just-silence retweeted
I made a site that tracks open-source AI attack harnesses.
Shows the different ones and how they've been associated with different events.
offsecharnesses.ai/
just-silence retweeted
My @offensive_con Tokyo talk video youtube.com/watch?v=TzKsSvq9… and I have published slides in my blog naup.mygo.tw/slides/Offensiv…
Took the existing Pojun pentest platform and translated the entire codebase into english. Pojun is further developed and improved continuation of ARTEX. Has its own proxy tool, C2 and much more. Check the translated english version here.
github.com/xaccefy/pojun-eng…
just-silence retweeted
Splits has a self-hosted bug bounty most hunters miss: up to $50k for crits.
• Crit/High only, user-fund loss
• Solidity, ~4.6k lines, 3 packages
• Email your report within 24h of discovery
Rules: splits.org/docs/resources/se…
Repo: github.com/0xSplits/splits-c…
just-silence retweeted
One week until the #Prompt2Pwn CTF. 🚩
It starts Nov 19 at noon PT and runs for 48 hours: 30+ challenges and $5,000+ in prizes.
Get your team ready: p2p.nahamcon.com/
just-silence retweeted
RUN UNCENSORED DEEPSEEK V4 PRO ON A NORMAL 8GB PC 🤯
a 9B model distilled from DeepSeek V4 Pro reasoning, with almost no refusals.
what you get:
• Runs on 8GB RAM with CPU only
• Can run on an 8GB VRAM GPU
• Refusals dropped from 100/100 to 3/100
• 5.6GB download (Q4_K_M)
• Supports LM Studio and Ollama
• Apache 2.0 license, open weights
model:
huggingface.co/xtools-at/Dee…
GGUF:
huggingface.co/mradermacher/…
just-silence retweeted
You can run a Uncensored 8B cyber model locally for blue-team/red-team work based on this datasets.
- No refusals on legit security questions.
- No made-up CVEs.
- CyberMetric & CTI scores held,
- (detection engineering, threat intel, authorized testing).
- Local, 8-bit,
- huggingface.co/samwandosamti…
The largest PhD-level cybersecurity dataset is now available on free for training your own Cyber model.
Most comprehensive whole cybersecurity knowledge base ever built for AI training.
- 30+ domains
- 95 GB
- 507 files
- Red team, Blue team, Web security, AI/LLM security, Forensics, Compliance etc...
- Full PortSwigger Academy (all 31 topics), HackTheBox, MITRE ATT&CK, OWASP, real bug bounty patterns, and defensive playbooks SOC, detection engineering, forensics, NIST/CIS/MITRE D3FEND all in one place.
- AI security, cloud, container, IoT, crypto, reverse engineering
Perfect for fine-tuning the next generation of cybersecurity AI models.
i see Already being used to fine-tune models.
just-silence retweeted
300 free cloud hours a month for Claude Code and Codex. 3 months of Pro, worth $297.
boxes.dev gives each Claude Code or Codex session its own cloud computer, with your whole project already set up on it.
Like Codex Cloud, but for both agents. Close your laptop, they keep building, and you approve changes from your phone.
You get:
→ Standard machine: 4 vCPU / 8 GB
→ Bigger machines up to 8 vCPU / 32 GB (use hours faster)
→ 12 machines awake at once, 10 projects
How to claim:
1. Sign up at boxes.dev/launch-yc
2. Add a card. It won’t auto-renew or charge you for extra usage
3. Paste their setup prompt into Claude Code or Codex in your project folder
No card? Skip it and get 20 free hours.
We tested it ourselves (not sponsored): one prompt, and a full Next.js app was running in the cloud.
New teams only. Your own Claude/Codex login. No Windows app.
Got some amazing Solidity developers in my network that are looking for new Smart Contract opportunities right now
I can share CVs to people hiring for such roles in DMs, and in case you are a dev - feel free to reach out as well, I'd always help in any way I can🫡
just-silence retweeted
AladdinDAO runs its own bug bounty, not on any platform: Critical up to $500k, High $20k.
Scope: f(x) Protocol V1 and V2 (fxUSD), Concentrator and CLever. Solidity. Submit by email with a mainnet-fork PoC.
docs.aladdin.club/security-b…
just-silence retweeted
2\ Cause: a hardcoded 8k-char cap on the skill listing.
Fix: make it env-configurable (~45k chars ≈ 12k tokens) or do lazy discovery.
Skills github.com/wgpsec/AboutSecur…
just-silence retweeted
OpenCode isn’t free anymore if you want serious usage
so if you’re looking for the cheapest ways to run strong coding models, these are the options i’d check first:
1. DeepRelay, $4.99/month
2.5B input + 250M output tokens/month for DeepSeek V4.1 Flash off-peak. there’s also a $0.99 plan with 500M input + 50M output if your usage is tiny.
2. Mistral student, €6/month
the Vibe API key can also give you access to DeepSeek V4.1, GLM 5.3 and GLM 5.3 Flash, not only to Mistral models.
3. CommandCode, $1/month
probably the cheapest subscription here, but the catch is that it only works with their own harness.
4. Alibaba Cloud, $3-$6/month
cheap access to models like Qwen 3.8 Max and GLM if DeepSeek isn’t a requirement.
5. Muse / MiMo, $5-$6/month
both are worth checking if you mainly want cheap coding inference without jumping to a $10 to $20 subscription.
if your usage is really light, i’d still skip subscriptions and just use direct API credits.
that can easily be cheaper than paying monthly for capacity you never use.
just-silence retweeted
Beyond ARTEX: 45 Agentic Attack Frameworks, What They Hunt For, and How to Find Them
blog.gitguardian.com/ai-pent…
































