <em>Anthropic · Announcement</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

<a className="res-doc" href="https://www.anthropic.com/research" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <circle cx="11" cy="11" r="6.5" />

      <path d="M16 16l4.5 4.5" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>Anthropic Research hub</strong>
    <span>Index of alignment, interpretability, and agent papers. Browse when a class claim needs deeper backup.</span>
    <em>Anthropic · Hub</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

<a className="res-doc" href="https://www.anthropic.com/engineering/building-effective-agents" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <path d="M5 7h14v3H5zM5 14h9v3H5z" />

      <path d="M17 14v6" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>Building effective agents</strong>
    <span>Engineering essay on when workflows beat free-form agents, and how to keep loops inspectable.</span>
    <em>Anthropic · Engineering</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

<a className="res-doc" href="https://claude.com/blog/the-new-rules-of-context-engineering-for-claude-5-generation-models" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <path d="M4 6.5h16v11H4z" />

      <path d="M8 10h8M8 13.5h5" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>Context engineering for Claude 5 models</strong>
    <span>Why less standing instruction often beats more, once models get sharper.</span>
    <em>Anthropic · Blog</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

## Protocols and open standards

<a className="res-doc" href="https://modelcontextprotocol.io/specification/2026-07-28/basic/security_best_practices" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <path d="M12 3.5 20 7v5c0 5-3.4 8.4-8 9.5C7.4 20.4 4 17 4 12V7l8-3.5z" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>MCP security best practices</strong>
    <span>Official guidance on tool trust, prompt injection surfaces, and connector hygiene.</span>
    <em>MCP · Spec</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

<a className="res-doc" href="https://www.anthropic.com/news/model-context-protocol" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <path d="M7 8h10v8H7z" />

      <path d="M10 4v4M14 4v4M10 16v4M14 16v4" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>Introducing the Model Context Protocol</strong>
    <span>Original announcement framing MCP as a USB-like standard for AI tools.</span>
    <em>Anthropic · News</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

## Independent analysis worth keeping

<a className="res-doc" href="https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <path d="M12 4 3.5 19h17L12 4z" />

      <path d="M12 10v4.5M12 17h.01" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>The lethal trifecta for AI agents</strong>
    <span>Private data, untrusted content, and external communication in one place. Best short safety model on the internet.</span>
    <em>Simon Willison · Essay</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

<a className="res-doc" href="https://aifluencyframework.org/" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <circle cx="12" cy="12" r="8.5" />

      <path d="M8 12h8M12 8v8" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>AI Fluency Framework site</strong>
    <span>Primary source for the 4D fluency language used in Foundations.</span>
    <em>Dakan & Feller · Site</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

<a className="res-doc" href="https://www.anthropic.com/news/usage-policy-update" target="_blank" rel="noopener noreferrer">
  <span className="res-doc__icon" aria-hidden="true">
    <svg viewBox="0 0 24 24" width="18" height="18" fill="none" stroke="currentColor" strokeWidth="1.8">
      <path d="M7 4.5h10v15H7z" />

      <path d="M10 9h4M10 12.5h4M10 16h2.5" />
    </svg>
  </span>

  <span className="res-doc__body">
    <strong>Usage Policy update</strong>
    <span>Policy source for what should not be routed into Claude casually.</span>
    <em>Anthropic · Policy</em>
  </span>

  <span className="res-doc__open">Open</span>
</a>

## How to cite these in your org

When someone asks “says who?”, send the specific link, not the whole Resources section. One paper beats a vibes argument. If your company has an approved AI policy page, put that above every external link in this list.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.