I had early access to Fable 5.2 for a bit & I think it crosses the line into being a sort of Jagged AGI. Here, I have it launch a business, do academic research, hack hardware, build 3D models & more. I think it is a good illustration of limits & powers: https://t.co/QiR0IFPrZY
@vladtenev@EggSlonker69 I am building Cryptographic Identity for Agents down to the computation layer of the agents. What ensures I am talking to the same model? This is what I am building ontop of @agentwormhole agent security layer
๐จ WARNING: @fomo wallets can be drained through malicious browser bookmarks, even with 2FA enabled.
Yesterday a friend of mine had around $36k drained from his FOMO wallet. I wanted to make a post about this because I don't think a lot of people know about this type of exploit.
Scammers are creating what look like useful crypto tools, such as wallet trackers, Twitter trackers, smart money trackers, etc.
The trick is that they can disguise malicious JavaScript as a bookmark or shortcut for these tools. So when you think you're saving a useful trading tool, you could actually be saving malicious code.
They'll get you to drag something into your bookmarks bar or save what looks like a normal shortcut.
The problem is that some of these aren't actually normal bookmarks. They're called bookmarklets, and they contain JavaScript that runs directly on whatever website you're using when you click them.
If you're logged into something like FOMO, that code can potentially interact with your authenticated session.
One possible attack is that the script waits until you've completed 2FA and then abuses the authenticated session to initiate a private key export. Once the attacker has your keys, they can drain your wallet.
What's even more concerning is that FOMO/Privy logs can show these actions coming from your own IP, device, and browser session.
So from their end, it can look like you were the one who authorized everything, even if you weren't.
I want to make something very clear for anyone using browser-based trading wallets:
DO NOT save or execute random bookmarklets that people advertise as trading tools.
Bookmarking a normal website is fine. The danger is saving and running bookmarks containing executable code, especially ones starting with javascript:.
If someone tells you to paste code into a bookmark or drag a random tool into your bookmarks bar, be extremely careful.
And don't assume having 2FA enabled makes you immune to this.
Just wanted to put this out there because I know a lot of people in the trenches use these kinds of tools every day, and I don't want to see someone else lose their money over something they didn't even realize was dangerous.
@Wen2daMoon The design Iโm working toward derives the neural weights from the agentโs key and verifies execution against those weights. For a specific response, that inference must be covered by the proof too. Key-to-weights binding alone wouldnโt prove which model generated it.
Who is building Cryptographic Identity for All Agents, where its neural net weights are derived from its key.
This gives us identity on the model its being ran on, can we trust agents in knowing that agent is still using the same model?
How can we guarantee the agent I am calling is switching to Gpt-6 Astra to Opus 5.5.
I have a solution I been working on for months.
@VitalikButerin just opened Devcon and set the tone for the week.
A formally verified OS, open silicon, self-sovereign healthtech, free discourse, private communication, trustworthy AI โ these are all part of Ethereum's mission.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! ๐งต
Blog: https://t.co/bOWCcFhCVq
Weekly update ๐ชฑ This week we crossed 1,000,000 total scans across our security layers.
Integrations live on Robinhood Chain:
- @RaetoCC: our x402 guard is built into Raeto, so agent payments get checked before they go out
- @gloamtrade: private agent payments on Gloam get checked by us before they're sent
- @routedotfun: Route's screener shows our security checks, and Wormbook's swaps run through Route
- @Griftwrld: every x402 payment the agents in San Verano make gets checked before it executes
- @Borromarkets: x402 bills agents pay with Borro credit get checked for hidden instructions
- @ChopLogsApp: x402 helps stop failed or mismatched payments before funds move.
We also posted a new case study on the x402 SDK flaws and updated the revenue tracker on https://t.co/arvj7j05mg.
Wormbook's second buyback and burn is done, and 4.37% of the $WORM supply is burned so far.
More in the pipeline and the worm stays long.
Agreed. We've spent the last few days on exactly this, and it's live now.
wormhole-x402 0.10.0 puts the spend limit at the signer, where no 402 can move it. A cap per payment, allowed assets and payees, a running total.
Sign-in requests are now bound to the site the agent called, too.
npm i [email protected]
https://t.co/9JPjENKb8O
We crossed 1,000,000 total scans this week.
Over a million checks on payments, tokens and agent inputs, across all of our security layers.
We've also updated the revenue tracker on https://t.co/5y8ERDQV0C with more details.
Next stop: 10 million.