DEV Community: Ahtisham Lala

Binance's $6.3 Billion Hack Started With a Job Offer — The Post-Mortem Is Out, and No Private Key Ever Leaked

Ahtisham Lala — Thu, 05 Nov 2026 19:35:31 +0000

Originally published on ConvertElementor: https://convertelementor.com/blog/binance-s-6-3-billion-hack-started-with-a-job-offer-the-post-mortem-is-out-and-no-private-key-ever-leaked-november-2026-20261105

Binance published its post-mortem on November 3, 2026, for the largest crypto theft ever recorded: about $6.3 billion drained from its hot wallets in roughly 20 minutes on October 29. No private key leaked. A job offer, a DevOps engineer's laptop, and an internal AI tool did the job — and the real lesson is about the approval layer, not the vault.

What actually happened on October 29

On October 29, 2026, attackers drained about $6.3 billion from Binance's hot wallets. At 20:05 UTC, five Binance hot wallets on the XRP Ledger sent 233.46 million XRP to the attackers. Binance suspended withdrawals that night. It did not matter — the money kept leaving anyway, because the attackers were inside the approval path, not breaking cryptography.

It started in June — with a job offer

Binance's post-mortem traces the story back to a fake job offer — the classic opening move of North Korea's TraderTraitor unit, which the FBI attributed the theft to on October 31, 2026. The laptop that got compromised belonged to a DevOps engineer. That detail matters more than almost any other: the attacker logged onto Engineer C's computer between 17:15 and 19:56 UTC on October 29, and from there reached the systems that could move funds.

The EIP-7702 trick, in plain English

The technical heart of this hack is a feature called EIP-7702, an Ethereum upgrade that lets an ordinary wallet temporarily delegate its powers to a smart contract. There is a cruel irony here: Binance's hot wallet used multi-party computation (MPC) — a setup designed so no single person holds a full private key. No key leaked, yet the attackers could still authorise transfers, because they controlled the session and the delegation, not the key itself.

What got frozen, what didn't, and who pays

About $2.8 to $3 billion of the stolen funds were frozen or intercepted, with help from Tether, Circle and other partners. Binance set aside $6.34 billion and began reopening withdrawals in phases on November 1. In the first nine and a half hours after withdrawals reopened, Binance processed $4.649 billion in withdrawals. Binance says no users will lose money, and about 95% of the impacted amount has been covered.

Why "no private key leaked" is the scary part

It is tempting to read "no keys leaked" as good news — the cryptography held. It is actually the warning: if an attacker can move $6.3 billion without ever touching a private key, then key management alone is not the security boundary. The boundary is who can approve a high-value action, from which device, to which destination, and how often.

What this means for anyone running infrastructure

You do not run a crypto exchange, and neither do I. But the pattern shows up in ordinary systems: service tokens need the same discipline as user credentials; high-value actions need a second approval, a destination allowlist, and a rate limit; and you should audit your AI tools' access. If your team uses AI coding assistants, internal chatbots, or automation with admin access, that access is now part of your attack surface.

What I would do this week

First, list every laptop, token, and integration with deployment or admin access. If the list surprises you, that is the finding. Then tighten the approval layer: second approval for high-value actions, allowlisted destinations, and rate limits — before you spend more on the vault.

Read the full article on ConvertElementor: https://convertelementor.com/blog/binance-s-6-3-billion-hack-started-with-a-job-offer-the-post-mortem-is-out-and-no-private-key-ever-leaked-november-2026-20261105

Moonshot Just Put a 2.8-Trillion-Parameter Model on Hugging Face — Kimi K3.1 Pro's Open Weights Are Real, and the Fine Print Matters

Ahtisham Lala — Thu, 05 Nov 2026 19:28:42 +0000

Originally published on ConvertElementor: https://convertelementor.com/blog/moonshot-just-put-a-2-8-trillion-parameter-model-on-hugging-face-kimi-k3-1-pro-s-open-weights-are-real-and-the-fine-print-matters-november-2026-20261105

Moonshot AI published the full weights of Kimi K3.1 Pro on Hugging Face on November 4, 2026 — 2.8 trillion parameters, a 1-million-token context window, and a license that is genuinely open for most of us but not for everyone.

What actually landed

Moonshot posted the weights on Hugging Face under the repository name moonshotai/Kimi-K3.1-Pro — no access gate, no form, no waitlist. The download is 96 safetensors shards totalling about 1.56 terabytes.

Kimi K3.1 Pro itself is not brand new. Moonshot announced it on October 23, 2026, as the first model in the K3.1 family. What is new is that the full weights are now public.

The numbers

The headline figure is 2.8 trillion total parameters, which makes K3.1 Pro the largest publicly downloadable model. It is a mixture-of-experts model: it has 896 experts, and only 16 of them activate for any given token, so 104 billion parameters are active at a time.

The context window is 1 million tokens — 1,048,576, to be exact. In 24 of the model's 93 layers, the usual attention mechanism is replaced by a different design, which is part of how it handles that long context.

The license: open for most of us, with a line drawn at the top

"Open weights" and "open source" are not the same thing. The weights ship under Moonshot's own Kimi K3.1-Pro License, not a standard open-source license like Apache 2.0. For independent developers, small teams, and most agencies, that licence is permissive enough to use and build on. Read it before you build a product on it, especially at larger scale.

The benchmarks

Moonshot's own numbers put K3.1 Pro near the top of the field. On Terminal-Bench 4.0, a coding-agent benchmark, it performed strongly. As with every vendor benchmark table, treat those numbers as a starting point, not proof — they measure what an agent can do, which is the result that matters most for developers.

What it takes to run: this is not a laptop model

The download button is the easy part. The weights are about 1.56 TB, and running them locally needs serious GPU hardware. The hosted price is genuinely competitive: $2 per million input tokens and $10 per million output tokens.

What this means for WordPress and Elementor builders

The single most useful thing about a model like K3.1 Pro for a WordPress developer is the combination of long context and strong coding ability. A 1-million-token context means a long-context code audit, or a debugging session across a whole theme or plugin codebase, is realistic in one conversation.

The open weights matter for a second reason specific to WordPress: sites that handle client data can, in principle, run capable models on infrastructure they control, rather than sending everything to a hosted API.

What I would do this week

First, do not download 1.56 terabytes on a whim. Try it hosted first — on Kimi.ai, through the API, or through a hosted provider.

Second, if you run a WordPress agency, look at the long-context coding angle before anything else. Try it on a real task: a long-context code audit, or a debugging session across a whole theme.

Third, read the license before you build a product on it. For personal use, small teams, and most agencies, the terms are workable — but check them yourself.

Kimi K3.1 Pro's weights release is the rare AI story this week where the headline and the reality line up — with fine print worth reading.

Read the full article on ConvertElementor: https://convertelementor.com/blog/moonshot-just-put-a-2-8-trillion-parameter-model-on-hugging-face-kimi-k3-1-pro-s-open-weights-are-real-and-the-fine-print-matters-november-2026-20261105

Run Your WordPress Site From ChatGPT or Claude — Free Until 30 November 2026

Ahtisham Lala — Thu, 05 Nov 2026 19:25:58 +0000

Originally published on ConvertElementor: https://convertelementor.com/blog/run-your-wordpress-site-from-chatgpt-or-claude-free-until-30-november-2026-20261104

WordPress.com and Jetpack are letting you connect ChatGPT or Claude to your site for free through November 2026 — no paid plan needed. This is not another AI writing plugin inside your dashboard. It is the reverse: your ChatGPT or Claude chat, connected to your site, able to read it and change it while you type in plain language.

What is actually free, and until when

On WordPress.com, connecting an external AI agent — ChatGPT, Claude, or any agent that supports MCP — is included with paid plans, and supported on the free plan through November 2026. Jetpack says the same for self-hosted sites connected to WordPress.com: until 30 November 2026, connecting ChatGPT or Claude to your Jetpack site is free, and you do not need a paid Jetpack plan.

Your ChatGPT or Claude subscription is separate. The free window covers the WordPress side of the connection, not your AI app's plan. After 30 November, WordPress and Jetpack say their pricing pages will show which plans include AI agents. Everything the AI made for you stays on your site either way. Nothing gets deleted when the window closes; you just may need a paid plan to keep the connection live.

How the connection works

The technology underneath is MCP, the Model Context Protocol — an open standard that lets an AI app talk to other software in a structured way. When you connect, the AI can see your site's posts, pages, media, and settings through declared tools, and it can only do what the connected WordPress user account is allowed to do.

For a WordPress.com site, the official route is the WordPress.com plugin for ChatGPT or the WordPress.com connector for Claude. For a self-hosted site, the route is Jetpack: install Jetpack free, connect it to WordPress.com, and then add the same WordPress.com app in ChatGPT or Claude. Coding agents work too — Claude Code, Cursor, VS Code, and Codex — with a one-line setup command that Jetpack publishes.

Connecting it

First, make sure the site is running Jetpack and connected to WordPress.com. Second, in ChatGPT add the WordPress.com app, and in Claude add the WordPress.com connector. Both ask you to sign in with the WordPress.com account linked to your site and approve the connection. At this step you choose the level of access — read only, or full access. Start with read only. You can widen access later, once you have seen what the agent does with read access.

Third, ask it something small and verifiable: list my five most recent posts and tell me which ones have no featured image. That tests whether the connection actually sees your real site, and the answer is something you can check in thirty seconds in your dashboard.

What actually worked

Drafting and editing posts worked best. Draft a post from a short brief, give it a category and tags, and save it as a draft — not publish it. Fixing a typo and tightening a paragraph on an existing page also worked cleanly. The useful pattern is: let the agent do the first pass, and keep publish in human hands.

Where it struggles

First, the agent is only as good as your instruction. "Improve my site" produces vague, confident changes you then have to undo. "Draft a 400-word post about our new service, in a plain tone, save it as a draft, do not publish" produces something usable.

Second, it does not know your taste, your brand voice, or what you would never publish. That is why I keep drafts as drafts. The agent saves time on the first 80 percent; the last 20 percent — the voice, the judgement, the decision to publish — is still yours.

It works best on standard posts and simple pages. For Elementor templates and heavily designed landing pages, let the agent prepare text and make layout changes in Elementor yourself, or test on staging first.

The safety checklist

Start with read-only access and prove the connection works. Before you switch to full access, make a backup you can actually restore. Use a dedicated user account for the connection where you can, rather than your main administrator login. An editor-level account can draft and edit content without being able to install plugins or change site settings. Never let the agent publish directly on a site that earns money or carries client work until you have reviewed several of its drafts.

What happens after 30 November

After 30 November 2026, the free access for Jetpack sites and WordPress.com free-plan sites ends, and the connection becomes a feature of the plans listed on the pricing pages. Your content does not disappear. Drafts stay drafts, published posts stay published, and images stay where they are.

If you run a WordPress site — especially if you build with Elementor and spend your days in the dashboard — connect it while it is free, start read-only, and give it one real week of small jobs.

Read the full article on ConvertElementor: https://convertelementor.com/blog/run-your-wordpress-site-from-chatgpt-or-claude-free-until-30-november-2026-20261104