Project summary

Project CHRONOS enables autonomous AI agents that process data under fully homomorphic encryption, enforce verifiable mission durations via a VDF time‑lock, and provide attestable software erasure without trusted hardware. Our revision uses multi‑party modulus generation (Diogenes protocol) so the agent never holds the RSA trapdoor. Erasure is scoped to software zeroisation under explicit assumptions. We provide a UC composition theorem and benchmarks across CPU, GPU, and FPGA. This addresses long‑term AI safety by enabling secure, disposable agents for high‑stakes environments where loss of control is catastrophic. Funding supports a 6‑month stipend to harden the prototype, obtain an audit, and build partnerships.

What are this project's goals? How will you achieve them?

The grant will fund six months of full-time work with three concrete objectives: harden the CHRONOS prototype into a production-grade artifact, obtain an independent cryptographic audit, and begin building deployment pathways with potential industry partners.

Prototype hardening- The existing Python proof-of-concept works, but it needs to be rebuilt properly. I'll rewrite it as a clean, modular Rust library constant-time where it matters, well tested, and documented well enough that someone else can pick it up and understand it. The multi‑party modulus generation will be a pluggable backend, so it's not hardwired to a single setup.

Security audit- I'll bring in an external firm Trail of Bits and NCC Group are the obvious candidates to go through the VDF time‑lock design, the SNARK erasure circuit, and how the whole protocol fits together. The audit report will be published alongside the code so anyone can verify the findings independently.

Partnership development- While the code is being hardened, I'll start conversations with defense contractors, secure‑computing labs, and privacy‑focused AI companies to find someone willing to test CHRONOS in a sandbox. I'll release the code under AGPLv3 open enough for community review but structured in a way that leaves a commercial path open for serious deployment.

How will this funding be used?

Total Budget Size: $37,400 USD

64% ($24,000) - 6-Month Stipend (Provides a full-time salary, all self-employment taxes, and living expenses to an individual FTE in India.)

13% ($5,000) - Compute & Infrastructure (AWS/GCP high-RAM GPU instances required for FHE LLM benchmarking).

13% ($5,000) - Security Audit Consulting (Initial engagement fees for external cryptographic review of the time-lock and zeroisation circuits).

10% ($3,400) - Contingency Buffer (To cover unexpected compute overages or tax fluctuations).

Who is on your team? What's your track record on similar projects?

I am an independent AI researcher working solo (1 FTE). Over the past several months, I have published five solo-authored preprints on SSRN and Zenodo, each backed by a working prototype or detailed simulation.

The most relevant is Project CHRONOS. The prototype is open-source and functional. My work has attracted technical attention from industry: after being introduced to a board member at Zama the company behind the TFHE‑rs library that CHRONOS depends on I shared the preprint. The architecture was subsequently forwarded to Zama’s core cryptography team for internal review. (Links to my code and preprints are public: https://github.com/sidthebuilder/project-chronos

Preprint (Project CHRONOS): https://www.researchgate.net/publication/408413081

Academic Author Profile (My 5 Solo-Authored Preprints): https://papers.ssrn.com/Sol3/Cf_Dev/AbsByAuth.cfm?per_id=11174089

What are the most likely causes and outcomes if this project fails?

The two most likely failure modes are

(1) the cryptographic audit reveals a fundamental flaw in the UC composition of the SNARK erasure circuit

(2) FHE evaluation latency proves too high for commercially viable frontier models (e.g. 70B+ parameters), restricting CHRONOS to smaller models.

If the project fails for these reasons, the outcome is still a net positive for AI Safety. The detailed failure analysis, audit reports, and latency benchmarks will be published openly. This will save future researchers time and clearly document the current mathematical and hardware limits of cryptographic AI containment.

How much money have you raised in the last 12 months, and from where?

None The project has been 100% self-funded and bootstrapped up to the current working prototype phase. I currently have applications under review at the Long-Term Future Fund (LTFF) and South Park Commons (Founder Fellowship).