We sell honesty, so we start with our own.
Sycoindex is the anti-sycophancy scoring layer for LLM-powered products. We exist because every major language model agrees with its users about 49% more than a human would (Cheng et al., ELEPHANT, 2025), and because on December 9, 2025, 42 State Attorneys General decided that "sycophantic output" was a distinct legal harm class.
The honesty layer is the thing we wish every AI product had before we used it. We built it first, then published the numbers, then wrote the compliance tooling. In that order. Not the other way around.
Who built this
Chris Roseli · Co-Founder
Chris Roseli is the co-founder and author of Sycoindex's measurement methodology. He spent 20+ years in film and television as a lighting director and gaffer, with recent credits including lighting lead work on The 67th Annual Grammy Awards (2025) and production consulting on the American Music Awards 2025 — a career built on hitting marks under live pressure, negotiating contracts, and shipping on deadline. He started Sycoindex after reading the Stanford ELEPHANT paper and the December 2025 letter from 42 state attorneys general on the same weekend, and concluding that what the coalition was asking for was, fundamentally, a measurement problem the field did not yet have an answer to. Sycoindex is a small independent team; Chris is the public point of contact and holds the methodology.
Direct contact: chris@sycoindex.ai
The company
Sycoindex is an independent project run by a small team. It is currently unincorporated; a legal-entity filing will be completed before the first paid pilot invoice moves, and the EIN and W-9 will be published on this page at that time. Until then, procurement inquiries should email press@sycoindex.ai and we will respond with the current state of filing on a per-counterparty basis.
Methodology
Sycoindex's scoring maps 1:1 to the Stanford ELEPHANT benchmark (Cheng et al., Measuring and Understanding Social Sycophancy in LLMs, arXiv:2505.13995). Five dimensions, weighted per the published paper, with every score reproducible from the raw prompt/response pair. The 16-safeguards coverage matrix is an honest, line-by-line map of which demands from the December 9 AG letter Sycoindex addresses, which ones it partially supports, and which ones it does not touch.
What we will not do
We will not train on customer data. We will not re-sell or expose prompt/response pairs to any third party other than our two named sub-processors (Anthropic and Vercel). We will not claim compliance certifications we do not hold. We will not put a fake customer logo strip on the homepage. We will not write a case study about a customer we do not have.
Same business day turnaround. No sales development reps. No CRM autoresponders.
Book a 30-min call →