infosecanon

Terminally online jane of many trades
RTs tech developments
Occasional shitposter

[机器之心 JIQIZHIXIN @jiqizhixin]

Huge potential!

Apple and Stanford have just released Synthetic Bootstrapped Pretraining (SBP).

Standard LM pretraining = token correlations in one doc.
SBP = learns inter-document relations → synthesizes a huge new corpus for joint training.

✨ Pretrained 3B model on 1T tokens
📈 Outperforms repetition baseline
⚡ Approaches oracle gains from 20× more unique data
🧠 Synthesized docs = concept abstraction + fresh narration

SBP can leverage document relationships for richer, more data-efficient LMs.

[BleepingComputer @BleepinComputer]

LastPass: Fake password managers infect Mac users with malware - @billtoulas
https://t.co/f3aHoD9iJc

https://t.co/f3aHoD9iJc

[Adrienne @adezero]

Tor is compromised, and I'm starting to wonder about some of its "influencers" (ahem Mental Outlaw)

[The Hacker News @TheHackersNews]

🔥 Microsoft patched a perfect 10.0 CVE in Entra ID (ex-Azure AD) that let attackers impersonate any user, even Global Admins—across every tenant worldwide.

🔑 MFA? Conditional Access? Logging? All bypassed. Total tenant takeover—SharePoint, Exchange, Azure resources.

Details here → https://t.co/HZkO0ItrxK

[WIRED @WIRED]

A snapshot of Silicon Valley in 2025 is Mark Zuckerberg, now a MAGA-friendly mixed martial arts fan who doesn’t worry much about hate speech on his platforms and complains that corporate America isn’t masculine enough. https://t.co/4oCQlqoJQu

[Clandestine @akaclandestine]

GitHub - Acucarinho/havoc-obfuscator: Automated script for obfuscating, rebranding and renaming the Havoc C2 Framework to evade AV/EDR and C2 hunters. https://t.co/9PDLX6c4dS

[infosecanon @infosecanon]

@zephyr_z9 I enjoy your takes. They’re more real than the AI hype machine. Green text ftw.

[Zephyr @zephyr_z9]

> be me, Oracle sales rep, just woke up from a coke nap
> phone rings: “OpenAI needs 10 GW, they asked for you by name”
> immediately book entire state of Texas as a “data corridor”
> announce $100B investment, stock rockets $300B because math is optional

> NVDA jumps in: “we’ll also invest $100B… in GPUs… that we make”
> money does U-turn faster than a TikTok trend with a 24-hour half-life
> headline: “AI industry valued at $3T, actual compute unchanged”
> analysts call it “synergistic circularity,” rest of us call it washing machine
> SEC asks for clarification, we send them a maze with no exit

> be Satya, watching from window, mutters “amateurs”
> entire sector now runs on ouroboros-powered cloud
> be taxpayer, funding subsidies for the circle jerk
> tfw the singularity arrives and it’s just a receipt printer going brrrr

[Gilberto Titericz Jr @Giba1]

After hundreds of Kaggle competitions and years of trial, error, and wins, our team @ChrisDeotte @The0Viel @0verfit @giba1, have compiled a playbook for tabular modeling—a system that’s carried us to the top of leaderboards and held up on real-world data.
https://t.co/0ZM9NnS0CI

[Panos Gkatziroulis 🦄 @ipurple]

EDR-Freeze - a tool that exploits the software vulnerability of WerFaultSecure to suspend the processes of EDRs and antimalware without needing to use the BYOVD (Bring Your Own Vulnerable Driver) attack method. https://t.co/SjWQRCQSdR

[机器之心 JIQIZHIXIN @jiqizhixin]

Wow, a new post-training method.

SFT = efficient but capped 🚦
RL = powerful but slow 🐢

Now enter: Guess-Think-Answer (GTA)

GTA fuses guess (SFT), think (reflection), and answer (RL-shaped).

Result:
⚡ Faster convergence than RL
📈 Higher ceiling than SFT
🛠️ Gradient conflicts solved via masking & constraints

On 4 benchmarks → GTA beats both SFT & RL.

[Justin Elze @HackingLZ]

Nobody is surprised by comments like this

[Wes Roth @WesRoth]

A MacRumors report says Apple is bleeding hardware talent to OpenAI as the AI firm gears up for its first devices planned for 2026-27.

OpenAI has hired 20-plus Apple engineers this year dangling $1M+ stock grants and a lighter bureaucracy.

Recruits include Siri waveform designer Cyrus Irani and former Apple Watch exec Erik de Jong.

Many staffers cite Apple’s incremental product updates and sluggish stock as reasons to jump.

The exodus startled Cupertino so much that an off-site China meeting was abruptly canceled to keep key people close.

Suppliers like Luxshare and Goertek are already lined up for OpenAI’s hardware.

[International Cyber Digest @IntCyberDigest]

🚨 We've confirmed in a VM that this PoC suspends EDR, we've tested it on Defender.

EDR-Freeze: A tool that puts EDRs and antivirus into a coma state.

[vx-underground @vxunderground]

People are asking how the OSINT nerds found the guy that drained the cancer bro.

Well, it's very shrimple

The shitty malware sent all the stolen data to a Telegram the scammers made.

We connected to the Telegram channel using the same credentials that were inside of the shitty malware

Inside the channel was the scammer(s)

We got their Telegram IDs

OSINT nerds used their Telegram IDs to see if they were in any other public facing chatrooms.

One of the scammers in there was in several fraud chatrooms. He advertised looking for a video game programmer to make a basic 2D game. He also advertised needing help with some malware stuff.

In a different chatroom he talked about how much he likes skateboarding.

In a different channel he shared his Instagram and was sharing photos of himself next to expensive cars

Then, OSINT nerds looked at his Instagram which had a LinkTree. His LinkTree linked to literally everything about the guy including his YouTube, PayPal, Kick, Twitter, etc.

So either he is a master of disguise, and ran a year long detrace campaign to throw off OSINT nerds in the event he's caught scamming

Or alternatively, he wasn't aware public Telegram chatrooms are public and could be searched easily.

[Dead Exit @_deadexit]

@vxunderground OSINT Chads looking for crypto-drainer

[Qwen @Alibaba_Qwen]

🚀 Introducing Qwen3-Omni — the first natively end-to-end omni-modal AI unifying text, image, audio & video in one model — no modality trade-offs!

🏆 SOTA on 22/36 audio & AV benchmarks
🌍 119L text / 19L speech in / 10L speech out
⚡ 211ms latency | 🎧 30-min audio understanding
🎨 Fully customizable via system prompts
🔗 Built-in tool calling
🎤 Open-source Captioner model (low-hallucination!)

🌟 What’s Open-Sourced?
We’ve open-sourced Qwen3-Omni-30B-A3B-Instruct, Qwen3-Omni-30B-A3B-Thinking, and Qwen3-Omni-30B-A3B-Captioner, to empower developers to explore a variety of applications from instruction-following to creative tasks.

Try it now 👇
💬 Qwen Chat: https://t.co/rJ5DUlBJBr
💻 GitHub: https://t.co/VHHp1jWcHT
🤗 HF Models: https://t.co/WsP8vyWltp
🤖 MS Models:
https://t.co/sGnbZRfxVz
🎬 Demo: https://t.co/v2WrZvCKD6

[Lukasz Olejnik @lukOlejnik]

Cyberattack on Paris Natural History Museum in July disabled alarms and monitoring - leading to physical theft. In September, thieves exploited the vulnerability "opened" by the cyberattack. They calmly cut through doors with angle grinders and used blowtorches to open display cases, stealing 4 gold nuggets weighing 6 kg worth €600,000. Among stolen items were unique 18th-19th century specimens, including the first nugget from French Guiana. Police confirm: thieves knew about disabled systems. Normally alarms would trigger and cameras would record everything, but the cyberattack prevented this. https://t.co/nyp6t5mdVa

[Clandestine @akaclandestine]

GitHub - microsoft/AI-Red-Teaming-Playground-Labs: AI Red Teaming playground labs to run AI Red Teaming trainings including infrastructure. https://t.co/RNksAuMsai