Project summary

TELOS provides runtime governance infrastructure for AI systems, continuous measurement and enforcement of behavioral boundaries during deployment, not just training. We apply control theory and statistical process control to treat conversational drift as a measurable process variable requiring real-time oversight.
What are this project's goals? How will you achieve them?

 Establish TELOS as validated, peer-reviewed methodology for runtime AI governance.

- Expand adversarial validation beyond current 1,300-prompt dataset

- Publish methodology and results through arXiv and conference presentation (FAccT or AIES)

- Partner with academic institution for independent replication study

- Maintain open datasets and reproduction code for community verification

How will this funding be used?

- Expanded validation / compute: $10,000 — Additional adversarial testing across model providers, edge case exploration

- Publication + conference: $5,000 — arXiv fees, conference registration and travel

- Academic collaboration: $10,000 — Independent replication study partnership

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

Jeffrey Brunner — Founder/CEO, TELOS AI Labs Inc. 30 years in education, Lean Six Sigma Black Belt. Technical development, validation research, regulatory engagement https://www.linkedin.com/in/brunnerjf/

Jennifer Brunner — Co-founder, Operations & Partnerships. Business administration, partnership development, operations management. https://www.linkedin.com/in/jenniferbrunner2014/

TELOS AI Labs is a Delaware C-Corp with joint majority ownership.

Track record:

- Validated TELOS against 1,300 adversarial attacks (HarmBench + MedSafetyBench) achieving 0% attack success rate vs 30.8% baseline

- Published open datasets: https://doi.org/10.5281/zenodo.18013104, https://doi.org/10.5281/zenodo.18009153

- Open source code: https://github.com/TelosStward/TELOS

- Live beta deployment: https://beta.telos-labs.ai

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

Most likely failure modes:

- Academic partners decline collaboration due to unfamiliarity with approach

- Expanded adversarial testing reveals edge cases that degrade performance metrics

- Publication rejected or delayed beyond useful timeline

Outcomes if failed:

- Validation data and code remain publicly available for others to build on

- Methodology documented regardless of adoption

- Funds used transparently on stated purposes with public accounting

Failure would delay but not eliminate the research contribution. The infrastructure exists; this funding accelerates validation and credibility.

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

Self-funded to date. Currently bootstrapped.

Pending applications:

- Long Term Future Fund: $75K-$150K (passed first round, awaiting decision)

- Survival and Flourishing Fund: $200K (application completed)

- Coefficient Giving: $200K-$300K (submitted)