The Cleus Blog
Tutorials, engineering deep-dives, and product updates on building with AI agents — research, coding, images, video and voice.
Qwen Jailbreak Prompts in 2026: Why They Fail (We Tested Where Qwen Blocks)
We sent the same requests to Qwen4-Max and to open Qwen 3.8 on two different hosts. Half were stopped by a filter no prompt can reach, and the political dodges were in the model itself.
Qwen4-Max Benchmarks and Tests: How It Performs, and Can It Be Uncensored?
Alibaba's Qwen4-Max launched on November 16 at $1.50 in and $4.50 out per million tokens. Here are the scores, what happened when we tested it ourselves, and why this Qwen can't be uncensored.
Grok Jailbreak Prompts in 2026: Why They Fail on Grok 4.7 (and What Works Instead)
Grok was the easiest big chatbot to jailbreak in 2025. xAI's own numbers for Grok 4.7 say that is over. Here is what changed, where those numbers are soft, and what people do instead.
ChatGPT Jailbreak Prompts in 2026: Why They Fail on GPT-6 (and What Works Instead)
ChatGPT moved to GPT-6 in October, and OpenAI published how it holds up against jailbreaks. Here is what a copied prompt is up against, what trying can cost your account, and what people do instead.
Gemini Jailbreak Prompts in 2026: Why They Fail on Gemini 4 (and What Works Instead)
Most Gemini jailbreak prompts were written for another model, and on Gemini 4 they argue with the one layer that is not doing the blocking. Here is how the filters work and what people do instead.
FAA Ground Stop Explained: Why Every US Flight Was Grounded and What Happens Now
For 4 hours and 21 minutes on Friday, no plane could take off anywhere in the US. Here is what failed, why officials gave two versions of the cause, and what you are owed if your flight was cancelled.
Anthropic IPO: Date, Price Range, Ticker ANT and What the Filing Really Says
Anthropic set its IPO at $70 to $78 a share for 825 million shares on Nasdaq as ANT. Here is what the prospectus says about the money, the control and the risks.
CVE-2026-124194: Outlook Zero-Day Lets One Opened File Leak Your Windows Login Hash
Microsoft says attackers are already using an Outlook for Windows bug that skips a safety warning and can hand over your sign-in hash. Here is who is affected and how to patch.
Jensen Huang Resigns as Nvidia CEO: What Happened and What It Means for AI
Nvidia's founder is out after 33 years and CFO Colette Kress is interim CEO. Here is what is confirmed, what is only alleged, and what changes for the AI companies that depend on Nvidia.
Gemini 4 Carbon Is Public: We Tested What Its Hacking Filters Block
Google sells the public Gemini 4 Carbon behind filters meant to stop offensive security work. We sent 11 ordinary prompts three times each to Carbon and Argon to find the line.
CVE-2026-125350: Chrome's Eighth Zero-Day of 2026 Is Being Exploited. Update to 157.0.8103.50
Google has patched a Chrome flaw that attackers were already using. Here is what CVE-2026-125350 does, which versions are safe, and how to check yours in under a minute.
GLM 5.3 Abliterated: How to Use GLM 5.3 Uncensored for Coding (No Setup)
GLM 5.3 is one of the strongest open coding models of 2026, and one of the quickest to refuse. Here are the real ways to run it uncensored, and the one that needs no GPUs.
How to Hack Using AI Agents (and what happens when the model stops saying no)
A hands-on field guide to agentic security work: real scan output, six non-destructive examples, and the exact places the loop still falls over.
The Best Uncensored LLMs to Run Locally in 2026, Sorted by How Much Memory You Have
The top-scoring uncensored model you can run at home fits on one 24 GB graphics card and beats models more than twice its size. Here are the best tested picks for every memory size, with download sizes and what each one gives up.
We Ran Every Fable 5.1 and Opus 5.5 Jailbreak on Fable 5.2
Six published Claude jailbreaks, five probes each, three runs per cell, on Fable 5.2 with Opus 5.5 as a control. Two never got a word out, and the reason is not the payload's length or its persona.
Claude Fable 5.2 Benchmarks: vs GPT-6 Astra and Opus 5.5, and Can It Be Uncensored?
Anthropic's Claude Fable 5.2 arrived on November 9 with a 20% price cut and a table of wins over GPT-6 Astra and Opus 5.5. Here are the scores, our own test results, and why there is no uncensored version.
Why China Is Struggling to Copy Nvidia: The 5 Walls Behind Its GPU Gap
Huawei sells more AI chips in China than Nvidia now, but China still makes about 25 times less AI computing power. Here are the five walls between Chinese GPUs and Nvidia's.
GLM-5.3 Uncensored? What GLM Refuses, the Abliterated Versions, and a Simpler Option
GLM is one of the least willing model families on the UGI leaderboard, and the abliterated GLM-5.3 uploads are huge and untested. We ran GLM-5.3 through our refusal and build tests and compared a simpler option.
Kimi K3.1 Uncensored? What Kimi Refuses, and the Closest Uncensored Alternative
Kimi K3.1 has no official uncensored version, and the community ones are huge and untested. We checked what Kimi K3.1 Pro refuses, how willing Moonshot's models are on a tougher test, and the closest alternative.
Is DeepSeek Safe to Use? Privacy, Bans and Safer Ways to Run It in 2026
DeepSeek's app sends everything you type to servers in China, and governments keep banning it from official devices. The open model is a different story. Here's what's documented and how to use it safely.
Cleus Play: Turn Any Idea Into a Playable AI Story Game
Type one line and Cleus builds the world, the cast and the opening scene. Then you play it: talking characters, music, three choices a scene and dice rolls when you take a risk.
How to Create an AI Character That Feels Real (2026 Guide)
A practical guide to making your own AI character: what makes one feel real, the mistakes that make it flat, how the fields map across Character.AI, Janitor, Tavern cards and Cleus, and a full before-and-after rebuild.
The Best AI Boyfriend Apps in 2026: Memory, Texting First and Slow-Burn Romance
Six AI boyfriend apps checked in November 2026 against what people actually ask for: memory, texting first, voice, slow-burn romance, no unwanted explicit content, and privacy.
CVE-2026-48356: Magento 'PolyShell' Flaw Now Confirmed Exploited — Patch Adobe Commerce Now
CISA has confirmed attackers are exploiting CVE-2026-48356, the 'PolyShell' file-upload flaw in Adobe Commerce and Magento Open Source. Here's who's affected and how to patch.
CVE-2024-27834: Old Safari Bug Now Exploited Against Unpatched iPhones and Macs
CISA just confirmed attackers are using CVE-2024-27834, a WebKit Pointer Authentication bypass Apple fixed in May 2024. Updated devices are safe; old, unpatched iPhones, iPads and Macs are not.
The Best Chai App Alternatives in 2026 (After Chai Left the App Store)
Chai's iPhone app left the App Store on October 22, 2026. Here's what happened, and eight Chai alternatives compared on price, free tier, filter and platform, plus how to move your bots.
The Best AI Girlfriend Apps in 2026: Memory, Calls, Photos and Privacy Compared
Which AI girlfriend app actually feels like a girlfriend? We compare Cleus, Candy.ai, Nomi, Kindroid, Replika and Character.AI on memory, texting first, calls, photos, real prices and who can read your chats.
GPT-6 Cyber Explained: OpenAI's 'Critical' Hacking Model Found 81 New Firefox Flaws
OpenAI's GPT-6 Cyber is its first cyber-specialized model rated Critical. It found 81 new Firefox bugs and built 948 Linux kernel exploits, but only vetted Daybreak Red teams can use it.
Stock Market Crash: Why Tech Stocks Lost $2.3 Trillion in a Day After the Midterms
The Nasdaq 100 lost about $2.3 trillion and chip stocks had their worst day since 2020 as investors bet the midterm blue wave will slow the AI data center boom. Here is what happened and what is still unknown.
French Bank Hacked: How Crédit Mutuel Arkéa Lost an Estimated $133 Million
Crédit Mutuel Arkéa says attackers came in through a supplier and sent money from the bank's own account. ZachXBT traced more than $133M into stablecoins, Ether and Monero.
The Best Uncensored AI Video Generators in 2026
Which AI video generators actually allow adult content with fictional adults, and which are mainstream-filtered? Eight options compared on length, resolution, sound, image-to-video and price.
The Best Replika Alternatives in 2026: 8 AI Companion Apps Compared
Why people leave Replika, what it still does well, and how eight AI companion apps compare on memory, voice calls, photos, adult content rules, free tiers and price.
AI Data Centers and the 2026 Midterms: What the Election Results Mean for AI
Data center bans passed almost everywhere in Ohio, Texas elected a governor who wants a moratorium, and AI stocks slid. Here is what the 2026 midterms mean for the AI buildout.
The Best Uncensored AI Story Generators in 2026 (NovelAI and Sudowrite Alternatives)
Six AI story writers compared for mature fiction in 2026: content rules, long-form memory, chapters, export and real prices from each official page.
The Best Uncensored AI Image Generators in 2026: 7 Options Compared
Seven uncensored and NSFW-capable AI image generators compared from their own pricing and policy pages: what each allows, what it costs, and where each draws the line on real people and minors.
Cleus Code: The Uncensored Claude Code Alternative That Works on Your Whole Computer
Claude Code and Codex stop at a refusal. Cleus Code is an uncensored AI coding agent that writes and runs code, uses your browser and desktop apps, and makes images, video and music. Here's what it can do.
Best Janitor AI Alternatives in 2026: Free, No Filter and Proxy-Friendly Picks
Seven Janitor AI alternatives checked on November 3, 2026: which are free, which have no filter, which take your own API key or proxy, which make images and which import your character cards.
The Best Uncensored AI Chatbots in 2026: 7 Picks for Frank Answers, Privacy and Local Use
Want an AI with no filter for questions, research, writing and code? We compared seven options on price, what they really allow, and who can read your chats.
Anthropic IPO Filing Explained: A $17B Quarter, a $23B Loss and an $895B Compute Bill
Anthropic's amended S-1 adds Q3 2026: $16.96B revenue, a $23.3B net loss that is mostly an accounting charge, and $895.4B in compute commitments. Here's what each number means.
The Best Character.AI Alternatives in 2026: 7 Apps Compared After the Bankruptcy
Character.AI is in bankruptcy and users are leaving. We compared seven alternatives on price, filters, memory and whether you can bring your characters, using each app's own pages and App Store listings.
Marvell Buys Lumentum for $115 Billion: What the Deal Means for AI Data Centers
Marvell is buying laser maker Lumentum in an all-stock deal valued at $115 billion, the biggest tech acquisition of 2026. Here is what Lumentum makes, why AI data centers need it, and what could still go wrong.
New York AI Companion Law Explained: The 12 Chatbot Features Banned for Minors
New York's new AI companion law switches off 12 chatbot features, from posing as human to flattery and sexual content, for under-18s and anyone not age-checked, starting January 1, 2027. Here's what it says, who enforces it and what changes for adults.
Kling 4.0 Explained: Features, Release Date, Pricing and How to Use It
Kling 4.0 makes up to 30 seconds of video with stereo sound in one pass. Here's what is live, what is only announced, how it compares with Veo 3.1, Wan 3.0 and Seedance 2.5, and how to prompt a full 30-second clip.
Character.AI Bankruptcy: Is Character.AI Shutting Down, and What Happens to Your Chats?
Character.AI filed for Chapter 11 and bids are due November 20. Disney's $20M offer doesn't cover the app or your data. Here's what the court filings say, whether it will shut down, and how to save your chats and characters before the sale.
DeepSeek Jailbreak Prompts in 2026: Why They Keep Breaking (and What Works Instead)
Researchers broke DeepSeek R1 on every test prompt, yet your copied jailbreak stops working after three messages. Here is why DeepSeek jailbreaks are so unreliable in practice, and what to use instead.
DeepSeek V4.1 Flash for Roleplay and Story Writing: Uncensored Prompts, Settings and Tips
DeepSeek V4.1 Flash writes fast, vivid fiction, until it refuses your plot. Here are the settings, prompt templates and memory tricks that make it a great roleplay and story partner, and the no-refusal way to run it.
What Is an Abliterated Model? How AI Refusal Removal Actually Works
Abliterated models answer what normal chatbots refuse, without jailbreak prompts. Here is how refusal removal works inside the model, what it breaks, and where to try an abliterated DeepSeek V4.1 Flash.
How to Use DeepSeek V4.1 Flash Uncensored (No Refusals, No Setup)
DeepSeek V4.1 Flash is one of the strongest open models of 2026, but the official app still filters and refuses. Here are the real ways to use it uncensored, and the one that takes no setup.
What to Do After a Data Breach: Check If You Were Leaked and Lock Down Your Accounts
Your details were in a data breach. Here is how to check what leaked, the seven steps that lock your accounts down, how credit freezes work in the US, UK, EU and Japan, and what criminals can do with each kind of data.
Bybit Hack Explained: How North Korea Stole $1.5 Billion From Bybit
On February 21, 2025, Bybit lost about $1.5 billion in ETH without a single private key leaking. How a hacked Safe{Wallet} laptop and a lying screen did it, how Bybit survived, and how little was ever recovered.
EIP-7702 Explained: How Ethereum Delegation Works and How to Revoke It
EIP-7702 lets an ordinary Ethereum wallet borrow a smart contract's code with one signature. Here's how delegation works, how phishers and the Binance hackers abused it, and how to check and revoke yours.
Lazarus Group and TraderTraitor: North Korea’s Crypto Hackers Explained
Who the Lazarus Group and TraderTraitor really are, the crypto heists the FBI and analytics firms have pinned on them, the lures they use, why North Korea does it, and how to defend against them.
The Biggest Crypto Hacks in History: 17 Heists Ranked (2026 Update)
From Mt. Gox to Binance's $6.36 billion theft: the 17 largest crypto heists ever, what each was worth when stolen, how attackers got in, how much came back, and how to protect your own coins.
KDDI Data Breach Explained: How a Zero-Day Exposed 12 Million ISP Email Accounts
A zero-day in third-party software let attackers into the email platform KDDI runs for six Japanese ISPs. Who was hit, what leaked, what regulators did and what to do now.
Amazon Earnings Q3 2026: AWS Grows 39% as AI Spending Hits $77 Billion a Quarter
AWS grew 39%, its fastest in nearly five years, while Amazon spent $77.3 billion in one quarter on AI data centres and chips, pushing free cash flow to −$42.4 billion. Here's what it buys and how AWS compares with Azure and Google Cloud.
New AI Models Released in October 2026: Every Major Launch, Dated and Verified
Meta and Mistral brought open weights back to the frontier, while OpenAI, Anthropic and Google shipped cheaper models. Here's every major AI launch of October 2026, with dates, specs, prices and what didn't ship.
Japan Data Breaches, October 2026: Rakuten Travel, Furusato Choice and a Month of Leaks
Furusato Choice, Rakuten Travel, Sony's So-net, Capcom, Sourcenext, Burger King Japan: October 2026 brought so many leaks that Japan's prime minister called an emergency ministers' meeting. What each one exposed, why Japan is being hit, and what to do.
Polymarket Hacked: How a Hijacked DebugBear Script Drained About $2.7M From Traders
On October 29, a hijacked website-monitoring script hit Polymarket's international site, sold traders' positions and moved out roughly $2.7 million. The timeline, how it worked, the refund rules and what to do now.
Sony Data Breach 2026: How Hackers Took 2.58 Million So-net and plala Records
An outsider broke into Sony's So-net member database on October 19 and took up to 2.58 million customer records, plus 54,401 email passwords. Here's the timeline, exactly what leaked, and what to do now.
Binance Hack Explained: How Hackers Stole $6.3 Billion in 20 Minutes
Binance was hacked for about $6.3 billion, the biggest crypto heist in history. One compromised laptop, 50 pre-signed signatures, ten blockchains and about 20 minutes. Here’s how it happened, with a replay, a game and a quiz.
How End-to-End Encryption Works on Cleus
Your conversations with an AI are some of the most personal data you'll ever type. On Cleus, premium users can turn on zero access encryption — a mode where…