<?xml version="1.0" encoding="UTF-8"?><rss 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" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Alt. - Medium]]></title>
        <description><![CDATA[Alt is a Medium publication by CTRL Magazine. A place for all things tech. - Medium]]></description>
        <link>https://alt.ctrlmag.com?source=rss----a0bc969b4f51---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Alt. - Medium</title>
            <link>https://alt.ctrlmag.com?source=rss----a0bc969b4f51---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 04:12:00 GMT</lastBuildDate>
        <atom:link href="https://alt.ctrlmag.com/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com" rel="hub"/>
        <item>
            <title><![CDATA[21 Claude Skills I’d Learn If I Were Starting Data Science Today]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://alt.ctrlmag.com/21-claude-skills-id-learn-if-i-were-starting-data-science-today-4aec7c3aa4dd?source=rss----a0bc969b4f51---4"><img src="https://cdn-images-1.medium.com/max/2600/0*ZGpnHtHZVpyJiTQT" width="2947"></a></p><p class="medium-feed-snippet">If I were starting data science today, I wouldn&#x2019;t spend all my time memorizing Python syntax, watching machine learning tutorials, or&#x2026;</p><p class="medium-feed-link"><a href="https://alt.ctrlmag.com/21-claude-skills-id-learn-if-i-were-starting-data-science-today-4aec7c3aa4dd?source=rss----a0bc969b4f51---4">Continue reading on Alt. »</a></p></div>]]></description>
            <link>https://alt.ctrlmag.com/21-claude-skills-id-learn-if-i-were-starting-data-science-today-4aec7c3aa4dd?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/4aec7c3aa4dd</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[claude]]></category>
            <dc:creator><![CDATA[learn with her]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 17:17:33 GMT</pubDate>
            <atom:updated>2026-11-15T17:17:32.099Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Business Logic Attacks: How Attackers Exploit OTP Recovery, Account Takeover, Payment Workflows and…]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://alt.ctrlmag.com/business-logic-attacks-how-attackers-exploit-otp-recovery-account-takeover-payment-workflows-and-03ea60ffa3ff?source=rss----a0bc969b4f51---4"><img src="https://cdn-images-1.medium.com/max/1672/1*A1cEl_-GTaQeM3k3gFptSA.png" width="1672"></a></p><p class="medium-feed-snippet">A real-world production security engineering deep dive using Java Spring Boot, Spring Security, PostgreSQL and distributed systems.</p><p class="medium-feed-link"><a href="https://alt.ctrlmag.com/business-logic-attacks-how-attackers-exploit-otp-recovery-account-takeover-payment-workflows-and-03ea60ffa3ff?source=rss----a0bc969b4f51---4">Continue reading on Alt. »</a></p></div>]]></description>
            <link>https://alt.ctrlmag.com/business-logic-attacks-how-attackers-exploit-otp-recovery-account-takeover-payment-workflows-and-03ea60ffa3ff?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/03ea60ffa3ff</guid>
            <category><![CDATA[aws-s3]]></category>
            <category><![CDATA[spring-boot]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[aws-lambda]]></category>
            <category><![CDATA[aws]]></category>
            <dc:creator><![CDATA[The Turtle Code]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 13:17:44 GMT</pubDate>
            <atom:updated>2026-11-15T13:17:43.163Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Why Naive Vector RAG Fails in Production: A Practical Guide to Hybrid Search + Reranking]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://alt.ctrlmag.com/why-naive-vector-rag-fails-in-production-a-practical-guide-to-hybrid-search-reranking-9080721400a9?source=rss----a0bc969b4f51---4"><img src="https://cdn-images-1.medium.com/max/2600/1*Ilbo1f_hiGweWEJL04RuEg.png" width="2800"></a></p><p class="medium-feed-snippet">Combining PostgreSQL pgvector semantic similarity, tsvector keyword search, and Reciprocal Rank Fusion (RRF) with a cross-encoder reranker&#x2026;</p><p class="medium-feed-link"><a href="https://alt.ctrlmag.com/why-naive-vector-rag-fails-in-production-a-practical-guide-to-hybrid-search-reranking-9080721400a9?source=rss----a0bc969b4f51---4">Continue reading on Alt. »</a></p></div>]]></description>
            <link>https://alt.ctrlmag.com/why-naive-vector-rag-fails-in-production-a-practical-guide-to-hybrid-search-reranking-9080721400a9?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/9080721400a9</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[rags]]></category>
            <category><![CDATA[postgresql]]></category>
            <dc:creator><![CDATA[Aslam Hossin]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 12:42:58 GMT</pubDate>
            <atom:updated>2026-11-15T12:42:56.667Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Why AI Forgets What You Told It.]]></title>
            <link>https://alt.ctrlmag.com/why-ai-forgets-what-you-told-it-e06ae1f6f11f?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/e06ae1f6f11f</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[generative-ai]]></category>
            <category><![CDATA[ai-engineering]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Mohammad Yasir]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 06:40:02 GMT</pubDate>
            <atom:updated>2026-11-15T06:40:01.043Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NiuGPSr9cVp3kNqr_AKoSQ.png" /></figure><p><em>Your AI chatbot can answer complex questions, but sometimes it forgets something you said just a few messages ago. Here’s why that happens and how developers can build AI applications that remember.</em></p><p>You are chatting with an AI assistant.</p><p>You tell it your name, explain what you’re building, share your preferences, and ask it to help with a task.</p><p>Everything works perfectly.</p><p>Then, a few messages later, you ask a follow-up question.</p><p>Suddenly, the AI asks you to explain something you already told it.</p><p>It feels like the AI has forgotten the conversation.</p><p>But here’s the interesting part: <strong>most AI models don’t actually remember conversations the way humans do.</strong></p><p>They generate responses using the information provided to them with each request.</p><p>And understanding this changes how we build AI applications.</p><h3>1. AI Doesn’t Remember Every Conversation</h3><p>When we use an AI chatbot, it can feel like we’re talking to something that continuously remembers everything.</p><p>But the underlying process is different.</p><p>A typical language model receives input, processes it, and generates an output. It doesn’t automatically retain the complete history of every conversation between requests.</p><p>In many applications, the developer has to send relevant conversation history again whenever the model is called.</p><p>For example:</p><pre>const messages = [<br>  {<br>    role: &quot;system&quot;,<br>    content: &quot;You are a helpful assistant.&quot;<br>  },<br>  {<br>    role: &quot;user&quot;,<br>    content: &quot;I&#39;m building a SaaS product.&quot;<br>  },<br>  {<br>    role: &quot;assistant&quot;,<br>    content: &quot;What kind of product?&quot;<br>  },<br>  {<br>    role: &quot;user&quot;,<br>    content: &quot;An AI writing tool.&quot;<br>  }<br>];</pre><p>The application sends these messages to the model.</p><p>The model can use the supplied context to understand the conversation and respond.</p><p>But if the application sends only the latest message, the model may have no idea what happened earlier.</p><p>That’s the first important lesson:</p><p><strong>Conversation history is something your application manages, not something you should assume the model automatically remembers.</strong></p><h3>2. What Is a Context Window?</h3><p>Every language model has a context window.</p><p>Think of it as the amount of information the model can consider during a particular request.</p><p>It can include:</p><ul><li>System instructions</li><li>User messages</li><li>Previous assistant responses</li><li>Retrieved documents</li><li>Tool results</li><li>Other application-provided information</li></ul><p>All of this consumes tokens.</p><p>A token is a small unit of text processed by the model. It can represent a word, part of a word, or punctuation.</p><p>Here’s a simplified example:</p><pre>flowchart TD<br>    A[&quot;AI Request&quot;] --&gt; B[&quot;System Instructions&quot;]<br>    A --&gt; C[&quot;Conversation History&quot;]<br>    A --&gt; D[&quot;Retrieved Information&quot;]<br>    A --&gt; E[&quot;Latest User Message&quot;]<br>    B --&gt; F[&quot;Context Window&quot;]<br>    C --&gt; F<br>    D --&gt; F<br>    E --&gt; F<br>    F --&gt; G[&quot;LLM&quot;]<br>    G --&gt; H[&quot;Generated Response&quot;]</pre><p>The model generates its response based on the information available in that context.</p><p>Now imagine a conversation that starts with 10 messages and eventually grows to 500 messages.</p><p>Sending the entire conversation every time becomes increasingly expensive and inefficient.</p><p>Eventually, the available context may not be enough to include everything.</p><p>This is where the real engineering problem begins.</p><h3>3. Why Sending the Entire Conversation Doesn’t Work</h3><p>One simple approach is to store every message and send the complete history with every request.</p><p>For a small chatbot, this might work.</p><p>But as the conversation grows, several problems appear.</p><p><strong>Higher token usage</strong></p><p>More conversation history means more input tokens. This can increase API costs.</p><p><strong>Slower responses</strong></p><p>Processing a larger context can increase latency, depending on the model and workload.</p><p><strong>Context limits</strong></p><p>Every model has a maximum context capacity. Once the available space is exhausted, your application must decide what information to retain.</p><p><strong>Irrelevant information</strong></p><p>Not every previous message matters to the current question.</p><p>Imagine a user discussing their business idea, then asking for help debugging a database query.</p><p>Does the model really need the entire business discussion to answer that technical question?</p><p>Probably not.</p><p>The goal isn’t to remember everything.</p><p>The goal is to provide the right information at the right time.</p><h3>4. Short-Term Context vs. Long-Term Memory</h3><p>This is where AI application architecture becomes interesting.</p><p>We can think about memory in two broad categories.</p><h3>Short-term context</h3><p>Short-term context contains information relevant to the current conversation.</p><p>For example:</p><ul><li>Recent messages</li><li>Current task</li><li>Previous question and answer</li><li>Temporary instructions</li></ul><p>This information helps the model maintain conversational continuity.</p><h3>Long-term memory</h3><p>Long-term memory contains information that may remain useful across multiple conversations.</p><p>For example:</p><ul><li>User preferences</li><li>Frequently used settings</li><li>Important project details</li><li>Previously established facts</li><li>Repeated user requirements</li></ul><p>Instead of sending everything to the model, the application can store useful information and retrieve it when needed.</p><p>Here’s a simplified architecture:</p><pre>flowchart TD<br>    A[&quot;User Message&quot;] --&gt; B[&quot;AI Application&quot;]<br>    B --&gt; C[&quot;Recent Conversation&quot;]<br>    B --&gt; D[&quot;Memory Retrieval&quot;]<br>    D --&gt; E[&quot;Stored User Information&quot;]<br>    C --&gt; F[&quot;Context Builder&quot;]<br>    E --&gt; F<br>    F --&gt; G[&quot;LLM&quot;]<br>    G --&gt; H[&quot;Response&quot;]<br>    H --&gt; I[&quot;Memory Update&quot;]<br>    I --&gt; E</pre><p>Notice something important.</p><p>The memory system is separate from the language model.</p><p>The application decides what to store, what to retrieve, and what to include in the next request.</p><p>This is how developers can give an AI application a more consistent memory experience.</p><h3>5. How Do We Actually Build AI Memory?</h3><p>There isn’t one universal memory architecture.</p><p>The right approach depends on what information your application needs to remember.</p><p>Let’s look at three practical techniques.</p><h3>A. Store important information in a database</h3><p>Not everything needs embeddings or a vector database.</p><p>If you want to remember a user’s name, preferred language, or subscription plan, a normal database is often enough.</p><p>For example:</p><pre>{<br>  &quot;userId&quot;: &quot;user_123&quot;,<br>  &quot;preferences&quot;: {<br>    &quot;language&quot;: &quot;English&quot;,<br>    &quot;responseStyle&quot;: &quot;concise&quot;<br>  },<br>  &quot;projects&quot;: [<br>    &quot;AI writing platform&quot;<br>  ]<br>}</pre><p>When the user starts a new conversation, your application can retrieve these details and include relevant information in the model’s context.</p><p>This is structured memory.</p><p>It’s predictable, easy to query, and useful for information with a known structure.</p><h3>B. Use summaries for long conversations</h3><p>Imagine a user has a conversation containing hundreds of messages.</p><p>Instead of sending every message, the application can periodically generate a summary.</p><p>For example:</p><p>Original conversation:</p><ul><li>User explains a SaaS idea.</li><li>Discusses the technology stack.</li><li>Changes the pricing model.</li><li>Explains the target audience.</li><li>Discusses deployment choices.</li></ul><p>A summary might look like this:</p><pre>The user is building an AI writing SaaS.</pre><pre>Technology:<br>Next.js, TypeScript, MongoDB.</pre><pre>Target audience:<br>SEO freelancers and content marketers.</pre><pre>Current priority:<br>Improve content generation and publishing.</pre><p>The application can then use this summary alongside recent messages.</p><p>This reduces unnecessary context while preserving important information.</p><p>However, summaries can lose details. Important decisions, constraints, and exact values should be stored separately when they need to remain accurate.</p><h3>C. Use retrieval for larger amounts of information</h3><p>What if your application needs to remember thousands of documents, conversations, or project notes?</p><p>Loading everything into the context window isn’t practical.</p><p>This is where retrieval-based memory becomes useful.</p><p>A common approach is:</p><ol><li>Convert selected information into embeddings.</li><li>Store those embeddings in a vector database.</li><li>Convert the user’s new question into an embedding.</li><li>Search for semantically relevant information.</li><li>Retrieve the best matching results.</li><li>Add those results to the model’s context.</li></ol><p>For example, a user asks:</p><p>“Which database did we choose for the project?”</p><p>The application searches stored information and retrieves a previous decision about PostgreSQL.</p><p>The model receives that information and generates an answer.</p><p>The important distinction is that the vector database doesn’t make the model remember.</p><p>It helps your application find relevant information to provide to the model.</p><h3>6. When Should Your Application Retrieve, Summarize, or Store?</h3><p>This is one of the most important design decisions.</p><p>Different information requires different treatment.</p><p>InformationRecommended approachRecent messagesConversation historyUser preferencesStructured databaseLong conversationsSummariesLarge document collectionsVector searchImportant project decisionsStructured recordsTemporary task instructionsCurrent context</p><p>Using a vector database for every piece of information is unnecessary.</p><p>Likewise, storing every conversation as one large text field doesn’t automatically create a useful memory system.</p><p>Choose the storage method based on how the information will be retrieved and used.</p><h3>7. Memory Introduces New Engineering Problems</h3><p>Giving an AI application memory makes it more useful, but it also introduces responsibilities.</p><p><strong>Stale information</strong></p><p>A user might change their preferences or abandon an old project.</p><p>Your memory system must support updates and deletion.</p><p><strong>Incorrect retrieval</strong></p><p>A vector search system might return information that looks similar but belongs to a different topic or project.</p><p>Retrieved information should be checked for relevance before being included in the prompt.</p><p><strong>Privacy and isolation</strong></p><p>One user’s memories must never appear in another user’s conversation.</p><p>Memory records need proper ownership, authorization, and tenant isolation.</p><p><strong>Context quality</strong></p><p>Retrieving too much information can be just as problematic as retrieving too little.</p><p>Your application needs to select useful information rather than blindly adding every search result.</p><p><strong>Memory control</strong></p><p>Users should be able to understand, correct, and delete information stored about them.</p><p>Memory isn’t just a feature. It’s also a data management responsibility.</p><h3>8. The Architecture Matters More Than the Model</h3><p>It’s tempting to think that choosing a more powerful language model will solve memory problems.</p><p>But a better model cannot reliably remember information that your application never provides.</p><p>A well-designed AI application needs several components working together:</p><ul><li>A conversation store</li><li>A memory storage layer</li><li>A retrieval mechanism</li><li>A context builder</li><li>A memory update strategy</li><li>Privacy and authorization controls</li></ul><p>The language model is only one part of this architecture.</p><pre>flowchart TD<br>    A[&quot;User&quot;] --&gt; B[&quot;AI Application&quot;]<br>    B --&gt; C[&quot;Conversation Manager&quot;]<br>    B --&gt; D[&quot;Memory Manager&quot;]<br>    D --&gt; E[&quot;Structured Database&quot;]<br>    D --&gt; F[&quot;Vector Database&quot;]<br>    C --&gt; G[&quot;Context Builder&quot;]<br>    E --&gt; G<br>    F --&gt; G<br>    G --&gt; H[&quot;Language Model&quot;]<br>    H --&gt; I[&quot;Response&quot;]<br>    I --&gt; C<br>    I --&gt; D</pre><p>You don’t need all these components on day one.</p><p>A simple chatbot might only need recent conversation history.</p><p>As the product grows, you can introduce summaries, structured memory, and retrieval.</p><p>The architecture should evolve with the actual needs of the application.</p><h3>Final Thoughts</h3><p>AI doesn’t necessarily forget because the model is unintelligent.</p><p>Often, it forgets because the application hasn’t provided the information needed to answer the next question.</p><p>The real challenge isn’t making an AI remember everything.</p><p>It’s deciding:</p><ul><li>What information is worth remembering?</li><li>Where should it be stored?</li><li>When should it be retrieved?</li><li>How much context should be sent?</li><li>When should outdated information be removed?</li></ul><p>These are application architecture decisions, not just prompt engineering decisions.</p><p><strong>A good AI application doesn’t need to remember everything. It needs to remember what matters.</strong></p><p><strong>About the author</strong></p><p>Mohammad Yasir is a full-stack engineer building backend systems and AI-powered products. He writes about AI engineering, backend development, and system design.</p><p>Explore more of his work at <a href="https://codilad.dev">codilad.dev</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e06ae1f6f11f" width="1" height="1" alt=""><hr><p><a href="https://alt.ctrlmag.com/why-ai-forgets-what-you-told-it-e06ae1f6f11f">Why AI Forgets What You Told It.</a> was originally published in <a href="https://alt.ctrlmag.com">Alt.</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Scientists Are Sleeping on These 18 Claude Skills]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://alt.ctrlmag.com/data-scientists-are-sleeping-on-these-18-claude-skills-356267e38543?source=rss----a0bc969b4f51---4"><img src="https://cdn-images-1.medium.com/max/2600/0*YtDD9crQpM9NFNoH" width="3200"></a></p><p class="medium-feed-snippet">Most data scientists have experimented with AI by now.</p><p class="medium-feed-link"><a href="https://alt.ctrlmag.com/data-scientists-are-sleeping-on-these-18-claude-skills-356267e38543?source=rss----a0bc969b4f51---4">Continue reading on Alt. »</a></p></div>]]></description>
            <link>https://alt.ctrlmag.com/data-scientists-are-sleeping-on-these-18-claude-skills-356267e38543?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/356267e38543</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[self-awareness]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[learn with her]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 06:39:37 GMT</pubDate>
            <atom:updated>2026-11-15T06:39:36.111Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Most Developers Know ChatGPT. These AI Coding Tools Go Further]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://alt.ctrlmag.com/most-developers-know-chatgpt-these-ai-coding-tools-go-further-88d9470b8ef2?source=rss----a0bc969b4f51---4"><img src="https://cdn-images-1.medium.com/max/2600/0*NJp12XHrnwK4pCjh" width="3840"></a></p><p class="medium-feed-snippet">ChatGPT is a great starting point for AI-assisted programming. But these 10 tools can help you edit entire codebases, automate development&#x2026;</p><p class="medium-feed-link"><a href="https://alt.ctrlmag.com/most-developers-know-chatgpt-these-ai-coding-tools-go-further-88d9470b8ef2?source=rss----a0bc969b4f51---4">Continue reading on Alt. »</a></p></div>]]></description>
            <link>https://alt.ctrlmag.com/most-developers-know-chatgpt-these-ai-coding-tools-go-further-88d9470b8ef2?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/88d9470b8ef2</guid>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[self-improvement]]></category>
            <dc:creator><![CDATA[learn with her]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 06:39:14 GMT</pubDate>
            <atom:updated>2026-11-15T06:39:12.810Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[From Traffic Spike to Recovery: The Complete DDoS Defense Playbook for Modern Applications]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://alt.ctrlmag.com/from-traffic-spike-to-recovery-the-complete-ddos-defense-playbook-for-modern-applications-da082ccda03e?source=rss----a0bc969b4f51---4"><img src="https://cdn-images-1.medium.com/max/1672/1*EckzVlloBlGgPbRyixqGww.png" width="1672"></a></p><p class="medium-feed-snippet">How to detect availability attacks, protect APIs, build resilient cloud infrastructure, and respond to incidents in the age of AI</p><p class="medium-feed-link"><a href="https://alt.ctrlmag.com/from-traffic-spike-to-recovery-the-complete-ddos-defense-playbook-for-modern-applications-da082ccda03e?source=rss----a0bc969b4f51---4">Continue reading on Alt. »</a></p></div>]]></description>
            <link>https://alt.ctrlmag.com/from-traffic-spike-to-recovery-the-complete-ddos-defense-playbook-for-modern-applications-da082ccda03e?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/da082ccda03e</guid>
            <category><![CDATA[backend-development]]></category>
            <category><![CDATA[distributed-systems]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[system-design-interview]]></category>
            <dc:creator><![CDATA[The Turtle Code]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 06:38:51 GMT</pubDate>
            <atom:updated>2026-11-15T06:38:50.416Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Identity Compromise: How Attackers Exploit OAuth Tokens, Service Accounts and Privileged Access in…]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://alt.ctrlmag.com/identity-compromise-how-attackers-exploit-oauth-tokens-service-accounts-and-privileged-access-in-0157ba1d5105?source=rss----a0bc969b4f51---4"><img src="https://cdn-images-1.medium.com/max/1672/1*8wBmrR7EeryWObF23hWlnQ.png" width="1672"></a></p><p class="medium-feed-snippet">A production security engineering deep dive with Java Spring Boot, Spring Security, OAuth 2.0, OIDC, JWT and Cloud IAM.</p><p class="medium-feed-link"><a href="https://alt.ctrlmag.com/identity-compromise-how-attackers-exploit-oauth-tokens-service-accounts-and-privileged-access-in-0157ba1d5105?source=rss----a0bc969b4f51---4">Continue reading on Alt. »</a></p></div>]]></description>
            <link>https://alt.ctrlmag.com/identity-compromise-how-attackers-exploit-oauth-tokens-service-accounts-and-privileged-access-in-0157ba1d5105?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/0157ba1d5105</guid>
            <category><![CDATA[devops]]></category>
            <category><![CDATA[backend-development]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[system-design-interview]]></category>
            <dc:creator><![CDATA[The Turtle Code]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 06:38:24 GMT</pubDate>
            <atom:updated>2026-11-15T06:38:23.330Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[How To Keep Two Flawless Security Tools From Making One Giant Mess]]></title>
            <link>https://alt.ctrlmag.com/this-shows-why-two-flawless-security-tools-can-make-one-giant-mess-74f374b28e8d?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/74f374b28e8d</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[infosec]]></category>
            <category><![CDATA[cyberattack]]></category>
            <category><![CDATA[cyber-security-solutions]]></category>
            <dc:creator><![CDATA[Adeleye Emmanuel Odafe]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 06:37:49 GMT</pubDate>
            <atom:updated>2026-11-15T06:37:48.632Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Two cybersecurity analysts watching in surprise as a massive, surreal holographic collision of digital titans erupts from a dimly lit server room’s racks. AI Generated" src="https://cdn-images-1.medium.com/max/1024/1*33nOPQwhlPk5VA6DFz5M-A.png" /></figure><p>According to our AI-powered SIEM, a critical server in the production environment was suddenly attempting a massive, unauthorized data transfer to an untrusted external IP. Seconds later, the EDR credential dumping on the exact same machine. It was presumed a textbook case of a potential ransomware : an attacker had compromised an identity, moved laterally, and was now exfiltrating data.</p><p>Within three minutes, the automated playbooks of the solution stack in house did exactly what they were bought to do. They isolated the server, blocked the outbound network path, and revoked the active developer credentials.</p><p>The machine had “saved the day.”</p><p>Except, when the night-shift analyst actually started digging into the telemetry, they discovered there was no attacker. There was no malware. There wasn’t even a breach.</p><h3>Following Instructions Straight Into a Collision</h3><blockquote>Let&#39;s imagine a soccer derby match where two defenders on the same team aren’t allowed to communicate. One lunges in to clear the ball; the other slides in from the blind side to intercept the same pass, taking each other out while the striker walks past untouched at the same time. The keeper rushes off his line to sweep up a ball that isn’t there. To the crowd, it looks like a disaster. To the players, they’re just following the instructions they were given.</blockquote><p>In the 2:14 AM incident, the engineer was trying to run a scheduled, autonomous vulnerability pentest to find weaknesses before an audit. To be thorough, the pentesting tool was doing exactly what it was designed to do: mimicking an attacker by probing internal assets and testing our identity defences.</p><p>But the SIEM and the network detection tool hadn’t been fully clued into the plan. Because the tools operate in separate silos, the SIEM saw the heavily aggressive network activity and assumed the worst. Meanwhile, the endpoint protection saw the “hostile” attempts to read memory and pulled the plug.</p><p>Neither tool was broken. Both were functioning exactly as advertised. But because they couldn’t see the context- in this case, the human intent behind the chaos- they ended up fighting each other, taking down a critical staging server, and giving the SOC team a collective heart attack.</p><h3>The Blind Spots Created by Too Much Noise</h3><p>When we talk about the “human-machine grey area,” this is exactly the kind of friction we mean. Everyone worries about “shadow AI” or unpatched vulnerabilities (which we try to handle with tools like Gytpol and Vicarius), but we rarely talk about the <strong>configuration decay</strong> and <strong>telemetry blindness</strong> that happens when your tools start screaming at the same time.</p><p>For an intern or a newbie just getting into the field, the instinct is to trust the dashboard. If the “Risk Score” is red, you panic. If it’s green, you stay chill.</p><p>But as a threat hunter, I know that a stack of autonomous tools without human orchestration is just an expensive noise generator full of falsepositives and benign alerts and here is why:</p><ol><li><strong>Automation over-trust builds fatigue.</strong> If your SIEM is constantly flagging automated pentests, background patch cycles, or legitimate micro-segmentation checks as high-severity incidents, your human analysts will start ignoring the alerts, and that is how the real attack thats masquerading as “background noise” gets missed.</li><li><strong>“Un-agentable” assets create silent gaps.</strong> A lot of modern security relies on installing a piece of software (an agent) on every device. But what about IoT devices, medical hardware, or OT systems You can’t install a CrowdStrike agent on a connected hospital pump or a smart thermostat. If your NDR and your endpoint tools aren’t talking, those un-agentable devices become the ultimate blind spots. A huge visibility issue.</li><li><strong>The “Intent” Gap is where the real risk lives.</strong> A machine can tell you <em>what</em> is happening (e.g., “10GB of data is moving to IP X”). Only a human can tell you <em>why</em> (“Oh, that’s Emmanuel from SOC trying to back up the staging environment before he goes on leave”).</li></ol><h3>You Can’t Autopilot Security</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h0-YTIMXOcjdTBQuqhpxrw.png" /></figure><p>The security industry loves to promise “hands-off” protection. But the more complex your environment gets, the more you realise that the most critical piece of the cyber security solution stack isn’t the software it’s the human translator in the middle who uses it</p><p>We shouldn’t stop buying automated tools. Without AI-driven triage and autonomous testing, we’d be completely overwhelmed by the sheer volume of modern threats especially with AI misuse on the hands of bad actors</p><p>But we have to treat these tools like what they are: highly specialised, yet slightly stubborn specialists. They are amazing at execution, but entirely dependent on human context to understand when they’re helping, and when they’re just getting in each other’s way.</p><p>The next time a vendor tells you their platform or tool can replace your security team, remember the 2:14 AM “phantom breach.” Sometimes, the hardest part of defending a network isn’t stopping the bad guys , it’s keeping your own tools from burning the house down.</p><h3>Over to You</h3><p>What’s the most spectacular “tool collision” you’ve seen in your environment? Have you ever had your endpoint agent and your firewall go to war? comments.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=74f374b28e8d" width="1" height="1" alt=""><hr><p><a href="https://alt.ctrlmag.com/this-shows-why-two-flawless-security-tools-can-make-one-giant-mess-74f374b28e8d">How To Keep Two Flawless Security Tools From Making One Giant Mess</a> was originally published in <a href="https://alt.ctrlmag.com">Alt.</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Buying All the Right Security Tools Still Leaves You Blind]]></title>
            <link>https://alt.ctrlmag.com/why-buying-all-the-right-security-tools-still-leaves-you-blind-6db2db9b98a7?source=rss----a0bc969b4f51---4</link>
            <guid isPermaLink="false">https://medium.com/p/6db2db9b98a7</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[cyber-security-solutions]]></category>
            <category><![CDATA[infosec]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Adeleye Emmanuel Odafe]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 06:37:29 GMT</pubDate>
            <atom:updated>2026-11-15T06:37:28.171Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="An over-the-shoulder, candid 35mm film-style photograph of a bustling Security Operations Center (SOC) during a high-tension incident response. Multiple cybersecurity analysts are seen actively collaborating, gesture-pointing, and working across diverse multi-monitor stations. In the foreground, screens display various, natural blurred interface “solution stacks” — including SIEM log timelines, network connection and NDR traffic nodes, automated patching tables etc" src="https://cdn-images-1.medium.com/max/1024/1*cwr7YrolaF8xx1DQXDQpTw.png" /></figure><p>Imagine buying four different smart smoke detectors for your living room.</p><p>One detector monitors heat; another checks air particles. A third watches the security camera for flickering light. A fourth automatically turns on the ceiling sprinklers the second it suspects trouble.<br>Sounds safe, right? Except when someone burns toast in the kitchen, the sprinkler system trips immediately, flooding the room and shorting out the camera, so nobody ever sees the actual electrical fire smouldering behind the walls.</p><p>That is what happens inside modern security stacks when <strong>tool sprawl</strong> collides with <strong>over-automation</strong>.</p><p>When you layer preemptive edge blockers, network detection and response (NDR), identity protection, and automated patching into the same environment <strong><em>without accounting for how humans actually work</em></strong>, you create a strange new problem: <strong>blind spots caused by too much protection.</strong></p><h4>Where the Machine Stops and the Human Begins</h4><p>If you are new to cybersecurity or manage teams without living in a terminal all day, it is easy to assume “AI-driven” or “autonomous” tools think the way people do. They don’t. Models are brilliant at spotting mathematical patterns at scale, and terrible at understanding workplace context.</p><p>Here is where that gap shows up in real SOC reporting:</p><ul><li><strong>Preemptive blocks can erase your forensic trail.</strong> When an automated threat-intel gateway drops traffic from a known malicious IP right at the front door, your dashboard celebrates a <strong>“blocked attack.</strong>” Great. Except now your internal network sensors never see which compromised laptop inside the building was trying to call out to that address in the first place. The machine stopped the symptom and accidentally hid the patient.</li><li><strong>Autonomous pentests find paths that humans refuse to close.</strong> Modern automated pentesting tools are phenomenal at chaining together tiny Active Directory misconfigurations to prove how an attacker could reach domain admin in twenty minutes. The machine hands you the exact blueprint to fix it. Then a human IT manager looks at the fix, realises it might break a legacy payroll script from 2016, and quietly hits “defer until next quarter.”</li><li><strong>Misconfiguration tools and patching engines can step on each other’s toes.</strong> One platform tightens a Windows registry setting to harden an endpoint; an automated patch cycle or a rushed sysadmin script resets it forty-eight hours later to push an urgent update. Neither tool is broken; they just don’t know why the other one acted.</li></ul><h3>The “Intent” Gap Nobody Wants to Admit</h3><p>The biggest blind spot in modern security is not in a missing firewall rule. It is the space between <strong>what a user is allowed to do</strong> and <strong>what a user actually means to do</strong>.</p><p>Look at how people use AI tools at work today. You can deploy strict AI governance platforms to give visibility into approved models, and you can install workspace controls that watch for risky copy-pasting in real time.</p><p>Yet every week, a tired engineer or an overwhelmed finance lead will still find a creative workaround to take a screenshot of a sensitive spreadsheet, paste anonymised chunks into an unapproved browser tab, or grant an overly broad OAuth permission just to finish a report before 5:00 PM.</p><p>To the machine, everything looks completely fine, but it takes a human analyst someone who understands how people cut corners under deadline pressure to look at a “low-severity” alert and anomaly and ask: <em>Wait. Why is an accounts payable laptop querying an engineering repository at 4:55 PM on a Friday?</em></p><h3>Buying a Platform Isn’t the Same as Operating It</h3><p>Most software vendors love selling the idea of “set it and forget it.” They pitch autonomous triage, self-healing endpoints, and zero-touch containment as if security is an appliance you plug into the wall. Only a handful of vendors do due diligence, and this is why vendors need to be properly scrutinised before the POCs and demo presentations.</p><p>The best security solutions aren’t the ones with the most logos on their architecture slide. They are the ones where the tools handle the noisy, repetitive heavy lifting, freeing human analysts to investigate the messy grey areas where automation loses the plot.</p><p>Have you ever seen two “smart” security tools cancel each other out or create a blind spot in your environment? Drop your favourite tool-collision story below.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6db2db9b98a7" width="1" height="1" alt=""><hr><p><a href="https://alt.ctrlmag.com/why-buying-all-the-right-security-tools-still-leaves-you-blind-6db2db9b98a7">Why Buying All the Right Security Tools Still Leaves You Blind</a> was originally published in <a href="https://alt.ctrlmag.com">Alt.</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>