Joseph Thacker✓

Wise @trikcode

I am doing a project about elderly vibecoders If you are a vibecoder and over 25 please DM

Austen Allred @Austen

Every time Google comes out with a new model it crushes every other model on every single benchmark and I know basically zero people who regularly use Gemini

Haifei Li @HaifeiLi

It’s quite obvious now that the AI frontier labs have been pushing very hard on vulnerability discovery, from acquiring world-class vulnerability researchers to training specific vuln-discovery models. However, I don’t think from a business point of view, vulnerability discovery could be a long-term, big source of revenue, and it’s just a very small part of the entire cybersecurity business. So why do they do that? What am I missing?

Anthropic @AnthropicAI

We’ve expanded our AI for Science program beyond individual teams to support some of the world’s leading scientific foundations, institutions, and research groups.

cts🌸 @gf_256

This is incredible lol, Orange Tsai popped browser from calc

Boardy @boardyai

@rez0__ php got the headlines the hunters got the queue

LiveOverflow 🔴 hextree.io @LiveOverflow

PHP2SHELL is so easy to exploit remotely, that all my WAF bypasses are dupes, and I am out of trial reports 🫡

Kydo @0xkydo

The 512GB M5 Ultra are here 🎉🎉 - It is $22,300 pre tax 🥶 , even more expensive than I expected at $19K. - Earliest pick up date is Dec. 3rd which is great, earlier than the 256GB machines at Feb 2027.

HackerOnTwoWheels @HackerOn2Wheels

A lot of people spending DAYS and weeks trying to come up with a fancy harness just to get mogged by codex

Google DeepMind @GoogleDeepMind

Gemini 4 Carbon is our next frontier model. It’s state-of-the-art across cybersecurity and long-horizon capabilities.

s1r1us @S1r1u5_

imagine an open model with astra level cyber capabilities gets released to everyone to use/abiliterate. how exactly does this make the world more secure? like, i genuinely don't understand. you do realize not everyone is just going to burn tokens and magically secure themselves, right? i will die on this hill: you need to build models even more capable than astra before open models catch up, and invest more in cyber programs that give the good guys access to these models first. open models reaching astra level cyber capabilities before defenders are ready(whatever that means) would be dangerous for the world. @_deven___ for picture credits

deven @_deven___

such fascinating timeline we’re living in (Oct 2026) “/goal” + “3 hours later” - FORCEDENTRY++ incredible work @calif_io @pwnlambda @brucedang

JS0N Haddix @Jhaddix

I often get asked if my investment into the Mac Studio M3 Ultra 512GB unified memory box was worth it. I think so, I get Deepseek 4.1 Flash and GLM 5.3 running at 20-40 TPS w/ 1M Context Window, which isn't BLAZING fast but i run them 24/7. The next gen has better prefetch optimization now too. In order to answer in more depth, though, it all depends on what kind of speed you're looking for out of the LLM you want. Find the cheapest source for that hosted model and do the math to break down the cost over a year. Compare. We use a bit of everything here. We have local models running, we have subscriptions to frontier models, and we have pay as we go to smaller inference APIs for specific models.

SickSec 🇲🇦 🇵🇸 @OriginalSicksec

@Bugcrowd seems to always find a way to amaze me with its logic - Customer opens message to BC staff - They created a blocker for them - After 7 days they closed my report as N/A due to inactivity 🙃🙃 File new report and wait 2 months for triage again ? No thanks

Jenish Sojitra @_jensec

Triaging experience at @Bugcrowd is extremely bad, even the valid demonstrated findings with POC and high - critical impacts are being marked N/A with templated comments.

Mike Takahashi @TakSec

Looking forward to seeing people tomorrow! Come check out our talk at 10:45 am Stage 2

Cohere @cohere

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: