Pinned Tweet
Give your agent a card and let it shop for you, according to your rules and preferences, with each payment authorized in real time on the Fast network.
Cards are coming to Fast.
We’re partnering with @PulsarMoneyApp to bring card payments into the Fast ecosystem.
From intent to purchase, without breaking the flow.
Musab retweeted
What will it take to formally verify the Linux kernel?
And how soon are we going to see Rust and Lean models of it?
Here's what our experience leads us to believe 👇
Musab retweeted
This year's Meridian is shaping up to be the best one yet.
If rooftop views over Lisbon, delicious food & drinks, and a guest list of @BuildOnStellar "who's-who" speak to your heart, then this is a can't-miss event!
Join us and our friends from @XoxnoNetwork & @theahaco in 2 weeks, spots are limited👇
Our RWA x DeFi event during Meridian in Lisbon brings together friends, partners, and industry leaders.
We're pleased to welcome @rv_inc as our security partner!
@rv_inc is auditing XOXNO's lending protocol on @StellarOrg, bringing deep expertise in formal methods and smart contract security.
Join us to discuss security on Stellar/Soroban and the challenges of building secure DeFi infrastructure.
RSVP: luma.com/rqt96zxr
Musab retweeted
As part of our Linux kernel formal modeling work, we found a policy bypass in the io_uring sandbox filters.
The fix was applied by Jens Axboe on the same day, now in mainline & queued for stable.
See our patch below, technical breakdown to come next.
lore.kernel.org/io-uring/202…
Musab retweeted
@RosuGrigore on AI:AM, on safe AI: it's "impossible to prove the weights correct."
But you can write safety specs — this should not happen — and prove the monitor that keeps the system inside the safety envelope. A police system around the AI.
xitter.cf/i/broadcasts/1DxleVgbP…
Musab retweeted
Super excited to see K formal semantics of Rust's MIR enter the big arena of Rust formal tools announced at @rustconf 2026. Congrats @rv_inc!
I take this opportunity to recall the unique advantages of having an explicit semantics of a PL:
1. It is open to the scrutiny of the community, as a standalone and self-contained artifact, so a higher chance to get it right;
2. It can be tested against a variety of tools (interpreters, compilers, model checkers, symbolic execution engines, deductive program verifiers) at no additional effort, so a higher chance to get it right.
KMIR is open source. Abuse it, please, and help others by reporting issues and pushing fixes.
KMIR, our formal verification tool for Rust, is now an accepted tool in the "Verifying the Rust Standard Library" challenge, as announced at @rustconf 2026!
We first submitted KMIR for consideration in April 2025, when it was still early. Over the following months, we expanded its Rust semantics and infrastructure while putting it to work on real-world verification.
Earlier this year @danielkcumming brought a much stronger KMIR back to the challenge.
That work got it accepted, and Daniel has since joined the committee that reviews community submissions.
Much of Rust's standard library rests on unsafe code, and each independent tool that checks it gives developers more confidence in what they build on.
KMIR now joins Kani, VeriFast, ESBMC, Flux, and RAPx in that work 🦀
youtu.be/rOR6JBsSnck?t=452
To explore KMIR, check out its GitHub repo 👇
Musab retweeted
Formally verifying Linux always exposes something interesting. Usually bugs.
Bugs in the kernel itself, but also in the formal verification tools.
If you do formal verification or develop formal verification tools and look for a challenge, Linux is your garden.
Join @rv_inc in their effort to have a formally verified Linux, for the benefit of all.
What happened when our engineers kept poking under the hood of the Linux kernel's isolation?
We proved 89 theorems in Lean 4, the spec exposed a silent kernel bug, and we crashed Aeneas.
The technical breakdown, from the proof to the kernel patch and the Aeneas fix, below 👇
runtimeverification.com/blog…
Musab retweeted
yesterday i’ve explored the @fastxyz ecosystem (ex PiSquared)
> set up a wallet
> minted my ID
> explored their shopping agent
> explored their AllSet app
what i’ve found most interesting are their ID system and the AllSet app, why?
- the ID system let’s you link your X, github and website to your FAST account, but the coolest thing is that you can sign a paper/work you’ve done, so that the owner is verified without exposing his real name
- the AllSet app let’s you deposit to and withdraw from FAST to and from some EVM chains, tek is cool because it’s not a bridge, it acts like a liquidity layer (new type of tek here)
this new tek inspired me to build a USDC bridge with 0 slippage that should work faster than other 0 slippage bridges
it’s not something i’ll build to make it a public product but i’m curious to see if it works and i’ll make it open-source, will keep you updated
One identity across the Fast Network. Linking this account to my Fast ID:
id.fast.xyz/gesoos.xbt?fast-…
Musab retweeted
Phenomenal questions, so let's unpack them together.
First the serious answer, then a more pragmatic take from Everett Hildenbrandt, our CEO, to summarize it in 2 words.
1. Exhaustively, almost never (but that's fine).
Simply speaking, you leverage what comes from proving the handful of invariants that would be catastrophic if they broke. Then, you can compose those guarantees, module by module, instead of specifying the whole system at once.
2. As far as better properties, if we start from asking 'what must never happen?' rather than 'what should happen?' then we're on the right path!
Fuzzing (or property-based tests) are also a cheap way to sharpen them before you commit to a proof.
With AI now handling more of the proof work in tools like Lean, Dafny and Verus, the bottleneck has shifted to the spec itself, but that raises a perhaps more interesting question: when code and proofs both get cheap to generate, does writing the spec become the new "coding", aka stating precisely what correct means?
But if you ask Everett, you'll get a half-jokingly "get good", as the simplest answer on how to write better properties in 2026 😅
Musab retweeted
BIG NEWS: Fast → @PulsarMoneyApp.
In this new phase of our partnership, your agent can move the money. Pulsar lets you spend it. Our Fast-branded cards are next.
Deposit from Fast and get your free card from Pulsar, and get up to 100% cashback.
Code: BFAST
Live today, FAST on @PulsarMoneyApp.
Sign-up → deposit from Fast → start spending.
✅ up to 100% cashback, free card, free USD/EUR accounts, and more perks coming for Fast users.
Limited invite code: BFAST
Get Pulsar → pulsar.money/download
Musab retweeted
We have already been working with @fastxyz on agentic payments, helping agents take a purchase from intent through to checkout on @Arc.
Now we are building on that collaboration to bring more ways to fund and move money into your pocket.
Pulsar and @fastxyz are partnering to make agentic payments actually feel agentic.
A seamless experience from intent to checkout, co-built with @crossmint agentic stack
Musab retweeted
Your money is about to move Fast. 👀
We are bringing card top-ups, deposits, and withdrawals to Pulsar with @fastxyz, connecting the Fast network to your everyday payments.
Musab retweeted
Live today, FAST on @PulsarMoneyApp.
Sign-up → deposit from Fast → start spending.
✅ up to 100% cashback, free card, free USD/EUR accounts, and more perks coming for Fast users.
Limited invite code: BFAST
Get Pulsar → pulsar.money/download
Your money is about to move Fast. 👀
We are bringing card top-ups, deposits, and withdrawals to Pulsar with @fastxyz, connecting the Fast network to your everyday payments.
Musab retweeted
Deep dive into Formal Verification (FV) in the era of AI.
Formal Verification (FV) has become, again, a hot topic. But sadly, AI folks live in the past. I often hear (from those who spent time to read about FV):
➡️ "FV means VC-generation using an abstract logic of your PL and proving the VCs using SMT"; or
➡️ "FV means translating your program to Lean/Rocq/Dafny/Boogie and proving it there"
This is what FV was 15 years in the past!
Modern FV builds upon complete and well-tested formal semantics of real programming languages as ground truth or trust base, without having to trust translators or abstract/convenient semantics.
If you are interested in large-scale FV in the context of AI, reach out to @rv_inc. Especially if you have domain-specific languages or applications.
RV has the most mature FV technology for AI generated code, built over more than 25 years, transitioned from a UIUC research lab in 2010.
Musab retweeted
founders should post more, post often, post when you feel the inspiration. every tweet is like a free lottery submission for attention
Musab retweeted
Just ran into this article by @KaiyuYang4: yangky11.github.io/blog/ai-a….
Good read, strongly recommend it. Especially to those who think that "I proved it in Lean" addresses all the correctness/safety/security concerns.
Besides functional correctness of AI generated code itself, Runtime Verification (RV) is another promising approach: you do not prove the code or the agent correct, but you provably correctly do not let it go wrong. By generating provably correct monitors that steer the agent to stay within its specified safety envelope.
I discussed this recently with @Jacob_Tsimerman, in the context of AI Safety.
RV is a low-hanging fruit, imo. @rv_inc
Musab retweeted
Two insights here for the coding & formal verification folks, besides due congrats to @typesafeai for Jev:
1. Frontier models can and will be outperformed by specialized models when it gets to generating provably correct code.
2. Increasing evidence that humans will soon not look at code anymore, that they will judge higher-level specs not code.
