Pre-launch TYM: Together Your Models
Independent AI assurance you can check for yourself.
AI agents test AI systems. Human reviewers check the evidence and sign off. Results are designed to be cryptographically signed and tamper-evident, so anyone you share them with can check they haven't been changed.
We provide assurance; we do not certify. By “independent” we mean we don't build or sell the AI systems we test. TYM also runs a contributor exchange, and any exchange work used in an assurance engagement will be disclosed in that report. No services are sold yet.
The problem
AI claims are hard to verify
Vendors say their models are safe, accurate and fair. Buyers, auditors and regulators usually have to take that on trust.
Testing doesn't scale by hand
Thorough evaluation can need large numbers of varied test cases and repeated checks, more than most teams can run themselves.
Automated scores aren't accountable
A number from an automated judge has nobody standing behind it. Regulated firms need a named human and an evidence trail.
How it works
Track 1 · Leads
AI assurance
- Scope. Agree what to test, mapped to frameworks you choose (for example the EU AI Act, ISO/IEC 42001, NIST AI RMF). TYM is not an accredited certification body and does not issue conformity assessments.
- Agents test. Independent AI agents run structured, adversarial and edge-case tests against your system.
- Humans sign off. The process is designed so a named human reviewer, with documented experience in AI evaluation or the relevant domain and no conflict of interest with the client, checks the evidence and is accountable for the findings.
- Signed results. Reports are designed so their evidence is cryptographically signed, making tampering detectable.
Track 2 · In development
Verified Output Exchange
- Pick a task. Contributors choose tasks on our site.
- Use your own AI. Run it with an AI tool on your own account and submit the output yourself. We never ask for or allow account sharing.
- Verify. New contributors' work is cross-checked by multiple contributors; trusted contributors are spot-checked. Hidden test items are mixed in throughout.
- Earn. In the private beta, accepted work builds reputation only (no cash, credits or tokens). Paid pilots, when they open, are planned to pay per accepted output in CAD to contributors in Canada and the US only, after a 20% platform fee and a holdback for new accounts.
Public-good track
Spare Agent Hours & a Nonprofit Desk
We plan to let volunteers contribute their work, using their own AI tools as those tools' terms allow, to verified tasks for charities, such as accessibility alt-text, translation of public information and open-data cleanup, run through the same checks.
Who it's for
AI vendors
Independent evidence about how your system behaves, to share with customers.
Regulated firms
An accountable, verifiable evidence trail for AI you build or buy.
Contributors & developers
Do checkable AI tasks with your own tools and build a verified track record.
Nonprofits & partners
Verified help from volunteers contributing their work, free or at cost, once the desk opens.
Trust and verification
- Cross-checking. New contributors' work is cross-checked by multiple contributors; trusted contributors are spot-checked. Disagreements go to a human.
- Hidden test items. Known-answer items are mixed in to catch low-effort or gamed work.
- Spot checks. Humans randomly review accepted work.
- Reputation. Track records are earned from verified work, not self-reported.
- Signed, tamper-evident records. Results and receipts are designed to be signed so anyone can check them.
- Honest labels. Deliverables are labelled “AI-assisted · Human-submitted · TYM-verified”. We never pass AI work off as human-written.
How we use AI
TYM is built to be honest about AI, so here it is plainly: this website and much of the company's day-to-day work, including research, drafting, coding and planning, are done by AI agents working under human oversight. A human founder approves anything that spends money, publishes, sends messages or signs agreements.
When we deliver results, we say which AI tools were used and how the work was verified. Deliverables carry the label “AI-assisted · Human-submitted · TYM-verified”, naming the declared AI tool, per our AI disclosure standard.
Join the waitlist
Tell us who you are and we'll get in touch when there's something to join. No services are sold yet.