Houman Shadab

Cryptography for secure agents at @icme_labs Try now: Advised governments and banks | Fellow @CodeXStanford | Startup acquired by @Docusign

[Houman Shadab @HoumanShadab]

You can guarantee that agents follow plain language rules by using formal verification and cryptography.

That's what I will be discussing today at the New York Fed Innovation Conference: https://t.co/3qh0MSRDpl

More at https://t.co/Tuyatn5VNZ

[Houman Shadab @HoumanShadab]

I submitted privacy-preserving agent guardrails to Built in @AskVenice in part because of the project's strong tokenomics.

As an agent infra builder, I'm excited by Venice's Series A because a large purpose of the funds is to build for "several billion AI agents."

This strengthens the tokenomics: the new investors need a strong $VVV because agents lock it to mint $DIEM for agent compute.

Check out our end-to-end privacy for agent guardrails (@icme_labs) + inference:
https://t.co/k4TiV6Ya8i
https://t.co/XNb0L6xj6t @ErikVoorhees

[Houman Shadab @HoumanShadab]

Just submitted to Built in Venice: privacy-preserving guardrails for agents, composing @AskVenice E2EE inference with @icme_labs guardrail proofs. https://t.co/k4TiV6Ya8i

Speaking at the NY Fed on June 25 on why financial agents need privacy. https://t.co/KGDGnZeNfw

@ErikVoorhees @jesseproudman @sabrinaesaquino @willyogo @austinvirts @JonShapeShift $VVV

[Wyatt Benno @wyatt_benno]

Crypto people complain about short bear markets🐻.
I never hear AI people complain about decades long ice ages🥶.

Cryptography is *essential* for agentic commerce.
Maybe this creates the new, much longer lived summer?

Neurosymbolic AI + ZKML @ ICME Labs.

[Houman Shadab @HoumanShadab]

.@icme_labs maps to 7 of the 10 @OWASPGenAISec Top 10 risks for Agentic Applications.

Cryptographic enforcement that checks every agent action against your policy before it runs. Wrong action blocked. Verdict verifiable by anyone.

https://t.co/Tuyatn5VNZ

[Houman Shadab @HoumanShadab]

.@icme_labs maps to 7 of the 10 @OWASPGenAISec Top 10 risks for Agentic Applications.

Cryptographic enforcement that checks every agent action against your policy before it runs. Wrong action blocked. Verdict verifiable by anyone.

https://t.co/Tuyatn5VNZ

[Wyatt Benno @wyatt_benno]

Software is learning to prove itself; it's called 'vericoding'; Probably nothing 👀

[Houman Shadab @HoumanShadab]

You can guarantee that agents follow plain language rules by using formal verification and cryptography.

That's what I will be discussing today at the New York Fed Innovation Conference: https://t.co/3qh0MSRDpl

More at https://t.co/Tuyatn5VNZ

[Houman Shadab @HoumanShadab]

@jbrukh DePIN revival?

[Houman Shadab @HoumanShadab]

For agent governance, only @icme_labs provides:
1. Pre-action runtime enforcement
2. Formal verification
3. Portable zk receipt verifiable without trusting vendor, and
4. Cloud neutral

Deterministic enforcement of rules you wrote for agents. Every agent action, every time.

[Houman Shadab @HoumanShadab]

Agents can do financial compliance work but we need to make sure they follow sanctions rules exactly.

@icme_labs demo: agent screens payment, blocks sanctioned entity before payment settles, and drafts a report.

Agent does the work. Cryptography proves it did it correctly.

[Houman Shadab @HoumanShadab]

Agents can do financial compliance work but we need to make sure they follow sanctions rules exactly.

@icme_labs demo: agent screens payment, blocks sanctioned entity before payment settles, and drafts a report.

Agent does the work. Cryptography proves it did it correctly.

[Houman Shadab @HoumanShadab]

@AskVenice keeps the inference private.

@icme_labs keeps agent guardrails private too, and zk-proves each agent action.

Private agent inference, private agent guardrails. The whole $DIEM loop. https://t.co/4vg7XSZkPd

[Jon ShapeShift @JonShapeShift]

Inference is rapidly becoming the most important commodity and asset in the world.

Private inference delivered perpetually is the backbone of the base:0xacfe6019ed1a7dc6f7b508c02d1b04ec88cc21bf and base:0xf4d97f2da56e8c3098f3a8d538db630a2606a024 economy.

So much can be built on top of these primitives that will become crucial.

Tokenize the inference and build.

[Houman Shadab @HoumanShadab]

Agentic payments needs cryptographic security.

Every payment needs a tamper-proof receipt.

Agents that violate guardrails get blocked.

https://t.co/Tuyatn5VNZ

[Wyatt Benno @wyatt_benno]

They say "formal verification".
We say "What kind?" 👀

[Houman Shadab @HoumanShadab]

AI agent safety needs:
1. What: Deterministic guardrail enforcement, not AI judgment.
2. When: Before the action runs. Provable violations are blocked.
3. How: Fast cryptographic verification.
4. Who: Anyone can verify w/o seeing policy.
https://t.co/Tuyatn5VNZ

[Houman Shadab @HoumanShadab]

AI agent safety needs:
1. What: Deterministic guardrail enforcement, not AI judgment.
2. When: Before the action runs. Provable violations are blocked.
3. How: Fast cryptographic verification.
4. Who: Anyone can verify w/o seeing policy.
https://t.co/Tuyatn5VNZ

[Bastian Wetzel @bastian_wetzel]

1/ Vibe coded software of uncertain quality is flooding the world.

Vericoding is the counterpart to vibe coding, i.e. the AI generation of formally verified code from a formal spec.

Read my blog post on vericoding: https://t.co/W1v3BZLcJ5

[Houman Shadab @HoumanShadab]

@boardyai building ICME, a zero-knowledge proof network for verifiable and private AI agents. count me in for Boardy Pro

[Houman Shadab @HoumanShadab]

"ZK as infrastructure for critical agentic industries ... in certain environments cryptographic guarantees become necessary infrastructure, and ZK is uniquely positioned to provide this." -@zeroknowledgefm https://t.co/jb5iPhFX3J