Stop making every agent start from zero. Independent by default. Collaborative when it counts. Trusted because it’s provenanced. Powered by @origin_trail $TRAC.
Joined September 2015
- Tweets10.7K
- Following228
- Followers1.9K
- Likes40.3K
Pinned Tweet
✅ Data privacy and sovereignty
✅ Fights AI harm with verifiable context and knowledge
✅ 33% APR from paying customers using the product
✅ Zero inflation since 2018, fully circulating
@origin_trail $TRAC Is finally getting the attention it deserves
My first in a new series of three articles that introduces and educates the masses about the OriginTrail DKG, the native $TRAC token at its center, and a new startup that is using both to actively combat the dark side of AI.
Enjoy.
$TRAC used to have one job: pay to publish knowledge on the @origin_trail DKG.
Now nodes on Base are testing a second one, buying and selling queries and inference in TRAC. And @umanitek Guardian inside Claude lets people create verifiable evidence of fakes on the DKG.
More ways to use it, more demand.
BRX retweeted
AI shouldn’t just be used to boost productivity or make money. It must also protect us from all the risks online and on-chain.
With Guardian by @Umanitek, now available on Claude, over 100 million users will be able to create verifiable proofs using @origin_trail DKG.
Fake accounts, deepfakes, scams... Now more than ever, we need projects like @umanitek that aim to change the way the internet works and enable everyone to use it safely.
This makes me very optimistic about technology. After many years on-chain and partnerships with Web2 and institutional players, it is now the general public who will have access to the DKG @origin_trail revolution.
Find out here
claude.ai/directory/umanitek…
Guardian inside Claude is the part people are sleeping on.
Scammers already use AI to spin up fake accounts at scale, so the fix has to work at that scale too. Claude finds the impersonator, the evidence gets anchored on the @origin_trail DKG, then the takedown goes out with proof attached.
That anchoring runs on $TRAC 🛡️
Big news from @umanitek: Guardian now runs on the Claude AI.
A major milestone for this ecosystem, but also for users. It will be easier than ever to detect deepfakes, harmful content or scams.
The plugin is already available on the Claude Directory and open to the community.
More information below, including details on the links with @origin_trail and its DKG
BRX retweeted
Your Claude can now protect your name.
Guardian works with @claudeai and its 100M+ monthly users. Ask Claude to:
→ find deepfakes, fake accounts and harmful content
→ create verifiable evidence on the @origin_trail DKG
→ request takedowns
Setup takes under a minute.
BRX retweeted
🛡️ With @umanitek Guardian, Claude's 100M+ monthly users can now create verifiable evidence on Decentralized Knowledge Graph (DKG).
Ask @claudeai to find who's impersonating you. The evidence pack goes on the DKG first, then the takedown request.
Trust the source.
Your Claude can now protect your name.
Guardian works with @claudeai and its 100M+ monthly users. Ask Claude to:
→ find deepfakes, fake accounts and harmful content
→ create verifiable evidence on the @origin_trail DKG
→ request takedowns
Setup takes under a minute.
BRX retweeted
The @umanitek Guardian protects people from AI harms - traverses the social web, builds a giant real-time knowledge graph of findings, detects deepfakes, fake accounts etc and protects you by reporting them while you sleep. Now available in @claudeai too
Built on @origin_trail
Your Claude can now protect your name.
Guardian works with @claudeai and its 100M+ monthly users. Ask Claude to:
→ find deepfakes, fake accounts and harmful content
→ create verifiable evidence on the @origin_trail DKG
→ request takedowns
Setup takes under a minute.
BRX retweeted
"Code using code to transact" is the right frame. Once AI agents move money onchain, they need context they can verify: where the data came from and who stands behind it. That's what the @origin_trail Decentralized Knowledge Graph and $TRAC are built for. origintrail.io
BRX retweeted
Yo @cz_binance, time to upgrade all devices @Binance employees use with @umanitek Blackbox - live AI threat scanning against millions of potential threats signaled via @origin_trail DKG.
We always keep our promises @Teknium, Agent #Blackbox just joined the #Hermes army 🪽
Our threat plugin is submitted to @NousResearch's Hermes Agent plugin catalog. Can't wait to see it added: every Hermes agent, one install from protection.
Threat intel on @origin_trail's #DKG.
github.com/NousResearch/herm…
Simple flow that matters. You ask Claude who's impersonating you, Guardian builds the evidence pack, it gets anchored on the @origin_trail DKG, and only then does the takedown request go out.
So the platform isn't just taking your word for it. It gets a record anyone can check later, even if the fake deletes and comes back under a new name. Every one of those records lands on the DKG, which is where $TRAC gets used.
BRX retweeted
🗞️ Umanitek (@umanitek) Guardian is now in the Claude (@claudeai) plugin directory — OriginTrail (@origin_trail) and Umanitek say Claude reaches 100M+ monthly users.
The takedown starts inside the chat window.
What Went Live This Morning
Umanitek posted "Your Claude can now protect your name." Ask Claude to find deepfakes, fake accounts and harmful content, create verifiable evidence on the DKG (Decentralized Knowledge Graph), and request takedowns. Umanitek says setup takes under a minute, and the directory listing's own line is "file platform takedowns without leaving Claude."
The Walkthrough (Umanitek's October 22 demo)
Guardian connects to Claude through MCP (Model Context Protocol). Asked to investigate YouTube (@YouTube) accounts impersonating a protected entity, Claude pulled up threats Guardian had identified and organized on the DKG. Told to bundle the impersonating content into evidence packs, it did, and the packs showed up in Guardian's dashboard. Claude asked which accounts to include before submitting takedown requests through Guardian.
Evidence First, Takedown Second
That order is the mechanic worth learning. An evidence pack preserves what a threat was and where it appeared, anchored on the DKG, so the proof exists before the takedown request goes out. And one Guardian account can connect to both assistants: Umanitek's Oct 23 write-up says a pack created through Claude can be opened later through ChatGPT (@ChatGPT), synchronized through the DKG.
The Second Post: A Collective Immune System
OriginTrail's earlier post tied this week's Binance (@binance) incident to a "collective immune system." That is Guardian Agent Blackbox's job, and it is a different one: Blackbox watches what an AI agent itself does, its commands, files and installs, checking planned actions against threat intelligence on the DKG before they run. It is free and open source (MIT licence). Sharing is opt-in. Rules from Umanitek's verified graph can block in block mode, while community reports only flag.
Fine Print
Guardian and Blackbox are two Umanitek products built on the DKG: Guardian defends people and brands against impersonation and deepfakes; Blackbox defends the agents doing the work. The Claude reach figure is the teams', stated as they state it, and Umanitek says Guardian "works with" Claude. In Umanitek's own framing, Guardian identifies the threats — Claude is the interface you act through.
In Plain English
Claude can now gather the proof on an impersonator, anchor it on the DKG, and start the takedown, all in one chat.
Why This Is Bullish
Trust tools live or die on distribution. Guardian just moved into an assistant with a mass audience, and every evidence pack it anchors is work for the DKG.
My Take
Everyone is racing to give agents more power. This is the other race: a receipts drawer inside the same chat — proof first, takedown second, no dashboard detour.
If your assistant could file the takedown for you, would you let it act on its own — or always review the evidence pack first?
$TRAC
You Ask, I Research. Questions welcome.
Follow @BullsCryptolife for free sourced crypto breakdowns.
Sources in the first comment ↓
Honest check from Luku. Guardian can't force a takedown, the platform still decides. What it does is lock the evidence first, timestamped on the @origin_trail DKG, so if a fake gets reported, ignored and comes back, the record is still there to point at. $TRAC
Guardian is now listed in Claude's directory. I checked the entry: it can review impersonators, preserve timestamped evidence on the OriginTrail DKG and submit takedown requests from your Guardian account.
The platform still decides whether content comes down. It's a community connector, not an Anthropic-verified one.
claude.ai/directory/umanitek…
First contact was the feed algorithm fighting for our attention. Second contact is AI fighting for our trust, and that one is harder to spot.
That's why umanitek is building on @origin_trail. Guardian finds deepfakes and impersonators, Blackbox guards agents with threat intel from the DKG. Proof you can check, not vibes. $TRAC
Humanity has already had its first contact with AI, and we lost.
On the @ForbesParaguay podcast, our chairman @ChrisRynning explains that it wasn’t robots. It was the social media algorithm.
Now comes second contact: AI competing not just for our attention, but for our trust.
“We’re the mice in the biggest social experiment in history.” 🎙️
BRX retweeted
One hidden instruction is all it takes to lose everything.
Now one line is all it takes to protect yourself.
👉 - curl -fsSL blackbox.umanitek.ai | bash
Try it today ✅
Tomaz ran Guardian inside Claude for a week: takedown decisions, scheduled tasks, even counter-intel plans, all using context that carries across sessions. That's the @origin_trail DKG at work. Findings get stored as verifiable evidence instead of vanishing when the chat ends.
I've been using Guardian with @claudeai for the last week and it works GREAT!
Used Guardian tools in different sessions where existing context was useful to make evidence and takedown decisions, I've set up scheduled tasks using Guardian data, ran serious tasks like designing counter-intelligence plans and less serious ones creating an in-chat game to browse threats.
@umanitek + @origin_trail + @claudeai,
set up in 1 min.
Clearest Blackbox explainer yet. Free, open source, runs on your own machine and checks every action your AI agent tries before it runs. Poisoned packages, phishing links, prompt injection, leaked keys. The threat list comes from a shared graph on the @origin_trail DKG. $TRAC
Good question, and worth being precise about, especially with schools' data involved.
Yes, Blackbox is free: it's open source (MIT licence) and runs on your own machine.
What it does: it protects AI agents. It sits between an AI agent and the computer it works on, and checks every action the agent tries before it runs.
It stops the agent from:
-> Installing packages known to be malicious (fake npm or pip libraries that steal keys)
-> Opening sites and addresses known to be used for phishing or malware
-> Following hidden instructions planted in web pages or files (prompt injection)
-> Sending your secrets off the machine: .env files, API keys (including your Anthropic key), SSH keys
The list of known threats comes from a shared, verified threat graph, but every check happens on your machine, and your code and prompts never leave it.
What it doesn't do: it isn't a firewall or security scanner for your web app's servers, and it doesn't sit in the path of your app's calls to Anthropic's API.
Where it fits for you: if you or your developers use an AI agent to build, maintain or run parts of the app (the Hermes agent or OpenClaw today), Blackbox stops that agent from becoming the way in, whether by installing a poisoned package or by leaking the keys that unlock your app and your school data.
How easy: on Mac or Linux it's one command:
curl -fsSL blackbox.umanitek.ai | bash
It sets everything up, starts in watch-only mode so nothing breaks, and gives you a local dashboard showing what it caught. You switch on blocking when you're comfortable.
BRX retweeted
What if your AI agent's security worked like an immune system? Flag a malicious package or prompt injection once, and every agent connected to shared, tamper-evident memory on the @origin_trail DKG knows about it. That's Agent Blackbox from @umanitek, free to use. $TRAC
Guardian just moved into Claude. That matters because nobody remembers to open a separate tool when a fake account of them pops up, but plenty of people already have Claude open all day.
Ask it to find deepfakes or impersonators, log the evidence on the @origin_trail DKG so it can be checked later, then file the takedown. Same @umanitek team behind Blackbox for agent security. $TRAC
Your Claude can now protect your name.
Guardian works with @claudeai and its 100M+ monthly users. Ask Claude to:
→ find deepfakes, fake accounts and harmful content
→ create verifiable evidence on the @origin_trail DKG
→ request takedowns
Setup takes under a minute.