You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$100,000 left to reach the funding goal.
I created SE6 to take AI to the next level with bounded responsibility. I've contributed time and money. I want to realize a responsible, live, interactive autonomous AI system represented by a digital avatar.
That is my long-term vision. In this proposal, "self-aware" means operational awareness of the system's own state, permissions, actions and limitations. It is not a claim of consciousness. This grant would fund a six-month step toward that vision, not a promise to complete it.
SE6 Verify Commons will develop open-source tools for checking whether an AI agent actually did what it claims, recording the evidence behind actions, and retaining useful operational memory across restarts. The question is whether these checks improve reliability enough to justify their costs.
GOALS AND PLAN
I will commit 600 hours over six months, approximately 23 hours per week, beginning a few days after funding arrives.
Month 1: define the threat model, evaluation tasks, metrics and simple baselines.
Months 2–3: improve consequence checks and provenance records; test stale or tampered evidence, missing permission and restart behavior.
Months 4–5: run comparative evaluations and obtain contracted independent replication and review.
Month 6: publish the toolkit, reproducible evaluation suite, failure cases, limitations and final report.
The interface will expose verified state and action receipts. A general conversational avatar and unrestricted autonomy are not promised deliverables. Actions remain subject to explicit authority, bounded resources and independent verification.
HOW I WILL KNOW IT WORKED
We will measure false completion claims, detection of failed actions, false alarms, verified task completion, restart reuse, runtime and resource overhead. Comparisons will use the same tasks and resources for SE6 and simpler baselines. Held-out cases and an independent reviewer will test whether results reproduce. Negative results will be published.
This could help developers and evaluators identify unreliable agent actions. Its contribution to safety of more capable systems remains a hypothesis to test, not an established reduction in catastrophic risk.
BUDGET
Minimum: $54,000 for a reduced six-month scope.
- $36,000 founder stipend: 600 hours at $60/hour.
- $6,000 essential equipment.
- $5,000 contracted development.
- $5,000 independent review.
- $2,000 evaluation and release costs.
At this level we will prioritize software verification and a narrower evaluation suite, with less contractor time and no expanded electronics bench.
Full goal: $100,000.
- $36,000 founder stipend: 600 hours at $60/hour.
- $25,000 equipment: $20,000 compute, verification hardware, storage, networking and power; $5,000 for a conditional low-voltage electronics bench only if justified by the verification work and permitted by the grant.
- $24,000 contract development: estimated 300 hours at $80/hour, including expenses.
- $10,000 independent review: estimated 100 hours at $100/hour, including expenses.
- $5,000 evaluation and release: $2,000 evaluation/API usage, $1,000 incremental electricity/connectivity, $2,000 data preparation and release.
These are planning allocations, not vendor quotes or secured contracts. The stipend includes living costs and applicable personal taxes. There is no separate rent charge, travel, overhead or contingency. Existing housing support will be reconciled before grant-funded housing support begins. Purchases will be staged and scope reduced if costs rise. Spending and any platform arrangements must fit the award's terms.
TEAM AND TRACK RECORD
I am Sheldon Thom, publicly programmerONE, the originator and architect of SE6. I define requirements, direct AI-assisted implementation and testing, and evaluate results. No additional team members or contractors are confirmed. I work from home in Georgia, USA, with an HP Envy x360 Core i5 laptop. I have personally spent approximately $5,000 on SE6; this is an estimate and is not requested as reimbursement.
Public materials document bounded demonstrations of source-grounded actions, independent consequence verification, persistent operational memory and capability development. These are limited demonstrations, not evidence of general superiority. Some comparisons show ties and substantial overhead. Current work does not establish AGI, consciousness or new physics.
Public evidence:
https://github.com/foundationAIone/se6-competition-edition
https://devpost.com/software/se6-verify
FAILURE RISKS
The additional verification may cost more than it helps. Narrow tests may not transfer to harder environments. Evidence checks can be incomplete or compromised, and persistent memory can become stale. Contractor availability and my limited equipment may constrain delivery.
We will address these risks through explicit baselines, held-out tests, refusal when evidence or authority is missing, independent review and staged purchases. If the approach fails to outperform simpler methods, the outcome will be an honest negative result and reusable tests, not an unsupported safety claim.
OTHER FUNDING
No grant or competition awards have been received. Personal spending is approximately $5,000. Pending applications are Foresight ($100,000, October 31, 2026), Hacker Initiative ($25,000 for a distinct extension, October 31), BlueDot ($100,000, November 1) and EA Funds/Transformative AI Fund ($100,000, November 1). Nebius and UGREEN competition entries are also pending.
Foresight, BlueDot, EA Funds and this proposal are alternative sources for the core work. I will disclose any award and revise, reduce or withdraw overlapping requests before accepting duplicate support. The same expense will not be charged to multiple funders.
AUTHORSHIP
The motivation and vision above are mine. AI assisted with organizing and drafting this proposal, budget and milestones, which I am reviewing. SE6 implementation and testing are also AI-assisted.