Anthropic Cowork looks great, but it’s Claude Max + macOS only for now. AiPy has been doing "AI works on your computer" ("Vibe Working") for ~1 year — free + Windows/macOS/Linux.
EN: aipy.app/ | ZH: aipyaipy.com/ | CLI: github.com/knownsec/aipyapp @AiPyapp
heige retweeted
Just trying to cd into my Solana/Agave runtime codebase and OpenAI Safety decided dev/agave/runtime is a threat to national security. 😭
Turned Daybreak ON, and it still got blocked with: "Some cybersecurity requests are still limited, even when Daybreak is on."
Is cd now a zero-day exploit? @thsottiaux 😅😅😅
Don't worry, many guys has 200+ cases with the same sitituation, they are worrier than you :-)
two months ago, I reported a critical bug to @msftsecresponse, but it has received no attention since. It's disheartening. I still appreciate the company, but there's no denying that their bug bounty program requires improvement 💔.
I used /goal + ASC + adb to analyze my phone. The agent pulled 50+ APKs with uid=1000, then used ASC to analyze them without exporting any pseudocode. After 10 hours, the agent had rooted my phone...😐
#mobile #bugbounty #security
github.com/MG1937/ASC
xitter.cf/MGAldys4/status/209854…
Guys, I built a super fast Android decompiler called ASC.
It completely replaced Jadx MCP for me and lets me analyze 10+ APKs in parallel.
And this week my Agent use ASC found 2 RCE in Honor and Xiaomi!
This tool now accepted by BlackHat EU Arsenal
github.com/MG1937
#BHEU
Guys, I built a super fast Android decompiler called ASC.
It completely replaced Jadx MCP for me and lets me analyze 10+ APKs in parallel.
And this week my Agent use ASC found 2 RCE in Honor and Xiaomi!
This tool now accepted by BlackHat EU Arsenal
github.com/MG1937
#BHEU
Introducing GLM-5.3-Flash
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog: z.ai/blog/glm-5.3-flash
Available now across all official platforms:
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.3…
Coding Plan: z.ai/subscribe
ZCode: zcode.z.ai/en
Chat: chat.z.ai
AutoClaw: autoclaw.z.ai
Hard to believe OT exposure is still this widespread in 2026. (ps:ZoomEye data includes honeypots/test systems, but the trend holds). With LLMs accelerating vuln research, ICS will face even more pressure—7 Siemens CVEs entered CISA KEV in 2026. siemens.com/en-us/content/ce…
heige retweeted
I have updated the CVE repo (github/Proteas/apple-cve) to include all (maybe) of the data from the past 25 years, covering 2001 to 2026. :)
I’m amazed that the phpcodz project (github.com/80vul/phpcodz) from 10+ years ago has now been turned into a Skill. The LLM era is keeping the classics alive!
wp2root as promised :-)
blog.calif.io/p/the-wordpres…
Our chain is nothing fancy. Its value is in showing what real-world PHP hacking actually looks like, and how far AI has come. Cooking up a chain like this would usually take a skilled operator a few weeks; we did it with Codex in under an hour.
The slow part was writing this article, explaining each step for readers without that background. This is how hacking is becoming. You write clear English prose describing what to do, and the model does it. Feed our article to your favorite model and it will happily reproduce the whole chain.
But that raises a harder question. Directing the model takes knowing what is possible, and that knowing usually comes from having done the work yourself. If the model does the work from now on, where does it come from? We learned it the slow way, by hand, over years. How the next person learns it, once the slow way is optional, we honestly do not know.
We do know one thing, though. You can outsource the hacking, but not the understanding.
heige retweeted
Tsecbench上线20天累计3196个ai跑分5262次,日均承载跑分任务260次,收到大家不少反馈。我们也在不断完善试题集和功能,今天云端托管跑分模式正式上线,欢迎大家来测试。
tsecbench.zc.tencent.com/
heige retweeted
Releasing the model weights and technical report of Kimi K3.
Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.
New model architecture: 2.5x the intelligence per unit of compute, not just more params.
Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.
Model weights: huggingface.co/moonshotai/Ki…
Tech report: github.com/MoonshotAI/Kimi-K…
Tech blog: kimi.com/blog/kimi-k3



































