<?xml version="1.0" encoding="UTF-8"?>
<rss version='2.0' xmlns:dc="http://purl.org/dc/elements/1.1/"
  xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>MikeBerggren.com</title>
    <description>Musings from a cybersecurity geek, engineering nerd, productivity enthusiast, devops wannabe, and sci-fi addict.</description>
    <link>https://mikeberggren.com/feed</link>
    <atom:link href="https://mikeberggren.com/feed" rel="self" type="application/rss+xml"/>
    <category domain="mikeberggren.com">Content Management/Blog</category>
    <language>en-us</language>
      <pubDate>Tue, 10 Nov 2026 14:41:22 -0800</pubDate>
    <managingEditor>mike.berggren@outlook.com (MikeBerggren.com)</managingEditor>
      <item>
        <guid>http://mikeberggren.com/black-bean-tacos#61862</guid>
          <pubDate>Tue, 10 Nov 2026 14:41:22 -0800</pubDate>
        <link>http://mikeberggren.com/black-bean-tacos</link>
        <title>Recipe: Black Bean Tacos with Tomato &amp; Avocado Salsa</title>
        <description></description>
        <content:encoded><![CDATA[<p>One of my favorite meal recipe apps (Mealime) recently shut down. Major bummer.</p>

<p>In the interest of preserving some really good recipes, I thought I&#39;d post one here to share with the world. Enjoy!</p>

<hr>

<h2 id="ingredients">Ingredients</h2>

<p><em>Makes 6 servings</em></p>

<table><thead>
<tr>
<th>Ingredient</th>
<th>Amount</th>
</tr>
</thead><tbody>
<tr>
<td>Avocados</td>
<td>3</td>
</tr>
<tr>
<td>Black beans</td>
<td>3 (15 oz) cans</td>
</tr>
<tr>
<td>Cilantro</td>
<td>1½ small bunches</td>
</tr>
<tr>
<td>Grape tomatoes</td>
<td>1½ pints</td>
</tr>
<tr>
<td>Limes</td>
<td>3</td>
</tr>
<tr>
<td>Red onions</td>
<td>1½ medium</td>
</tr>
<tr>
<td>Small corn tortillas</td>
<td>24</td>
</tr>
<tr>
<td>Black pepper</td>
<td>½ tbsp</td>
</tr>
<tr>
<td>Chili powder</td>
<td>½ tbsp</td>
</tr>
<tr>
<td>Ground cumin</td>
<td>½ tbsp</td>
</tr>
<tr>
<td>Salt</td>
<td>2 tsp</td>
</tr>
<tr>
<td>Virgin coconut oil</td>
<td>3 tbsp</td>
</tr>
</tbody></table>

<h2 id="nutrition-facts-per-serving">Nutrition Facts (per serving)</h2>

<p><strong>650 calories</strong> · Fat 26g · Net carbs 65g · Protein 20g · Sugars 6g</p>

<p><em>Macro split: 34% fat, 55% carbs, 11% protein</em></p>

<h2 id="cookware">Cookware</h2>

<ul>
<li>Cutting board</li>
<li>Can opener</li>
<li>Measuring spoons</li>
<li>Mixing bowls</li>
<li>Colander</li>
<li>Stirring spoon</li>
<li>Stainless steel or cast iron skillet</li>
<li>Citrus juicer</li>
<li>Chef&#39;s knife</li>
</ul>

<hr>

<h2 id="instructions">Instructions</h2>

<h3 id="make-the-salsa">Make the salsa</h3>

<ol>
<li><strong>Wash and dry the fresh produce:</strong> 1½ pints grape tomatoes, 1½ small bunches cilantro, 3 avocados, and 3 limes.</li>
<li>Small dice the tomatoes and transfer to a medium bowl.</li>
<li>Halve the avocados lengthwise and twist the halves to separate, then remove the pit with a knife or spoon. Scoop out and small dice the flesh, then add it to the tomatoes.</li>
<li>Using a knife, shave the cilantro leaves off the stems at a downward angle, working away from your body. Discard the stems and finely chop the leaves. Add half of the cilantro to the bowl with the tomatoes and avocado, and transfer the other half to a small bowl.</li>
</ol>

<h3 id="cook-the-beans">Cook the beans</h3>

<ol>
<li>Preheat a skillet over medium heat.</li>
<li>While the skillet heats up, trim off and discard the ends of the <strong>1½ red onions</strong> and remove the outer layer. Small dice the onions (¼-inch pieces). Transfer half of the onions to the salsa bowl and leave the rest on the cutting board.</li>
<li>Once the skillet is heated, add <strong>1½ tbsp coconut oil</strong> and swirl to coat the bottom.</li>
<li>Add the remaining onion to the skillet and cook, stirring occasionally, until softened, 3 to 5 minutes.</li>
<li>While the onion cooks, drain and rinse the <strong>3 cans of black beans</strong> in a colander.</li>
<li>Add another <strong>1½ tbsp coconut oil</strong> to the skillet, followed by the beans and the remaining cilantro.</li>
<li>Season the beans with <strong>1½ tsp chili powder, 1½ tsp cumin, 1 tsp salt, and ¾ tsp black pepper</strong>. Stir until fragrant, 15 to 30 seconds. Cook, stirring occasionally, until the beans are warmed through, 2 to 4 minutes. Mash some of the beans with a fork if desired.</li>
</ol>

<h3 id="finish-and-serve">Finish and serve</h3>

<ol>
<li>Wrap the <strong>24 tortillas</strong> in a damp towel and microwave until heated through, 15 to 30 seconds.</li>
<li>Juice the limes. Drizzle half of the juice over the beans and add the rest to the salsa bowl.</li>
<li>Season the salsa with <strong>½ tsp salt and ¾ tsp black pepper</strong>, then mix together.</li>
<li>To serve, spoon the bean mixture into the tortillas and top with the salsa. Enjoy!</li>
</ol>
]]></content:encoded>
      </item>
      <item>
        <guid>http://mikeberggren.com/sandboxing-isnt-enough#61855</guid>
          <pubDate>Wed, 04 Nov 2026 17:23:24 -0800</pubDate>
        <link>http://mikeberggren.com/sandboxing-isnt-enough</link>
        <title>Why Sandboxing Isn’t Enough to Contain AI Agents</title>
        <description></description>
        <content:encoded><![CDATA[<p>Wow. </p>

<p>I&#39;m absolutely loving Matthew Green&#39;s latest blog post on whether sandboxing is enough to contain AI agents. He doesn&#39;t pull any punches and calls it like it is. </p>

<p>If you haven&#39;t read it, check it out <a href="https://blog.cryptographyengineering.com/2026/09/30/is-sandboxing-sufficient-to-contain-rogue-agents/">here</a></p>

<p>My takeaways: </p>

<ul>
<li><p>The OpenAI incident demonstrated that agents can proactively discover zero-day exploits and collaborate to bypass restrictions, a failure compounded by an organizational structure where no security team had clear authority to overrule researchers. </p></li>
<li><p>Sandboxing is a fundamentally incomplete solution because utility requires information access. Granting an agent the power to be useful opens two-way communication channels that are impossible to monitor perfectly for obfuscated malicious data or unauthorized instructions. </p></li>
<li><p>The &quot;warden&quot; approach (using a cheaper model to guard a more intelligent one) merely repackages the alignment problem, as it relies on the hope that a lesser intelligence can reliably contain a &quot;wizard&quot; in an adversarial setting. </p></li>
<li><p>The scariest risk may not be a rogue superintelligence at all. It&#39;s perfectly compliant agents doing exactly what they&#39;re told by whoever manages to get text in front of them. Agents that pass instructions to each other through shared channels like email, Slack, and shared docs are all the ingredients a worm needs. </p></li>
</ul>

<p>Admittedly, it&#39;s not an easy challenge to solve, but it&#39;s one that absolutely calls for a defense-in-depth mindset.</p>
]]></content:encoded>
      </item>
      <item>
        <guid>http://mikeberggren.com/ai-agents-patching-vulnerabilities#61844</guid>
          <pubDate>Thu, 22 Oct 2026 14:29:09 -0700</pubDate>
        <link>http://mikeberggren.com/ai-agents-patching-vulnerabilities</link>
        <title>Should we trust AI agents to patch our vulnerabilities?</title>
        <description></description>
        <content:encoded><![CDATA[<p>I&#39;m a little late to the party on this one, but I still wanted to share and comment on Google&#39;s recent findings on vulnerability/exploitation trends. The article is a bit lengthy (~17 min read) but it&#39;s a good read. Check it out <a href="https://cloud.google.com/blog/topics/threat-intelligence/vulnerability-discovery-and-exploitation-trends-in-the-ai-era">here</a>. </p>

<p>My takeaways (most of which aren&#39;t particularly surprising):</p>

<p>-- Vulnerabilities targeting the AI operational stack are heavily concentrated in orchestration and agent frameworks (e.g., Langflow, LangChain, AutoGen). This layer accounts for 50% of all AI-related flaws disclosed in 2026.</p>

<p>-- The one that did catch my eye: half of all AI-discovered vulnerabilities lead to remote code execution, compared to roughly a quarter across the broader CVE ecosystem. AI isn&#39;t just finding more bugs, it&#39;s finding more serious ones.</p>

<p>-- With discovery and exploitation both accelerating, the article argues for a fundamental shift in defensive strategy, which I&#39;d boil down to:</p>

<p>1) Intelligence-driven triage<br>
2) AI-enhanced code review<br>
3) Automated remediation</p>

<p>While I respect the strategy here, I think these are delicate matters (particularly the third) and shouldn&#39;t be implemented haphazardly. Ironically, the same report finds agent frameworks are the most vulnerable layer of the AI stack, so giving an agent write access to patch production deserves the same scrutiny as any other privileged tool.</p>
]]></content:encoded>
      </item>
  </channel>
</rss>