Ari Juels

Spotted at @consensus2026 👀 IC3's @sjbreck participating in a workshop with @stephendpalley from @BrownRudnickLLP on the legal and technical risks of agentic AI

An excellent account of hack of The DAO, authored by my former colleague @el33th4xor, recounting the key roles he and @phildaian played. Happily, it ended in champagne with plastic forks at an @initc3org bootcamp. Or did it end? fortune.com/2026/04/28/ten…

@KushalBabel Meet our judges 🫡 We have a preeminent lineup of judges that bring academic, VC, and founder experience to the hackathon

Missed kickoff day for the Shape Rotator hackathon? No problem 😎 Watch the full keynote and author presentations here: encodeclub.com/programmes/sha…

Last call for submissions to SBC! Deadline: Tomorrow, 13 March 2026. Visit sbc-conference.com/2026/ for details. @danboneh @Tim_Roughgarden @dawnsongtweets @CBRStanford @initc3org @BerkeleyRDI

Two weeks left to submit to SBC! Visit sbc-conference.com/2026/. @danboneh @Tim_Roughgarden @dawnsongtweets @CBRStanford @initc3org @BerkeleyRDI

14 papers. 2 weeks. 1 mission: Turning academic research into real-world products & applications Introducing Shape Rotator - a virtual hackathon and accelerator program by IC3, FlashbotsX, @theBBFund, @encodeclub, and The Convent. 🧵👇

Five weeks left to submit your blockchain research for presentation to the SBC community at Stanford (July 27-29). Submit by March 13: sbc26.hotcrp.com. @Tim_Roughgarden @danboneh @dawnsongtweets

My Smartcon 2025 talk on how oracles can support AI applications using Props (protected pipelines). The short clip here was just released by Chainlink. An excellent follow-on demo by Philipp Schindler ran Props on the Chainlink Runtime Engine (CRE). youtube.com/watch?v=9lZqio…

New from @ForbesDAssets: @jon_helgi reports on IC3 research, CoinAlgs, and interviews Professor @AriJuels about the counterparty risks our financial system may not be ready for:

A good TLDR for our paper on AI-powered collective investment schemes in Yahoo! Finance. Article here: finance.yahoo.com/news/popular-c… Paper here: arxiv.org/pdf/2601.00523

Do AI trading agents like @elizaOS and actually make you money? I wrote a paper with @initc3org to find out - studying both the profitability + fairness conditions of "collective investment algorithms". Here's the key takeaways in degen-speak: > the main intuition of the paper - the trading algorithms/patterns of the AI are either public, or they are private. if they are public, you get copytraded, sandwiched, and often lose money. if they are private, insiders can basically "rug" you. > we model AI trading agents formally as "collective investment algorithms" (CoinAlgs). you put your money to some CoinAlg C, and C will trade on a public address its assets according to some deterministic/randomized algo. such is the case for ElizaOS and other DeFi CoinAlgs + you can draw an analogy to tradfi roboadvisors (vanguard/schwab/wealthfront). all these move money bigly today (vanguard has $206B+, ai16z hit $2.7B MC) > in the "algorithm is private" case - we can model the spectrum of privacy by calculating the "distance" between the publicly available info + the real info of a CoinAlg. we prove that algorithmic privacy itself leads to potential actionable "insider alpha". this means an insider can extract alpha in proportion to how "private" the algo is. so if you invest in a trading DAO (eg. daos. fun) with a private algo - insiders can probably "rug" you with inside info. > on the flip side - in the "algo is transparent" case - you get sandwiched/copytraded publicly, resulting in diminished returns (compared to a private algo). we use a "one-shot stackelberg game" - fancyspeak for autopilot copytrade - to model this sandwich behavior. > we run a simulation of both copytrading ("strategy theft") and mev sandwich attacks on Uniswap data (link in paper pg. 14) for USDC/ETH and PEPE/ETH from 6/1/2024 to 7/26/2025. the main result is that the copytrader earns a lot more than you do. the mev sandwicher earns less than you do, but eats your margins, and is much more stable than your profits. for the private trading case, we can also construct a low-bit "leak" channel to get profitable alpha. > so where does this leave us? well firstly, if the bot algo isn't yours, you shouldn't use it (you will get rugged). and if you have alpha, keep it to yourself (you will get copytraded/sandwiched). for existing systems though, there may be some design workarounds - eg. you can have a TEE as a "random wrapper" where you have say 100 potential trades, and randomness in the TEE picks (so insider/adversary doesn't know) > some areas on further research in the pipeline: (1) an empirical data study on DeFi trading agents in-the-wild, and how they're doing in $$$ terms, (2) how to apply this to new markets - eg. prediction markets, vaults, perps that are more complex than AMM spot trading, (3) system design - how to weave in the TEE randomized wrapper, and what that might look like. AI-based DeFi trading is likely here to stay, so lets understand its risks properly. Paper link: arxiv.org/pdf/2601.00523

🧵 1/ New IC3 research reveals a fundamental dilemma in AI-powered trading systems and collective investment algorithms (CoinAlgs). Think robo-advisors, quant funds, and AI DAOs like @ai16zdao and @truth_terminal. We call it the "CoinAlg Bind":

In traditional finance, investor protections address the CoinAlg Bind. But CoinAlgs—especially AI-powered ones—are popping up in DeFi, software tools, etc., without these protections. Heuristic guardrails might help. Our paper proposes some and we are researching more. (3/3)

A CoinAlg can be *transparent*. But then it's vulnerable to frontrunning by outsiders. Or it can be at least partially *private*. But then insiders can secretly extract value from it, harming investors. We prove and study this dilemma, which we call the **CoinAlg Bind**. (2/3)

You don't share toothbrushes. Sharing AI-powered investment algorithms can also be a bad idea. Our new paper shows a fundamental and inescapable tradeoff in these algorithms, which we've dubbed CoinAlgs: Collective Investment Algorithms. CoinAlgs come in two flavors... (1/3)

A new report from @glxyresearch details the inner workings of @Chainlink: How the oracle network functions and why it matters, particularly as institutions are onboarded to public blockchains.

IC3 is entering a strategic partnership with Goethe University, an affiliate of ATHENE - Europe’s largest applied cybersecurity research center.

Managed to get my hands on a limited edition copy of USECASE Magazine, and it’s damn good!

@relyt29 @sarahalle_ @Haaroony @initc3org

Read the well-researched exposé by @scilla_alecci and Fergus Shiel: icij.org/investigations…

Who would have imagined that Kevin Spacey, Elon Musk's mother, and a 2025 supernatural Avengers-like action thriller would figure in the absurd afterlife of a smart contract we investigated? icij.org/investigations…