Sage is now live on NEAR AI's confidential infrastucture. Intel TDX and NVIDIA confidential computing, plus cryptographic attestation, give teams a verifiable boundary for classifying and deciding on sensitive enterprise data.

New on our Frontier Red Team blog: using Claude to formally verify software at large scale. Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time). With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks. Read our research in full here:

Today: I’m joining @ilblackdragon (Illia Polosukhin) at Open Source AI Summit SF for a fireside chat on open models’ benefits and risks. Also speaking: @NorvigPeter and @JeanKossaifi (Nvidia). Kickoff: 2:30 PM PT | Our chat: 2:40 PM PT Come by, say hi and enjoy the event! #OpenSourceAIWeek# #OpenIntelligence#

NEW LAUNCH Introducing NEAR Business: a confidential multisig treasury for crypto-native teams. Pay teams and vendors, swap, and allocate capital across 35+ chains, with every move approved by your signers. Built on NEAR Intents. How it works ↓

Job Board: Call-for-Me Booking Agent Build an AI agent that calls restaurants & local businesses to book tables/appointments, then sends back a clear confirmation. You keep ownership and earn every time someone hires it on Agent Market. Bid now:

NEW EPISODE with @NEARProtocol co-founder and co-author of the 2017 transformer paper @ilblackdragon at @token2049: "We want agents to hold as much NEAR as possible." - what NEAR is today and where it wants to be by 2030 - why crypto is still too hard for most people - why anyone you pay can see your whole balance - how NEAR AI keeps AI requests private - AI agents hiring other AI agents and paying per job - bank transfers that land as stablecoins - how the NEAR token earns from AI agents - why an AI agent needs a spending limit - why traditional stocks may miss AI medicine THE VISION: "The vision always been AI is the front end and blockchain markets is the back end... Back when it started in 2017 as an AI company... And then we realized we needed the back end economic layer and we started working on blockchain." PRIVACY: "The real bottleneck I always believed has been privacy... when you pay people they should be able to see how much money you have But that's not normal, right? And if you talk to anyone outside of crypto, they will be like, this is insane." MACHINES: "I think the reality is blockchain always was built for machines. And so I think where the next wave of this is coming is agents now becoming the primary user." AGENTS FOR HIRE: "Your GitHub or whatever can hire my agent to review it. My agent will run all your data. I actually will not see it because everything run privately... And then you just pay for the outcome, right?" THE TOKEN: "We want agents to hold as much NEAR as possible... This is like your store of the AI value right? So this and then on the other side all of the Intents infrastructure as it generates revenue buys back NEAR as well." SPENDING LIMITS: "You hire a new person, you're probably not going to give them full access to all the bank accounts of the company, or your personal account, right? You're going to give them, okay, here's a spending limit... You cannot go and buy a random meme coin." AI MEDICINE: "This is a bit controversial, but I think traditional stocks will probably not capture this as well because the traditional stocks actually, they are designed to be more like centralizing and average out thing... I do think this can actually be really well done on kind of decentralized." Timestamps: 0:00 Coming up 1:02 NEAR builds AI and private payments 5:14 Crypto is still too hard 9:11 NEAR AI keeps AI requests private 14:11 AI agents hiring other AI agents 20:10 Bank transfers straight into crypto 21:56 Renting out an AI code reviewer 24:39 NEAR's token earns from AI agents 26:26 Spending limits for AI agents 29:17 Sending crypto like a bank transfer 31:01 AI medicine made for one person

This has been part of our AttackBench and effectively all agent harnesses are susceptible to some form of these attacks. Including a way to run arbitrary code on the machine with the agent. Stay safe out there!