The Governance and Advisory Council.
The Council is the group that reviews releases, capability changes, and capital partners. It can refuse a decision on safety or alignment grounds. Most seats are external, most decisions are written down, and the meeting calendar is set before the agenda is.
Four rules that hold every review.
A real veto, not advice.
The Council can refuse a release on safety or alignment grounds. The decision sticks. It is not advisory to the CEO or the board.
External seats outnumber internal.
Most seats are held by people outside Blankline. They are paid for their time, they hold no equity, and they cannot be removed for the substance of their votes.
Decisions on the record.
Every review produces a short written outcome with the votes and the reasoning. Where the decision affects a public product, the record is public.
Reviewed on a calendar.
Composition, scope, and procedure are reviewed twice a year. The dates are set before the meeting that sets the agenda, so no one can quietly skip a review.
Governance Topology.
A proposal comes in with its safety review and its check against the charter. The Council discusses and votes on the record. The outcome is one of three: ship, hold, or stop.
Inputs
Proposal, evidence, prior decisions
Council review
Discussion and vote on the record
Outputs
Decision, reasoning, public record
Five kinds of decision, every time.
Frontier model releases.
Any new model shipped publicly or significantly retrained. The Council votes on whether the safety evidence presented is sufficient.
Capability and permission changes.
A new tool, a new integration, or a new permission scope that meaningfully changes what our products can do or who they can affect.
Sensitive contracts.
Customer deployments in regulated, high-stakes, or dual-use settings. Reviewed against the charter before the deal is signed.
Capital partners.
New investors and material changes to the cap table. The Council can decline an investor on alignment grounds, regardless of valuation or check size.
Severity-one incidents.
Any incident where a user could be harmed, or where the root cause points to a flaw in the safety design rather than the implementation.
Five seats. Four external.
The Council is built so internal voices cannot out-vote external ones. We publish people as they are seated. Until then, this is the structure they will sit in.
External AI safety researcher
Active researcher in alignment, evaluation, or interpretability. Two seats.
External governance or policy expert
Background in technology policy, civil liberties, or regulation. One seat.
External deployment engineer
Senior practitioner with experience running AI systems in production at scale. One seat.
Internal
One seat held by a member of the Blankline leadership team. Cannot be the CEO.
Chair
Elected by the Council from among the four external seats. Three-year term, renewable once.
A living document.
The Council’s composition, scope, and procedure live on this page. The page is the working version, not a brochure. When it changes, the date and the reason change with it.