<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Ksatria Bintang Samudra</title>
    <description>The latest articles on DEV Community by Ksatria Bintang Samudra (@kstriabintang).</description>
    <link>https://dev.to/kstriabintang</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4171544%2F1cd0bbf8-91c6-45df-af36-cfdac232d4a5.png</url>
      <title>DEV Community: Ksatria Bintang Samudra</title>
      <link>https://dev.to/kstriabintang</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/kstriabintang"/>
    <language>en</language>
    <item>
      <title>Debloating Android without root — and proving it worked</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/debloating-android-without-root-and-proving-it-worked-3mj5</link>
      <guid>https://dev.to/kstriabintang/debloating-android-without-root-and-proving-it-worked-3mj5</guid>
      <description>&lt;p&gt;Phones ship full of software nobody asked for, and the usual "fixes" are worse than the problem: rooting voids warranties and breaks banking apps, and one-tap "boosters" mostly just show you an animation. I wanted a third option — remove bloat &lt;strong&gt;without root&lt;/strong&gt;, keep every change &lt;strong&gt;reversible&lt;/strong&gt;, and then &lt;strong&gt;prove&lt;/strong&gt; the phone actually got faster. That's adb-kit.&lt;/p&gt;

&lt;h2&gt;
  
  
  No root, fully reversible
&lt;/h2&gt;

&lt;p&gt;adb-kit drives &lt;strong&gt;wireless ADB&lt;/strong&gt;. It disables or uninstalls packages &lt;em&gt;per-user&lt;/em&gt;, which needs no root and touches nothing in the system partition. Because the operations are per-user, every change is reversible — you can bring any package back. Nothing is permanent, so there's no "I bricked my phone" failure mode.&lt;/p&gt;

&lt;h2&gt;
  
  
  Measure, don't vibe
&lt;/h2&gt;

&lt;p&gt;The part I care about most: adb-kit profiles the result with &lt;code&gt;gfxinfo&lt;/code&gt; so the effect is a number, not a feeling. Anyone can claim a phone "feels snappier." I'd rather show frame times.&lt;/p&gt;

&lt;p&gt;On a &lt;strong&gt;vivo V21&lt;/strong&gt;, here's the measured before/after:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Packages:&lt;/strong&gt; 327 → 275 (&lt;strong&gt;52 removed&lt;/strong&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Average frame time:&lt;/strong&gt; 18 ms → &lt;strong&gt;12 ms&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Janky frames:&lt;/strong&gt; 0.27% → &lt;strong&gt;0.00%&lt;/strong&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Those are documented in the repo with screenshots, on the same device, same workload. A ~33% drop in frame time is the difference between "mostly smooth" and "smooth."&lt;/p&gt;

&lt;h2&gt;
  
  
  Why reversibility changes the psychology
&lt;/h2&gt;

&lt;p&gt;The honest reason most people never debloat is fear — fear of removing the wrong thing. Making every action reversible removes the fear, which is what makes the toolkit actually usable rather than just technically correct. You can be aggressive precisely because you can undo it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The principle
&lt;/h2&gt;

&lt;p&gt;I don't ship "boosters." I ship tools that make a claim and then back it with a measurement you can reproduce. If I can't show the before and after, I don't get to say it got faster.&lt;/p&gt;

</description>
      <category>android</category>
      <category>privacy</category>
      <category>mobile</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Payment-native from day one — building with QRIS in Indonesia</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Sun, 08 Nov 2026 05:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/payment-native-from-day-one-building-with-qris-in-indonesia-2epa</link>
      <guid>https://dev.to/kstriabintang/payment-native-from-day-one-building-with-qris-in-indonesia-2epa</guid>
      <description>&lt;p&gt;Most payment tutorials assume a credit card and a Western checkout. In Indonesia that assumption quietly blocks a huge share of your users. The default here is &lt;strong&gt;QRIS&lt;/strong&gt; — one QR standard that every bank and e-wallet accepts. If your product can't take a QRIS payment, for many people it simply can't take their money.&lt;/p&gt;

&lt;p&gt;So I build payment-native from day one. Vensix tops up its balance with QRIS. A Telegram storefront bot I built takes QRIS orders. An ISP I work with collects subscriber bills over QRIS. Same primitive, three very different products.&lt;/p&gt;

&lt;h2&gt;
  
  
  The one rule: verify on the server
&lt;/h2&gt;

&lt;p&gt;The single most important thing about payments is also the easiest to get wrong: &lt;strong&gt;the client is not allowed to decide that a payment succeeded.&lt;/strong&gt; A "payment complete" message in the app is a UI state, not a fact. The fact lives on the server, confirmed against the payment provider.&lt;/p&gt;

&lt;p&gt;Concretely, the flow I use is:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The server creates the charge and stores it as &lt;code&gt;pending&lt;/code&gt; with the exact amount and an order reference.&lt;/li&gt;
&lt;li&gt;The user pays by scanning the QRIS code.&lt;/li&gt;
&lt;li&gt;The provider notifies my server (webhook) &lt;strong&gt;and&lt;/strong&gt; my server independently re-checks the charge status against the provider's API — never trusting the webhook payload alone.&lt;/li&gt;
&lt;li&gt;Only when the server confirms the amount and reference does the order flip to &lt;code&gt;paid&lt;/code&gt; and the goods get released.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If the app crashes, if the webhook is spoofed, if the user closes the tab — the truth is still correct, because it was never the client's to decide.&lt;/p&gt;

&lt;h2&gt;
  
  
  Idempotency and race conditions are the real work
&lt;/h2&gt;

&lt;p&gt;Two things bite everyone building payments:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Duplicate events.&lt;/strong&gt; Providers retry webhooks. If your handler isn't idempotent, one payment can credit a balance twice. Every charge gets a unique reference, and applying it is a no-op the second time.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Races.&lt;/strong&gt; Two requests for the last unit in stock, or two top-ups landing at once. I settle these in the database — a transaction plus a uniqueness constraint — so the outcome can't depend on timing.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why this matters for the market
&lt;/h2&gt;

&lt;p&gt;Building payment-native isn't just a technical nicety here — it's the difference between a product Indonesians can actually use and one they can't. QRIS removes the card requirement, the foreign-currency friction, and the sign-up-for-a-processor barrier. The engineering discipline — server-side truth, idempotency, transactional settlement — is what keeps it trustworthy once real money is moving.&lt;/p&gt;

&lt;p&gt;Money is the one place where "it works on my machine" is never good enough. Verify on the server, make it idempotent, settle it in the database.&lt;/p&gt;

</description>
      <category>fintech</category>
      <category>payments</category>
      <category>webdev</category>
      <category>indonesia</category>
    </item>
    <item>
      <title>Shipping a whole business on Cloudflare Workers</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Sun, 08 Nov 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/shipping-a-whole-business-on-cloudflare-workers-28dl</link>
      <guid>https://dev.to/kstriabintang/shipping-a-whole-business-on-cloudflare-workers-28dl</guid>
      <description>&lt;p&gt;For a solo builder, the traditional server is a tax you pay forever: patching, scaling, monitoring, a monthly bill whether anyone shows up or not. For most of what I ship, I've stopped paying it. I run on &lt;strong&gt;Cloudflare Workers&lt;/strong&gt; — code that executes at the edge, close to the user, with no server to babysit.&lt;/p&gt;

&lt;p&gt;Makmur Motor (a Denpasar dealership's showroom + inventory CMS) runs on Workers with D1 and KV. Venmail (a privacy-first temporary inbox) runs on Workers. This portfolio — including its contact form and a Workers-AI chat assistant — is a Worker serving static assets. Three different products, one operational model.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the edge actually buys you
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;No servers to manage.&lt;/strong&gt; There's no VM to keep alive, no OS to patch, no region to choose. The code runs in hundreds of locations automatically.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Latency is low by default&lt;/strong&gt; because the code runs near the visitor, not in one distant data centre.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The bill rounds to zero at low traffic.&lt;/strong&gt; A dealership site or a niche tool doesn't need to cost real money to keep online — that changes which projects are even worth shipping.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The storage primitives make it a real platform, not just a CDN: &lt;strong&gt;D1&lt;/strong&gt; (SQLite at the edge) for relational data, &lt;strong&gt;KV&lt;/strong&gt; for fast key-value reads, &lt;strong&gt;R2&lt;/strong&gt; for files, and Workers AI for inference right where the request lands.&lt;/p&gt;

&lt;h2&gt;
  
  
  The trade-offs I plan around
&lt;/h2&gt;

&lt;p&gt;The edge is not a normal server, and pretending otherwise causes pain:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;No long-lived connections or big background jobs.&lt;/strong&gt; A Worker is short-lived and request-scoped. Anything long-running belongs in a queue or a durable object, not a "just keep it running" process.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;D1 is SQLite, not Postgres.&lt;/strong&gt; It's excellent for edge-native apps, but when a product needs heavy relational work, row-level security and complex queries, I reach for Postgres/Supabase instead. Right tool, right job.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cold, distributed state&lt;/strong&gt; means you design for eventual consistency where it's fine and strong consistency (a transaction) only where money or correctness demands it.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The honest summary
&lt;/h2&gt;

&lt;p&gt;The edge isn't magic and it isn't right for everything — a data-heavy system of record still wants Postgres. But for the long tail of real products a small team ships, "no servers, low latency, near-zero idle cost" is a genuinely better default. It's why I can run several live products at once without a server bill or a pager.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>cloudflare</category>
      <category>edge</category>
      <category>programming</category>
    </item>
    <item>
      <title>One key, five LLMs — building Vensix, an OpenAI-compatible gateway</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Sat, 07 Nov 2026 12:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/one-key-five-llms-building-vensix-an-openai-compatible-gateway-4kei</link>
      <guid>https://dev.to/kstriabintang/one-key-five-llms-building-vensix-an-openai-compatible-gateway-4kei</guid>
      <description>&lt;p&gt;Every model provider wants to be your only provider. Different keys, different SDKs, different billing, different dashboards. If you build anything real on top of LLMs, you feel that friction fast — and your users feel it worse.&lt;/p&gt;

&lt;p&gt;Vensix is my answer: &lt;strong&gt;one API key that routes to Claude, GPT, Gemini, DeepSeek and Kimi&lt;/strong&gt;, speaks the OpenAI-compatible protocol, and meters cost as you go.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why OpenAI-compatible
&lt;/h2&gt;

&lt;p&gt;The single best interoperability decision you can make in this space is to expose the OpenAI Chat Completions shape. Almost every client library, framework and no-code tool already speaks it. So Vensix doesn't ask developers to learn anything new — you point your existing OpenAI client at a new base URL, swap the key, and change the model string. That's the whole migration.&lt;/p&gt;

&lt;p&gt;Behind that familiar surface, Vensix decides where a request actually goes. The model name selects the upstream; the gateway normalises the request, forwards it, and translates the response back into the shape the caller expects.&lt;/p&gt;

&lt;h2&gt;
  
  
  Metering is the hard part, not routing
&lt;/h2&gt;

&lt;p&gt;Routing is easy. &lt;strong&gt;Honest, real-time cost metering is the feature.&lt;/strong&gt; Each upstream prices tokens differently, and prompt versus completion tokens are priced differently again. Vensix reads the usage from every response, prices it against the specific model that served it, and debits a balance — so a user sees one number going down, not five invoices arriving later.&lt;/p&gt;

&lt;p&gt;For the Indonesian market that balance is topped up with &lt;strong&gt;QRIS&lt;/strong&gt;, verified server-side. No cards required, no foreign-currency friction — the thing that quietly blocks a lot of local developers from using frontier models at all.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I'd tell someone building the same
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Pick the compatible protocol first.&lt;/strong&gt; Your integration surface is your moat, and "works with what people already have" beats "elegant but new."&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Treat usage accounting as core, not telemetry.&lt;/strong&gt; If money moves, the numbers have to be right and reconcilable, not approximate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fail predictably.&lt;/strong&gt; When an upstream is down, the caller should get a clear, typed error in the shape they already handle — not a surprise.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;A gateway is unglamorous infrastructure. But making five models feel like one, billed in local money, is exactly the kind of plumbing that lets other people build.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>llm</category>
      <category>api</category>
      <category>programming</category>
    </item>
    <item>
      <title>How I audit AI-written code — and the bug that proved why</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Sat, 07 Nov 2026 05:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/how-i-audit-ai-written-code-and-the-bug-that-proved-why-bik</link>
      <guid>https://dev.to/kstriabintang/how-i-audit-ai-written-code-and-the-bug-that-proved-why-bik</guid>
      <description>&lt;p&gt;I build most of my software as an AI-augmented solo developer. I write the spec, I own the architecture, and I drive Claude Code and Cursor through the implementation. But there is one rule I never break: &lt;strong&gt;I audit what actually ships, not what the model tells me it produced.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Here is the bug that turned that rule into a habit.&lt;/p&gt;

&lt;h2&gt;
  
  
  The symptom that shouldn't exist
&lt;/h2&gt;

&lt;p&gt;CoalTrack is an attendance and payroll app for coal-mining crews. During closed testing, a handful of devices simply could not reach the server — no error, no retry loop that recovered, just silence. The code looked correct. The API was up. Every earlier build had "worked on my machine."&lt;/p&gt;

&lt;p&gt;So I stopped reading the source and started reading the &lt;strong&gt;artifact&lt;/strong&gt;. I pulled apart the signed app bundle and inspected the merged &lt;code&gt;AndroidManifest.xml&lt;/code&gt; that Google Play actually receives.&lt;/p&gt;

&lt;p&gt;The &lt;code&gt;android.permission.INTERNET&lt;/code&gt; line wasn't there.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why the source lied
&lt;/h2&gt;

&lt;p&gt;In a multi-module Flutter build, the manifest you write is not the manifest that ships. Plugins, build types and manifest merging all rewrite it. A change somewhere upstream had dropped the INTERNET permission from the merged result — so on a fresh install with no cached grant, the app had no network at all. On my phone it worked because the permission had been granted by an earlier build and survived the upgrade.&lt;/p&gt;

&lt;p&gt;Reading the source would never have found this. Reading the shipped bundle found it in minutes.&lt;/p&gt;

&lt;h2&gt;
  
  
  The discipline
&lt;/h2&gt;

&lt;p&gt;That is the whole point of AI-augmented work done responsibly. The model accelerates the typing; it does not absolve me of verification. Concretely, that means:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Verify the artifact, not the intent.&lt;/strong&gt; Inspect the built bundle, the merged manifest, the generated SQL migration — the thing that actually runs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Write the risky logic twice, on purpose.&lt;/strong&gt; CoalTrack computes Indonesian statutory payroll in a PHP service &lt;em&gt;and&lt;/em&gt; a Dart mirror, then cross-checks them with mirrored tests. If the model drifts on one side, the other catches it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Gate everything with tests.&lt;/strong&gt; 400+ automated tests run in CI before any signed build. AI writing more code is only safe if more code is also checked.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Read the diff like a reviewer, not an author.&lt;/strong&gt; The model is an eager junior engineer. Treat its output that way.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The takeaway
&lt;/h2&gt;

&lt;p&gt;"AI wrote it" is not a defect and it is not a badge. The engineering is still mine — the spec, the review, the decision about what is correct. The productivity is real, but so is the responsibility. Ship the audit, not the optimism.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>codereview</category>
      <category>security</category>
      <category>programming</category>
    </item>
    <item>
      <title>Vibe coding is fun — until it hits production</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Sat, 07 Nov 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/vibe-coding-is-fun-until-it-hits-production-do4</link>
      <guid>https://dev.to/kstriabintang/vibe-coding-is-fun-until-it-hits-production-do4</guid>
      <description>&lt;p&gt;"Vibe coding" is the perfect name. You describe what you want, the AI produces it, you run it, and if it works, you move on — steering by feel, never really reading the code. It is genuinely fun, genuinely fast, and genuinely fine… right up until it isn't.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where vibe coding is great
&lt;/h2&gt;

&lt;p&gt;I'm not here to gatekeep. Vibe coding is a legitimately good tool for:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Prototypes and spikes&lt;/strong&gt; — proving an idea in an hour instead of a day.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Learning&lt;/strong&gt; — poking at a new framework by generating and running things.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Throwaway scripts&lt;/strong&gt; — one-off tasks where "works once" is the whole requirement.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;For those, not reading every line is a &lt;em&gt;feature&lt;/em&gt;. Speed is the point, and the cost of being wrong is basically zero.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where it quietly bites
&lt;/h2&gt;

&lt;p&gt;The trouble starts when vibe-coded code sneaks into production — because the same "don't read it, just ship it" that made it fast becomes the liability:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Security holes you never saw.&lt;/strong&gt; The model happily writes code that concatenates user input into a query or trusts the client. If you didn't read it, you didn't catch it. (This is how classics like &lt;a href="https://ksatriabintangsamudra.com/glossary/#sql-injection" rel="noopener noreferrer"&gt;SQL injection&lt;/a&gt; get shipped.)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Edge cases nobody handled.&lt;/strong&gt; "It worked when I tried it" is not "it handles empty input, timeouts, and two users at once."&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Code nobody understands.&lt;/strong&gt; The most expensive outcome: production code that no human on the team actually comprehends. When it breaks at 2am, "the AI wrote it" is not a debugging strategy.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;None of these show up in the demo. All of them show up in production.&lt;/p&gt;

&lt;h2&gt;
  
  
  The mode switch
&lt;/h2&gt;

&lt;p&gt;The answer isn't to stop using AI — I ship with it constantly. It's to &lt;strong&gt;change modes when the stakes change&lt;/strong&gt;. Vibe for the prototype; engineer for the product.&lt;/p&gt;

&lt;p&gt;Graduating a vibe-coded prototype looks like this:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Write down what it must actually do&lt;/strong&gt; — the spec you skipped.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Read the code the AI wrote.&lt;/strong&gt; All of it. Understand it well enough to defend it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Test the edge cases&lt;/strong&gt;, not just the happy path.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Check the security-sensitive paths&lt;/strong&gt; — auth, input handling, anything touching money or data.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Verify the built artifact&lt;/strong&gt;, because &lt;a href="https://ksatriabintangsamudra.com/blog/auditing-ai-written-code/" rel="noopener noreferrer"&gt;what ships isn't always what the model said it produced&lt;/a&gt;.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;You keep the speed of AI and add the rigor of engineering. That's the whole trick.&lt;/p&gt;

&lt;h2&gt;
  
  
  The honest take
&lt;/h2&gt;

&lt;p&gt;Vibe coding isn't the enemy — unread code in production is. Let the AI move fast where being wrong is cheap, and slow down to actually understand things where being wrong is expensive. The developers who win with AI aren't the ones who vibe the hardest; they're the ones who know exactly when to stop vibing and start engineering.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>programming</category>
      <category>productivity</category>
      <category>webdev</category>
    </item>
    <item>
      <title>I wrote Indonesian payroll twice on purpose</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Fri, 06 Nov 2026 12:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/i-wrote-indonesian-payroll-twice-on-purpose-4nm0</link>
      <guid>https://dev.to/kstriabintang/i-wrote-indonesian-payroll-twice-on-purpose-4nm0</guid>
      <description>&lt;p&gt;"Don't repeat yourself" is good advice right up until the moment a wrong number costs someone their correct wage. CoalTrack calculates payroll for coal-mining crews, and I implemented the same statutory logic &lt;strong&gt;twice&lt;/strong&gt; — once in a Laravel/PHP service, once in a Dart mirror on the phone. On purpose. Here's the reasoning.&lt;/p&gt;

&lt;h2&gt;
  
  
  The rules are law, not preference
&lt;/h2&gt;

&lt;p&gt;Indonesian payroll isn't a formula you invent. It's legislation:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;PPh21&lt;/strong&gt; income tax under the TER scheme (PMK 168/2023),&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;BPJS&lt;/strong&gt; health and employment contributions,&lt;/li&gt;
&lt;li&gt;overtime under &lt;strong&gt;Kepmenaker 102/2004&lt;/strong&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Each has brackets, caps and rounding rules that have to be applied in a specific order. Get the order wrong and the total is wrong in a way that's hard to spot and easy to dispute.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why two implementations, not one
&lt;/h2&gt;

&lt;p&gt;A payroll figure shows up in two places: computed authoritatively on the server, and previewed instantly on the worker's handset — including when the handset is offline down a mine access road. I could have made the phone a dumb display that waits for the server, but a preview that's blank without signal is useless exactly when people want it.&lt;/p&gt;

&lt;p&gt;So the phone computes too. And the moment you have two implementations of a legal calculation, you have a new risk: &lt;strong&gt;they can disagree.&lt;/strong&gt; That risk is also the safety net. If the PHP service and the Dart mirror ever produce different numbers for the same inputs, something is wrong — and I want to know at build time, not at payday.&lt;/p&gt;

&lt;h2&gt;
  
  
  Mirrored tests are the contract
&lt;/h2&gt;

&lt;p&gt;The two implementations are held together by &lt;strong&gt;mirrored test suites&lt;/strong&gt;: the same input fixtures, the same expected outputs, run against both languages. The tests are the real specification. When the tax rule changes, I change the fixtures first, watch both sides fail, then fix both until they agree. Neither implementation is allowed to be "close enough."&lt;/p&gt;

&lt;p&gt;This is also why the numbers can be trusted after the fact. Attendance events and approvals live behind an &lt;strong&gt;append-only audit trail&lt;/strong&gt; — PostgreSQL triggers reject &lt;code&gt;UPDATE&lt;/code&gt; and &lt;code&gt;DELETE&lt;/code&gt; on those tables — so the inputs to payroll can't be quietly rewritten later.&lt;/p&gt;

&lt;h2&gt;
  
  
  When duplication is the right call
&lt;/h2&gt;

&lt;p&gt;Duplication is a cost, and most of the time DRY is right. But when a calculation is (a) defined by an external authority, (b) needed in two runtimes for real reasons, and (c) expensive to get wrong, a second implementation checked against the first isn't waste — it's a proof. I'd take a redundant, cross-verified payroll engine over a single elegant one every single payday.&lt;/p&gt;

</description>
      <category>fintech</category>
      <category>payments</category>
      <category>security</category>
      <category>programming</category>
    </item>
    <item>
      <title>Access control belongs in the database, not the UI</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Fri, 06 Nov 2026 05:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/access-control-belongs-in-the-database-not-the-ui-al3</link>
      <guid>https://dev.to/kstriabintang/access-control-belongs-in-the-database-not-the-ui-al3</guid>
      <description>&lt;p&gt;The most common access-control bug I see isn't exotic. It's this: the app hides a button, and the developer calls it done. But the data is still one crafted request away. If authorization only lives in the UI, it doesn't really exist.&lt;/p&gt;

&lt;p&gt;So I push access control down to where the data actually is — the database — using Postgres &lt;strong&gt;row-level security (RLS)&lt;/strong&gt;. I've built this into a campus platform (Cendekia), an ISP operations portal, and a mining workforce app, each with several roles that must only ever see their own slice.&lt;/p&gt;

&lt;h2&gt;
  
  
  What RLS actually does
&lt;/h2&gt;

&lt;p&gt;With RLS on, a table doesn't return rows a user isn't allowed to see — even if the query asks for them. The policy is attached to the table, so it holds no matter which screen, endpoint, or ad-hoc query does the asking. A student sees their own grades; a lecturer sees their classes; an admin sees the registry — and the rule is enforced by Postgres, not by remembering to add a &lt;code&gt;WHERE&lt;/code&gt; clause in every handler.&lt;/p&gt;

&lt;p&gt;That's the key shift: &lt;strong&gt;authorization becomes a property of the data, not a habit of the code.&lt;/strong&gt; New endpoints inherit it for free. A forgotten filter can't leak anything, because the database refuses.&lt;/p&gt;

&lt;h2&gt;
  
  
  Two lessons from doing it for real
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;RLS has a performance cost — design for it.&lt;/strong&gt; A policy is evaluated per row. On one production system a Supabase slowdown traced straight to missing indexes and per-row policy evaluation; I fixed it with a handful of well-chosen indexes and a policy rewrite. RLS is not free, but it is optimisable — treat the policy like the hot path it is.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Roles and regions compose.&lt;/strong&gt; Real orgs aren't just "admin vs user." Access is role &lt;em&gt;and&lt;/em&gt; scope (which region, which class, which site). Modelling that in the policy — rather than in scattered app logic — keeps it consistent as the app grows.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Then make the record refuse to lie
&lt;/h2&gt;

&lt;p&gt;Access control decides who can read and write. The other half is making sure what was written can't be quietly rewritten later. For attendance and approvals in the mining app, the audit trail is &lt;strong&gt;append-only at the database level&lt;/strong&gt;: Postgres triggers reject &lt;code&gt;UPDATE&lt;/code&gt; and &lt;code&gt;DELETE&lt;/code&gt; on those tables. The app can add facts; it cannot edit history. When payroll or a dispute depends on those records, "trust me, nobody changed it" isn't good enough — the database guarantees it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The principle
&lt;/h2&gt;

&lt;p&gt;Hiding a control is UX. Enforcing access in the database is security. Do the second one, index it properly, and back it with records the application isn't allowed to alter — then the guarantee holds no matter what the front end does.&lt;/p&gt;

</description>
      <category>database</category>
      <category>security</category>
      <category>backend</category>
      <category>programming</category>
    </item>
    <item>
      <title>Trojans: the malware you invite in, and how the FBI uninstalled it</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Fri, 06 Nov 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/trojans-the-malware-you-invite-in-and-how-the-fbi-uninstalled-it-3end</link>
      <guid>https://dev.to/kstriabintang/trojans-the-malware-you-invite-in-and-how-the-fbi-uninstalled-it-3end</guid>
      <description>&lt;p&gt;The original Trojan Horse did not knock down the walls of Troy. The Trojans &lt;strong&gt;pulled it inside themselves&lt;/strong&gt;, because it looked like a gift.&lt;/p&gt;

&lt;p&gt;Computer trojans work the same way, and that is why they are still one of the most successful kinds of malware ever made. They do not need a fancy exploit. They need you to click.&lt;/p&gt;

&lt;h2&gt;
  
  
  What a trojan actually is
&lt;/h2&gt;

&lt;p&gt;A trojan is malware wearing a costume. It might look like:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;an invoice in a Word document that asks you to "enable content",&lt;/li&gt;
&lt;li&gt;a cracked game or a free version of paid software,&lt;/li&gt;
&lt;li&gt;a browser update, a PDF reader, a "wallet tool" for crypto,&lt;/li&gt;
&lt;li&gt;a job test or a repo from a friendly recruiter.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Once you run it, the real payload starts: stealing passwords, logging keystrokes, giving a remote attacker control, or downloading more malware. Some are called &lt;strong&gt;RATs&lt;/strong&gt;, remote access trojans, because they quietly give someone else the keyboard.&lt;/p&gt;

&lt;p&gt;Quick vocabulary check: a &lt;strong&gt;virus&lt;/strong&gt; infects other files, a &lt;strong&gt;worm&lt;/strong&gt; spreads by itself, and a &lt;strong&gt;trojan&lt;/strong&gt; gets you to install it. I go deeper into that family tree in &lt;a href="https://ksatriabintangsamudra.com/blog/history-of-computer-viruses/" rel="noopener noreferrer"&gt;the history of computer viruses&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  The banking trojans
&lt;/h2&gt;

&lt;p&gt;Around 2007, &lt;strong&gt;Zeus&lt;/strong&gt; became the template. It sat inside your browser and stole online banking logins, sometimes changing the page you saw so you would type in extra details. Infected machines were linked into a &lt;strong&gt;botnet&lt;/strong&gt;, a remote-controlled army of computers. Variants of Zeus, like GameOver Zeus, are why one of its alleged operators sits on the &lt;a href="https://ksatriabintangsamudra.com/blog/fbi-cyber-most-wanted/" rel="noopener noreferrer"&gt;FBI's Cyber Most Wanted list&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Emotet: from trojan to cybercrime delivery service
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Emotet&lt;/strong&gt; appeared in 2014 as a banking trojan. Then its operators had a business idea: why steal bank logins yourself when you can &lt;strong&gt;rent out access&lt;/strong&gt; to every computer you have infected?&lt;/p&gt;

&lt;p&gt;Emotet became a &lt;strong&gt;loader&lt;/strong&gt;. It spread through convincing reply-chain emails, often continuing real conversations stolen from earlier victims. Then it sold its foothold to other criminals, including ransomware gangs. Europol called it one of the most dangerous pieces of malware in the world.&lt;/p&gt;

&lt;h2&gt;
  
  
  The takedowns: police using the malware against itself
&lt;/h2&gt;

&lt;p&gt;Here is the twist that makes these stories so satisfying.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Emotet, January 2021.&lt;/strong&gt; An international operation coordinated by Europol and Eurojust took control of Emotet's infrastructure from the inside. Investigators pushed an update to infected machines that disconnected them from the criminals, and the malware was scheduled to remove itself in April 2021. Emotet later came back in a weaker form, but the hit was massive.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Qakbot, August 2023.&lt;/strong&gt; The FBI led &lt;strong&gt;Operation Duck Hunt&lt;/strong&gt; against Qakbot, another trojan turned loader that had fed ransomware groups for years. The FBI redirected infected computers to servers it controlled, then sent them an uninstaller. More than &lt;strong&gt;700,000 computers&lt;/strong&gt; were cleaned, and around &lt;strong&gt;$8.6 million&lt;/strong&gt; in cryptocurrency was seized.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Operation Endgame, May 2024.&lt;/strong&gt; Europol called it the largest ever operation against botnets, hitting several loaders at once, including IcedID, SmokeLoader, Pikabot and Bumblebee.&lt;/p&gt;

&lt;p&gt;Think about how wild that is: the delivery channel criminals built to install malware became the channel police used to uninstall it.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to not be the one who opens the horse
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Never "enable macros" on a document you did not expect.&lt;/strong&gt; That single button has started thousands of incidents.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Download software only from official sources.&lt;/strong&gt; Cracks and "free" premium tools are the most common trojan costumes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Treat unexpected attachments in old email threads as suspicious.&lt;/strong&gt; Reply-chain hijacking makes them look familiar on purpose.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Run untrusted code in a sandbox or VM.&lt;/strong&gt; Especially "take-home tests" from strangers. North Korean teams love that trick.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep endpoint protection on and updated,&lt;/strong&gt; and give users non-admin accounts so a trojan cannot do everything.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The honest summary
&lt;/h2&gt;

&lt;p&gt;Trojans win because they attack the one thing no patch can fix: our habit of trusting things that look normal. The good news is that the same infrastructure that makes them powerful can also make them fragile. Take over the command server, and an army of infected machines can be told to go home.&lt;/p&gt;

</description>
      <category>malware</category>
      <category>cybersecurity</category>
      <category>botnet</category>
      <category>security</category>
    </item>
    <item>
      <title>How police hacked LockBit, the world's biggest ransomware gang</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Thu, 05 Nov 2026 12:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/how-police-hacked-lockbit-the-worlds-biggest-ransomware-gang-3gh5</link>
      <guid>https://dev.to/kstriabintang/how-police-hacked-lockbit-the-worlds-biggest-ransomware-gang-3gh5</guid>
      <description>&lt;p&gt;Ransomware gangs love a leak site. It is their shop window: a dark-web page listing victims, a countdown clock, and a threat to publish stolen data unless the ransom is paid.&lt;/p&gt;

&lt;p&gt;In February 2024, visitors to &lt;strong&gt;LockBit's&lt;/strong&gt; leak site found something new on the page. The logos of the UK's &lt;strong&gt;National Crime Agency&lt;/strong&gt;, the &lt;strong&gt;FBI&lt;/strong&gt; and a long list of international police forces. The site had been seized. And the police were just getting started.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who LockBit was
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;LockBit&lt;/strong&gt; appeared around 2019 and grew into the most prolific ransomware operation in the world. The US Justice Department later said it had attacked more than &lt;strong&gt;2,000 victims&lt;/strong&gt; and received more than &lt;strong&gt;$120 million&lt;/strong&gt; in ransom payments.&lt;/p&gt;

&lt;p&gt;It worked like a franchise, a model called &lt;strong&gt;ransomware-as-a-service&lt;/strong&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The &lt;strong&gt;core team&lt;/strong&gt; built and maintained the ransomware, the negotiation panel and the leak site.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Affiliates&lt;/strong&gt;, independent criminals, broke into companies and deployed it.&lt;/li&gt;
&lt;li&gt;Ransoms were split, with the affiliate taking the larger share.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It was a business with support staff, software updates, even a "bug bounty" for its own malware. The leader used the handle &lt;strong&gt;LockBitSupp&lt;/strong&gt; and was so confident in his anonymity that he publicly offered a &lt;strong&gt;$10 million&lt;/strong&gt; reward to anyone who could reveal his identity.&lt;/p&gt;

&lt;p&gt;That turned out to be a mistake.&lt;/p&gt;

&lt;h2&gt;
  
  
  Operation Cronos
&lt;/h2&gt;

&lt;p&gt;On &lt;strong&gt;19 and 20 February 2024&lt;/strong&gt;, an operation led by the NCA, working with the FBI, Europol and partners in around ten countries, hit LockBit at the core. They:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;took control of LockBit's &lt;strong&gt;administration environment&lt;/strong&gt; and its &lt;strong&gt;leak site&lt;/strong&gt;,&lt;/li&gt;
&lt;li&gt;seized servers and cryptocurrency accounts,&lt;/li&gt;
&lt;li&gt;obtained &lt;strong&gt;decryption keys&lt;/strong&gt;, around a thousand at first, to help victims recover without paying,&lt;/li&gt;
&lt;li&gt;and gathered data on affiliates, the people actually doing the attacks.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;But the most memorable part was the &lt;strong&gt;trolling&lt;/strong&gt;. Instead of just showing a seizure banner, police rebuilt the leak site in LockBit's own style. They posted countdown clocks, this time for revelations about the gang. They published details about the operation, mocked LockBitSupp, and announced the identity reveal like a movie trailer.&lt;/p&gt;

&lt;p&gt;This was psychological warfare. Ransomware runs on trust between criminals. Affiliates suddenly had to wonder: what does the NCA know about me?&lt;/p&gt;

&lt;h2&gt;
  
  
  The deleted data that was never deleted
&lt;/h2&gt;

&lt;p&gt;One discovery matters for every company that has ever considered paying a ransom.&lt;/p&gt;

&lt;p&gt;LockBit's model promised that if you paid, your stolen data would be deleted. Investigators found that LockBit &lt;strong&gt;kept data from victims who had paid&lt;/strong&gt;. The "deletion" was a lie.&lt;/p&gt;

&lt;p&gt;So when someone says "paying is the fastest way to make the problem go away", remember Cronos. Paying buys a criminal's promise, and those promises are worth exactly what you would expect.&lt;/p&gt;

&lt;h2&gt;
  
  
  The unmasking
&lt;/h2&gt;

&lt;p&gt;In &lt;strong&gt;May 2024&lt;/strong&gt;, the UK, US and Australia named LockBitSupp as &lt;strong&gt;Dmitry Khoroshev&lt;/strong&gt;, a Russian national. He was sanctioned, charged in the US, and the US offered up to &lt;strong&gt;$10 million&lt;/strong&gt; for information leading to his arrest. He joined the &lt;a href="https://ksatriabintangsamudra.com/blog/fbi-cyber-most-wanted/" rel="noopener noreferrer"&gt;FBI's Cyber Most Wanted list&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;He offered $10 million for his own identity. The police revealed it for free.&lt;/p&gt;

&lt;h2&gt;
  
  
  Did it kill LockBit?
&lt;/h2&gt;

&lt;p&gt;Not completely. LockBit tried to come back with new infrastructure and loud messages. But its reputation took a hit it never really recovered from. Affiliates moved to other groups, and the brand that had dominated ransomware turned into a cautionary tale.&lt;/p&gt;

&lt;p&gt;That is the realistic win in cybercrime: not a final arrest, but breaking the business model.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I take from Cronos as a builder
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Assume ransomware is a business looking for easy customers.&lt;/strong&gt; Most affiliates get in through stolen VPN passwords, unpatched edge devices and exposed remote desktop. Close those, and you fall off the target list.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Backups are your real negotiation power.&lt;/strong&gt; If you can restore, the ransom note is just an incident, not a hostage situation.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Report to the authorities.&lt;/strong&gt; Operations like Cronos depend on victim reports, and they can bring back decryption keys.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Never trust the "we deleted it" promise.&lt;/strong&gt; If data was stolen, plan as if it is out there.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The honest summary
&lt;/h2&gt;

&lt;p&gt;Operation Cronos showed that even the biggest ransomware gang is just a website, a few servers and a lot of fragile trust. Take those over, and the "untouchable" hackers start to look very touchable. Our side of the deal is to make sure they never get the first foothold.&lt;/p&gt;

</description>
      <category>ransomware</category>
      <category>cybersecurity</category>
      <category>lockbit</category>
      <category>security</category>
    </item>
    <item>
      <title>Inside the FBI's Cyber Most Wanted list</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Thu, 05 Nov 2026 05:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/inside-the-fbis-cyber-most-wanted-list-1377</link>
      <guid>https://dev.to/kstriabintang/inside-the-fbis-cyber-most-wanted-list-1377</guid>
      <description>&lt;p&gt;The FBI has a "Ten Most Wanted Fugitives" list that most people know from movies. Fewer people know it also keeps a list just for hackers: &lt;strong&gt;Cyber's Most Wanted&lt;/strong&gt;. Reading it is like reading the credits of every big cybercrime story of the last fifteen years.&lt;/p&gt;

&lt;p&gt;I go through it now and then, not because I expect to meet anyone on it, but because every name is a lesson in how real attacks are built.&lt;/p&gt;

&lt;h2&gt;
  
  
  The banking-malware godfather
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Evgeniy Bogachev&lt;/strong&gt; is probably the most famous face on the list. US authorities say he ran &lt;strong&gt;GameOver Zeus&lt;/strong&gt;, a network of infected computers used to steal banking credentials, and helped spread &lt;strong&gt;CryptoLocker&lt;/strong&gt;, one of the early ransomware hits. Losses were estimated at more than $100 million.&lt;/p&gt;

&lt;p&gt;The US offered up to &lt;strong&gt;$3 million&lt;/strong&gt; for information leading to his arrest, at the time one of the biggest rewards ever for a cyber suspect. He is believed to live in Russia, and he has never been arrested.&lt;/p&gt;

&lt;h2&gt;
  
  
  The guy with the "thief" plate
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Maksim Yakubets&lt;/strong&gt; is accused of leading &lt;strong&gt;Evil Corp&lt;/strong&gt;, the group behind the &lt;strong&gt;Dridex&lt;/strong&gt; banking malware. When the US charged him in 2019, it offered up to &lt;strong&gt;$5 million&lt;/strong&gt; for information about him.&lt;/p&gt;

&lt;p&gt;What made the story famous were the videos: a customised Lamborghini, reportedly with a number plate reading "thief" in Russian. Cybercrime as a lifestyle brand. The charges and sanctions made that lifestyle a lot harder to spend.&lt;/p&gt;

&lt;h2&gt;
  
  
  The state hacker
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Park Jin Hyok&lt;/strong&gt; is different. He is not a freelancer. The US Justice Department charged him in 2018 as a programmer working for a North Korean government front company. The indictment ties him to the &lt;strong&gt;Sony Pictures&lt;/strong&gt; hack, the &lt;a href="https://ksatriabintangsamudra.com/blog/bangladesh-bank-heist-typo/" rel="noopener noreferrer"&gt;Bangladesh Bank heist&lt;/a&gt; and &lt;a href="https://ksatriabintangsamudra.com/blog/wannacry-kill-switch/" rel="noopener noreferrer"&gt;WannaCry&lt;/a&gt;. In other words, to &lt;a href="https://ksatriabintangsamudra.com/blog/lazarus-group-hackers-who-fund-a-nation/" rel="noopener noreferrer"&gt;Lazarus Group&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;He will almost certainly never stand in a US courtroom. So why bother charging him?&lt;/p&gt;

&lt;h2&gt;
  
  
  The ransomware bosses
&lt;/h2&gt;

&lt;p&gt;The modern list is full of ransomware names. &lt;strong&gt;Mikhail Matveev&lt;/strong&gt; was charged in 2023 over attacks using LockBit, Babuk and Hive, with a reward of up to &lt;strong&gt;$10 million&lt;/strong&gt;. &lt;strong&gt;Dmitry Khoroshev&lt;/strong&gt;, named as the leader of LockBit in 2024, carries the same reward. I covered how police got inside his operation in &lt;a href="https://ksatriabintangsamudra.com/blog/lockbit-operation-cronos/" rel="noopener noreferrer"&gt;Operation Cronos&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why the list works anyway
&lt;/h2&gt;

&lt;p&gt;It is easy to laugh at a "most wanted" list where most people live out of reach. But I think it works, in three quiet ways.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Travel becomes a gamble.&lt;/strong&gt; An indictment follows you forever. &lt;strong&gt;Roman Seleznev&lt;/strong&gt;, a Russian carder, was arrested in the &lt;strong&gt;Maldives&lt;/strong&gt; in 2014 while on holiday, flown to the US and later sentenced to &lt;strong&gt;27 years&lt;/strong&gt;. Every beach trip is now a risk calculation.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Money gets frozen.&lt;/strong&gt; Charges often come with sanctions. Exchanges, banks and partners stop touching your wallets. Stolen money is only useful if you can spend it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Trust breaks.&lt;/strong&gt; Cybercrime runs on reputation. Once a leader is unmasked, affiliates start to wonder what else the police know, and who is talking.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I take from it as a builder
&lt;/h2&gt;

&lt;p&gt;None of these people broke into systems with movie magic. Banking trojans stole saved passwords. Ransomware crews bought access from other criminals. A state team sent convincing emails. The defences are still the unglamorous ones: MFA everywhere, patched systems, least privilege and backups you have actually tested.&lt;/p&gt;

&lt;h2&gt;
  
  
  The honest summary
&lt;/h2&gt;

&lt;p&gt;The Cyber's Most Wanted list is less a list of geniuses and more a list of business models. Each name represents a way of turning weak security into money. Close those gaps, and you are simply not worth their time.&lt;/p&gt;

</description>
      <category>cybersecurity</category>
      <category>cybercrime</category>
      <category>security</category>
      <category>fbi</category>
    </item>
    <item>
      <title>The $25 million video call where everyone was a deepfake</title>
      <dc:creator>Ksatria Bintang Samudra</dc:creator>
      <pubDate>Thu, 05 Nov 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/kstriabintang/the-25-million-video-call-where-everyone-was-a-deepfake-3kfb</link>
      <guid>https://dev.to/kstriabintang/the-25-million-video-call-where-everyone-was-a-deepfake-3kfb</guid>
      <description>&lt;p&gt;For years the advice for spotting a scam was simple: if you are not sure, get them on a video call. If you can see and hear the person, it is probably real.&lt;/p&gt;

&lt;p&gt;In 2024 that advice died in a meeting room in Hong Kong.&lt;/p&gt;

&lt;h2&gt;
  
  
  What happened
&lt;/h2&gt;

&lt;p&gt;In &lt;strong&gt;January 2024&lt;/strong&gt;, a finance employee at the Hong Kong office of &lt;strong&gt;Arup&lt;/strong&gt;, the British engineering firm behind landmarks like the Sydney Opera House, received a message that appeared to come from the company's &lt;strong&gt;UK-based chief financial officer&lt;/strong&gt;. It talked about a confidential transaction.&lt;/p&gt;

&lt;p&gt;The employee was suspicious. Good instinct. So they joined a &lt;strong&gt;video call&lt;/strong&gt; to check.&lt;/p&gt;

&lt;p&gt;On the call: the CFO, plus several colleagues they recognised. Faces, voices, mannerisms. Everything looked right. The instructions were clear, so the employee followed them.&lt;/p&gt;

&lt;p&gt;Over the following days they made &lt;strong&gt;15 transfers&lt;/strong&gt; to &lt;strong&gt;five bank accounts&lt;/strong&gt;, totalling &lt;strong&gt;HK$200 million&lt;/strong&gt;, around &lt;strong&gt;US$25 million&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Every other person on that call was an &lt;strong&gt;AI-generated deepfake&lt;/strong&gt;. The fraud was only discovered when the employee later checked with head office. Hong Kong police said it was the first time they had seen a fake multi-person video conference used this way, and Arup confirmed it was the victim in May 2024.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why it worked
&lt;/h2&gt;

&lt;p&gt;What I find brilliant, in a terrible way, is that the attackers did not hack Arup's systems at all. No malware. No stolen passwords. Arup itself later said its systems were not compromised. They hacked &lt;strong&gt;trust&lt;/strong&gt;.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Authority.&lt;/strong&gt; The request came from the CFO. Most employees do not argue with the CFO.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Social proof.&lt;/strong&gt; It was not one fake person, it was a whole meeting. If everyone else on the call acts normal, you assume it is normal.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Secrecy and urgency.&lt;/strong&gt; A "confidential" deal gives a reason not to ask around.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Public raw material.&lt;/strong&gt; Executives appear in videos, interviews and conference talks. That footage is the training data for their clone.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This is classic &lt;a href="https://ksatriabintangsamudra.com/blog/ai-orchestrated-cyberattack/" rel="noopener noreferrer"&gt;social engineering&lt;/a&gt;, just upgraded with AI. The playbook is decades old. The special effects are new.&lt;/p&gt;

&lt;h2&gt;
  
  
  It is not the first, and it will not be the last
&lt;/h2&gt;

&lt;p&gt;Back in 2019, the Wall Street Journal reported a UK energy company being tricked into sending around €220,000 after a call where AI imitated the voice of its parent company's boss. That was voice only. Five years later it was a full video meeting.&lt;/p&gt;

&lt;p&gt;Today, voice-cloning tools need only a short sample. Real-time face-swap software runs on a gaming laptop. The cost of faking a person has collapsed.&lt;/p&gt;

&lt;h2&gt;
  
  
  How I would defend against this
&lt;/h2&gt;

&lt;p&gt;Here is my honest take: &lt;strong&gt;do not rely on spotting the fake&lt;/strong&gt;. Detection tools help, but the tells keep disappearing. Design the process so that even a perfect fake cannot move money.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Out-of-band verification.&lt;/strong&gt; If someone asks for money on a call or chat, hang up and call them back on a number you already had. Not the one in the message.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Two-person rule.&lt;/strong&gt; No single employee should be able to send a large or unusual payment alone, no matter who asked.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Verification phrases.&lt;/strong&gt; Some teams and families now agree on a private code word for emergencies. Low tech, very effective.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Treat secrecy plus urgency as a red flag.&lt;/strong&gt; Real CFOs can wait ten minutes for a callback.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Limit the public footage of key people&lt;/strong&gt; where it makes sense, and teach executives that their voice is now a security asset.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The honest summary
&lt;/h2&gt;

&lt;p&gt;The Arup scam is the moment "I saw them with my own eyes" stopped being proof. The attackers did not need to break any system, because the human approval process was the system. Make that process require more than one channel and more than one person, and a deepfake becomes just an expensive video.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>deepfake</category>
      <category>cybersecurity</category>
      <category>fraud</category>
    </item>
  </channel>
</rss>
