AKVC Labs
The verification platform
AKVC Labs is part of AK Venture Corp (AKVC). It runs a verification platform for AI agents.
What the platform does
The platform checks claims made about AI agents. It pulls claims out of a company’s
materials, scores the ones that are still uncertain, sends the ones that can be
formalised to a proof kernel, and keeps an audit trail of both.
AKVC Labs runs it as part of the company’s work.
Verification stack
Three parts, in plain language.
01
Calibrated evidence scoring
A claim is paired with an evidence file. A decision model scores how well
the file supports the claim. The number is a calibrated judgement about that
claim, not a star rating for a company, and not a promise that the claim is true.
When support is thin, the model should abstain.
02
Formal proofs
Claims that can be written exactly — algorithmic guarantees, policy invariants,
safety properties — are checked in Lean 4 and Mathlib, or a similar kernel.
A model may draft the proof. Only the kernel can accept it. A confident
paragraph is not a proof.
03
Audit trail
Inputs, model versions, scores, proof artefacts, and human sign-offs are
written to an append-only log. Each entry is chained to the one before it
by a hash, so a later reader can detect a changed record.
Programmes
Five programmes sit on this platform. Register interest to ask for more.
Longer notes are on the programmes page.
Verification-first studio
A time-bounded programme for teams building agentic products where a correctness
claim matters. It includes platform access, a claim ledger, formal specification
of a few core invariants, a calibrated evidence dashboard, and an audit log.
Register interest in the verification-first studio
Verified-claims diligence
A written review of specific claims, each with an evidence file, a calibrated
score, formal-check results where a claim can be formalised, analyst commentary,
and stated limits. A person remains accountable for the report.
Register interest in verified-claims diligence
Evidence-weighted syndicate
Materials for a prospective deal include an evidence file and calibrated scores.
Registering interest asks for information. It does not reserve a place.
Register interest in the evidence-weighted syndicate
Infrastructure tollbooth
An audit and formal-verification layer for teams building agent
platforms: hash-chained logs, proof checks, and claim ledgers.
Register interest in the infrastructure tollbooth
Proof bounties
Fixed-scope work by AKVC Labs: formalise a claim, attempt a proof
or a counterexample in Lean, and return the artefact with a written report.
Register interest in proof bounties