␋

@IntCyberDigest @Jake_Hanrahan

@Jake_Hanrahan Sounds like they are just doing community service.

Anti-loan shark vigilantism in Colombia.

@DanielMiessler Not hate, just that these words are different things, and tbf we are not even close to AI, these models are limited by the same nature of their reinforcement learning strategy, we should call things by their name.

People who hate AI love to call it "Generative AI", or "LLMs". It's a screeching tell. It's like calling the person who cleans your teeth a "Not a Doctor", instead of a dentist. If you're an AI researcher who deals with lots of different types of AI, then obviously that makes sense. But in 2026, in the vast majority of conversations, it contributes absolutely nothing to the discussion while waving a neon protest sign in your face that says, "I HATE ALL OF THIS"

@jun_song Organic DDoS

The only thing I'm worried about with Kimi-K3. Since this model is on GPT-5.6 level and better than Opus, a ton of people are going to buy the subscription. But they simply don't have enough compute to serve it. Even with Arena right now, the model got significantly slower because it’s crowded. Everyone is going to rush in at once and it'll be impossible to use.

@OpenAIDevs @work_louder * Slaps the laptop keyboard "This babe can connect to any other keyboard extending your keyboard range, into another keyboard, one keyboard to rule them all"

Meet kbd-1.0-codex-micro, built with @work_louder. Map the buttons and joystick to your workflow, and keep your pinned chats in view. Get yours before stock returns 410.

@OpenAIDevs @work_louder

@bot59751939 @ctjlewis "... without breaking my flow", sure it didn't

@ctjlewis "so we invented... um well we made a keypad... yeah chatgpt came up with it, how could you tell?"

“Babe, your goycattle clicker finally arrived!”

@RussianPanda9xx @akses_0x00 damn

I have been seeing reverse engineering / malware analysis courses that cost thousands of dollars. Do you know you can actually learn that for free, right? There are also a bunch of other courses out there that are way cheaper or even free: - courses.0ffset.net - malwareanalysis-for-hedgehogs.com - invokere.com - github.com/RPISEC/MBE - guyinatuxedo.github.io/index.html - openanalysis.net - p.ost2.fyi - taomm.org

@github From September 2025 to June 2026, still waiting for that security bulletin.

tl;dr: you should probably be at GitHub Universe.

@xploitrsturtle2 oh man, no friends, no future.

Tell me your friends ill tell you your future

@DailyDarkWeb Anyone confirmed? this seems to be fake, from a previous breach plus LLM sloppiness.

🇺🇸 Duolingo Alleged 500 Million User Database Advertised on Underground Forum A threat actor is advertising what they claim is a database containing more than 500 million Duolingo user records. The listing includes a sample of the alleged dataset and contact details for potential buyers. The seller claims the database contains: * Usernames and display names * Email addresses and verification status * Password hash indicators * Phone numbers * Country, timezone, and language preferences * Learning progress and streak statistics * XP, leagues, achievements, and subscription tier * Device identifiers and registration metadata * Account creation and last activity timestamps At the time of writing, there is no public confirmation from Duolingo validating these claims. The authenticity, origin, and scope of the alleged dataset remain unverified. Analyst Note: Large consumer platform databases are frequently recycled, repackaged, or combined from previous incidents. Organizations and users should avoid assuming the data is new until its provenance and freshness can be independently verified. #DDW #Intelligence #DarkWeb #Duolingo

@quxiaoyin Show us the numbers around this claim.

I went a dinner a few weeks ago with a bunch of enterprise execs who told me "we will never use Chinese models." "Even if it's 100x cheaper?" "No, we care about safety and security." 1. They don't understand when they host open-source models with their own GPUs or US data centers, they won't share their data to China. 2. They are giving away all their data to OpenAI and Anthropic rather than owning it privately themselves. 3. They don't understand math. 100x is a big number and lots of profits. It's almost July 2026 now. If your execs still talk like that, fire them now.

@KettlebellDan Achualy, these models are vulnerable to the same type of attacks that any other LLM, no safe switch exist for private or open models, they have foundational vulnerabilities due to the type of statistical approach used for regression patterns, at the end not a single model is safe

not sure why anyone would trust China open AI models when: undetectable backdoors survive safety training no kill switch once weights are open opaque unverifiable training data/intent

@bot59751939 @ZackKorman you can automate a process for centralize raw dara according to the context, but in the same way can be tricked with false info, at the end you have an expensive MCP processing noise data.

@ZackKorman Aside from the fact that you can download their and other backdoored models from HuggingFace, I've worked in some pretty extreme environments. Very little surprises me anymore. "Only the paranoid survive" is not just a cute saying.

The fact that this has 2.8k likes tells you that people have zero understanding of AI agent security. This sounds spooky. It’s also the least likely way an AI agent might cause a data breach. Please live in the real world.

@bot59751939 @ZackKorman This is a PT learning strategy, unless that you are working with a MoE based in synthetic data and a foundational base like BERT, but that's not the case, the risk is associated to the ingested data, but malicious actors are still abusing of misconfigurations and bad practices.

@yuhasbeentaken isaiprofitable.com

silicon valley: ai is the biggest money machine ever meanwhile... estimated cumulative ai spend vs revenue: 1. amazon: $331b spent → $22b revenue 2. google: $303b spent → $25b revenue 3. meta: $238b spent → $3b revenue 4. microsoft: $256b spent → $31b revenue 5. openai: $62b spent → $26.7b revenue 6. anthropic: $33b spent → $22b revenue maybe it’s the most expensive infrastructure war in history. the direct payback is still mostly a promise.

@rekdt They are actively working to seize that one too.

Hearing there might be a chance .io survives so we can still do netnut November

@RiderHomie @vxunderground Smelly shared a sample some time back

@RiderHomie @vxunderground Vol.3, insane

"hey smelly, out of curiosity, how did you learn so much? have you read every paper on vxug?" fuck no. i probably havent read 10% of the library. if you unironically read everything there you would be two things 1. ultra mega fuck off malware brain 2. profoundly depressed

@vxunderground 2, sir, thanks

@vxunderground - "Make no mistake" CIRCA 2026

Chat, I don't want to jump to conclusions, but I have a sneaking suspicion this malware stager was vibe coded. Historically, malware hasn't left extremely descriptive comments in their stagers.