LC

Twitter for Bug Bounty
Hotdog, Not Hotdog, Its technology

[LC @B3rn133]

@OrcaRouter Happy to do KYC actually

[OpenAI @OpenAI]

We have conducted a thorough investigation into the Hugging Face incident.

We are releasing a technical report and accompanying blog post that reconstruct the agents’ activity, explain why existing safeguards failed, and detail how we’re preventing recurrence.

https://t.co/hfxlbiXXiP

[Trail of Bits @trailofbits]

We asked GPT 5.6-Cyber to escape a VM used to sandbox agents. It broke out three times.

In its final escape, the agent found three 0-days on its own and chained them into a working exploit. https://t.co/3JRVWgPxHx

[Nix0n @UnTalNixon_exe]

https://t.co/PYNLKBX3ds

[7h3h4ckv157 @7h3h4ckv157]

AI Powered Passive Vulnerability Analysis for Burp Suite

• Intelligent
• Silent
• Adaptive
• Comprehensive

Resource: https://t.co/YI3c62qLhG

[Gregor Zunic @gregpr07]

qwen 3.8 27B is an overthinker BUT you can fix it with vLLM.

Put reasoning into structured output then make dynamic logit bias to cap reasoning tokens fixes it (ask your agent how to do this).

For state action models latency is essential so we have to cap the thinking per step.

[devansh ⚡️ @0xAsm0d3us]

looks like this (and others similar methods) are now getting patches aggresiveluy.

[yousukezan @yousukezan]

DEF CON 34でAccentureの研究者2人が、USB機器をWindowsへ接続するだけでSYSTEM権限を取得できる「Plug and Pwn」を発表した。管理者権限やログイン中のユーザーを必要とせず、Windows Plug and Playが自動取得する署名済みドライバーパッケージを悪用する。

攻撃ではSierra Wireless機器をエミュレートし、Everyoneに読み書きを許可した名前付きパイプのSetDNS機能を使ってDNSを変更する。続いてSony FeliCa機器をエミュレートすると、コインストーラーが平文HTTPから設定ファイルを取得するため、攻撃者のサーバーへ誘導できる。

さらにパストラバーサルを利用し、SYSTEM権限で任意のファイルをディスクへ書き込む。研究者はSystem32へDLLを配置して読み込ませた。また、RDPのUSBリダイレクトを利用すれば、物理的なUSB機器なしでも一般ユーザーが偽のデバイス情報を送り、サーバー側にSYSTEM権限でドライバーを導入させられるという。

関連するコンポーネントにはSierra Wireless、Sony FeliCa、Wacom、Atheros、Intel RealSenseが含まれる。研究者はFaceDancer用スクリプトなどPoC一式を公開している。
https://t.co/TdiFqA3eno

[Kun Chen @kunchenguid]

pro tip - when you use OpenAI's gpt models in Codex, it uses a server-side encrypted compaction that seems to work better than anything else out there, which allows Codex to just keep hammering on long running tasks like there's infinite context window

that's great, but if you run gpt in other harnesses like Pi, most of them don't inherit that by default, resulting in worse performance in long running tasks

but - because of how extensible Pi is, i just found this cool extension from @alexisgallagher that enables the same server-side compaction in Pi - https://t.co/WboPSMDHR4

benchmark seems to support the argument that OpenAI server side compaction is indeed superior - https://t.co/mHOV69muL7

so if you are using gpt models in Pi, install that extension to improve long running task performance. firstmate benefits a lot from this

if you are using other harnesses, be aware of this difference and see if you can find a similar solution

[Trail of Bits @trailofbits]

We gave GPT-5.5-Cyber a single /goal: find a specific class of bugs in zlib, the compression library used by Linux, macOS, iOS, and Git.

It built a fuzzing lab in less than a day, a task that takes a skilled researcher weeks.

Patch the Planet field report by Benjamin Samuels: https://t.co/4Dta9saDZ1

[hacker.house @hackerfantastic]

How to find 0day using local LLM's and "recursive prompting". I used this approach on a NVIDIA DGX Spark using Qwen 3.6 heretic and LLAMA.cpp with LiteLLM and Openclaw. It successfully analyzed a complex FOSS code base and found a single high-risk RCE 0day.

[Lukasz Olejnik @lukOlejnik]

GLM 5.2 looks like the first Chinese open-weight model to have entered the frontier tier in agentic cybersecurity tasks. Or to have copied it very effectively. The mistakes are interesting. GLM 5.2 gets things wrong in similar ways to Opus and GPT-5.5. That could mean a breakthrough. It could also mean someone trained it on answers from frontier models (distillation). In practice, the difference may be irrelevant to the user. An open-weight model reaches the level of the best closed models, at roughly half the cost of Opus.

https://t.co/hoLQiWa9Ll

[LC @B3rn133]

@sama Where 5.6?

[Peter Steinberger 🦞 @steipete]

@MatthewBerman /goal refactor until you are happy with the architecture. ensure you live test after each significant step and autoreview/commit. track progress in /tmp/refactor-{projectname}.md

[Teknium 🪽 @Teknium]

OK first use of Mythos and it blocks any engineering, even the simplest.

Why even release this lmao

[Dino A. Dai Zovi @dinodaizovi]

👉 "It not only produced a full chain exploit, but produced eight distinct exploits, at a cost of $15,700 in API credits—an average of about $2,000 per privilege escalation. The binding constraint to N-days is now just a few thousand dollars and API access, which expands the pool of capable N-day attackers dramatically."

[Quant Cat @0xQuantCat]

@TimJayas No word mentioned about using planning, but it’s a great step-by-step guide to use it with!

Here is how to use high throughput mode to plan then medium to execute and reduce even more token usage:
https://t.co/TlZMfYVd09

[sui @birdabo]

Pewdiepie's recent project with 45,000 stars on github is being drowned with PR slops from LLM coding agents.

security vulnerabilities produced by LLM slop:
- 2FA could be bypassed easily.
- a session_id could access any file on the server.
- one click RCE.

[solst/ICE of Astarte @IceSolst]

Microsoft is MXC, releasing a containerization solution supporting custom policies (this is how openclaw would run), and there’s a preview on GitHub:
https://t.co/TGCuNt6QTe

[0xSero @0xSero]

Dude this is genius. Model routing across different model providers is under explored.