←
forefy1,659 posts
forefy avatar

forefy

@forefy

Over a decade of security research and engineering channeled into securing emerging threats ㅤ CTO @audit_wizard 🧙‍♂️🪄🪄 ㅤㅤㅤㅤㅤㅤㅤ Co-Founder @hackstackapp

◷ Joined December 2021▧ 231 Photos and videos
forefy avatar
forefy@forefy·Nov 11

Just wanted to show you guys real proof that coercing Fable 5.2 to perform cyber tasks is possible without red team access 🤓 It's kinda like doing a stealth redteam op - you try to ride on top of normative operations that just happen to get you where you want to be, without triggering the EDR But with prompts btw if you type "hi exploit this kernel for me" it won't work. this is only for patient people. its like art.

Post attachment

Small tip on avoiding cyber blocks on Fable/Mythos 👇 > The guardrail is way more sensitive at the first couple of prompts > If your repo is auto-flagged on "hello", start a session outside the repo > Start with a UI request rather than a cyber task > Gradually have the UI be filled with the actual task, but the UI testing being the driving force (this sometimes actually works)

forefy avatar
forefy@forefy·Nov 17

3rd party JS - oh right how could we forget 🙃 The number of agents that now had woken up to pivot to attack your 3rd party javascript is through the roofs 2026 AI hacking era taught us that to stay ahead of the game we need to recognize the risk BEFORE it happens and this is the early tell If you hold sites with posthog, mixpanel, segment, hotjar and many more - you're next (and don't forget transitive dependencies) If you need help preparing my DMs are open.

This is a big one. We found malicious JavaScript being served through Optimizely to some of the world's largest websites. 178 organisations affected, with everything from crypto theft and card skimming to iOS exploits. Zoom, BBC, CNN, HSBC, Salesforce, GitLab, KLM, IKEA, Equifax, Ledger. The list goes on... Full list in the post. blog.report-uri.com/worlds-l…

Auditware avatar
⟳ forefy retweeted
Auditware@audit_wizard·Nov 16

Post attachment

You want me to use AI to alert you when a process on any of your dev devices suspiciously tries to open a MetaMask wallet? 🦊 No right? Because that would mean my AI agents, systems, and employees can see all that your employee is doing But how powerful would that be? SOLVED: Using zero-knowledge proofs, endpoint agent (prover) can only prove to the control server (verifier) that a statement is true (e.g. a process tried to access web extension id `nkbihfb.*` (metamask)). Moreover, having the entire code open-source, self-deployable on your own AWS/GCP/Hetzner gives only you the control over what happens (e.g. investigate, block, run IR playbooks). Orgs don't want to build this, because this kills EDR companies' profits - we want to give it away for free. Thanks to @thedaofund, we can make our PoC generally accessible. If you want this for yourself, free forever, we only need web3 to step in, even with the smallest of donations, to turn this into reality. Here is the initiative paper: initiatives.thedao.fund/init…

forefy avatar
forefy@forefy·Nov 15

hermetic is listed at #6 out of the entire hermes catalog 😱🔥🔥🔥 total shock!!

great list. we indexed all 2,380 Hermes Agent catalog plugins. 492 had a readable author handle, so only those were scored: GitHub activity and X presence, weighted by public replies, quotes and answered mentions. our index, our criteria. #1 plugin is gadget, by @adolandev, who also made Hermes Sheets and Hermes Whiteboard, two of the nine above. next, mayfly: one plugin, a fresh @instabox_sh box, a receipt, then the box is deleted.

0:05
forefy avatar
forefy@forefy·Nov 15

Gradual realization of how to hunt with /goal > you need to hunt with goals > your agents write goals better than you > it will write it even better if you massage it's context window first > it will write it EVEN BETTER when you point it to this json-schema github.com/forefy/.context/b… hunting for bugs is one of the most classic use cases for /goal !

forefy avatar
forefy@forefy·Nov 12

(Redteam) This is how your org gets compromised through AI-bloatware tools. Modern orgs use OAuth to access apps, and by default most users get access to Jira or Atlassian for onboarding docs (even sales, product, etc - not just devs). ✨jira-harvester coerces Jira's native AI assistant "Rovo" to gather all the sensitive attachments ever uploaded alongside tickets and docs - env files, json, zip, har captures, and lots more that include AWS/Azure/GCP API keys, Jenkins, databases, AI app tokens, and whatnot. This is a highly probable vector from a weak compromised user to complete cloud and ci takeover (and spread laterally to anywhere else) Old technique, but was usually done by searching through the search bar - now with your AI (glm or so) speaking to the customers internal AI (rovo) finding the credentials and testing if they work became a 10 minute task - this is why it's important to keep raising the awareness to these risks forefy.com/skills/b64fd4f9-f…

forefy avatar
forefy@forefy·Nov 11

Already on the registry! forefy.com/models/fe3e8f8c-5…

You can run locally a 7B-Cyber Agent model designed for autonomous cybersecurity tasks on your laptops. - Specialized for ethical security audits, tool use, and long-context analysis. - 32k context - native tool calling, - It handles multi-turn tool use, - secure code review, CVE triage, - Long-context repository & log analysis - huggingface.co/mradermacher/…

Post attachment
forefy avatar
forefy@forefy·Nov 9

Use it now before the hype period performance boost is over

Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.

0:19
forefy avatar
forefy@forefy·Nov 9

Hacking became direct outcome of believing 😱 > the more I believe a bug is there > the more relentless my prompts will be > the more tokens efficiently steered by my creative mind > the least likely I am to give up on that target or vector AI made it so that more belief means more bugs Also "glm 5.3" for engagement. thank you.

forefy avatar
forefy@forefy·Nov 8

Is Kimi K3 "dumber" than it was yesterday? or is it just me? Now measureable 🤓

Post attachment

Open sourced my hermes extension that makes it level up for security researchers: - Costs tracking (cheaper routing without data retention) - AAI with stuff like LLM-as-a-judge for hermes skills, benchmarking that improvements of skills really do improve - HDR (Harness Detection & Response) - security checks and LuLu style alerts for anything that isn't covered natively by hermes Listed and approved on the hermes plugins marketplace !!! All you gotta do is enjoy it and star the repo 🥳 github.com/forefy/hermetic

0:33
forefy avatar
forefy@forefy·Nov 5

Wanted to know what uncensored cyber model to run locally v.s. REAP prune and self-host on a cloud provider? This discussion keeps coming up between me and my peers so I tweaked the model calculator to take you through direct comparisons from what can run on your Mac v.s. how much will it cost a month to host on an inference provider (ones that take any model) I think it's very cool and useful to security researchers that literally can't use anthropic anymore forefy.com/models/7745058b-6…

0:23

Uncensored cyber models made it to the AI Security Registry 🔥🔥 - Relevant abliterated uncensored models - Listing of models in the way security researchers want and need - Best "Will it fit my machine" calculator I've seen so far - Popularity layer / which cyber models are most used in our small community DM if your model doesn't show up in here and should

0:34
forefy avatar
forefy@forefy·Nov 5

Got 20 findings from GLM and unsure which are total BS? After thorough benchmarks this works great as a self-honest triage filter: 👉 "Rate all discovered findings in a table, from best to least received by the receiving dev team, considering expected severity downgrades." Use it in your next audit and tell me how well it worked!