September 25, 2025
September 25, 2025
We are releasing details on BRICKSTORM malware activity, a China-based threat hitting US tech to potentially target downstream customers and hunt for data on vulnerabilities in products. This actor is stealthy, and we've provided a tool to hunt for them. https://t.co/kuZ1UUUz6H
— John Hultquist (@JohnHultquist) September 24, 2025
Inside the Jaguar Land Rover hack: stalled smart factories, outsourced cybersecurity and supply chain woes
Inside the Jaguar Land Rover hack: stalled smart factories, outsourced cybersecurity and supply chain woes | Jaguar Land Rover | The Guardian
Being a carmaker where ‘everything is connected’ has left JLR unable to isolate its plants or functions, forcing a shutdown of most systems
Zack Whittaker: "* breathes in * Fuccccccccckkkkkk thiiiisssssss.…" - Mastodon
* breathes in * Fuccccccccckkkkkk thiiiisssssss. How this sketchy call recording app got to #2 in the Apple App Store is beyond me. https://techcrunch.com/2025/09/24/neon-the-no-2-social-app-on-the-apple-app-store-pays-users-to-record-their-phone-calls-and-sells-data-to-ai-firms/
neat malware trick, but if I had to choose my fighter I'm betting a cracked RE also clocks this. https://t.co/YPiieksCT8 pic.twitter.com/9raD8dxOE2
— J⩜⃝mie Williams (@jamieantisocial) September 25, 2025
There are some truly wild reasoning traces in @apolloaievals & OpenAI's recent paper
— Jeffrey Ladish (@JeffLadish) September 25, 2025
The models appear to have developed specific uses for the words "marinade" "overshadow" "illusions" "vantage" and others. This seems likely to be the result of RL training pic.twitter.com/h9vkLgxv4V
I highly recommend reading the paper and appendix G: Unusual Terminology in Chain-of-Thought.
— Jeffrey Ladish (@JeffLadish) September 25, 2025
There are a lot more interesting results in this paper than the strange reasoning scratchpad terminology! But the quotes above really stood out to mehttps://t.co/onBCnN7AXg
BIG claim. Giving an LLM just 78 carefully chosen, full workflow examples makes it perform better at real agent tasks than training it with 10,000 synthetic samples.
— Rohan Paul (@rohanpaul_ai) September 24, 2025
"Dramatically outperforms SOTA models: Kimi-K2-Instruct, DeepSeek-V3.1, Qwen3-235B-A22B-Instruct and GLM-4.5. "… pic.twitter.com/8G1zNXs45T
This may save someone else a few minutes. I've compiled a few random vibecoded tools for generating/analyzing SVGs commonly found in many phishing emails that exploit Microsoft Direct Send. https://t.co/kIvT2FaAqC
— Justin Elze (@HackingLZ) September 24, 2025