What should companies disclose about training, evaluations, incidents, energy use, safeguards, and deployment?
After the first Foundation Model Transparency Index publicly named laggards, measured disclosure rose — though training-data and compute details remain thin. Mandatory incident reporting — the aviation-safety model — is the most commonly proposed floor beneath voluntary disclosure.
View on the map → · Open in Browse →
Measured transparency went backwards: the Foundation Model Transparency Index fell 17 points against stiffened criteria, with training data the most opaque domain — reversing the index's earlier finding that naming laggards raised disclosure. A novel constitutional constraint arrived from the other direction: mandatory disclosure of trade secrets may trigger Fifth Amendment takings claims, an argument any disclosure regime now has to design around.
Companies score worse this year. The mean score this year is 41: this is a 17 point decrease over the previous year.
New edition of the canonical index reverses its own earlier finding that naming laggards raised disclosure: measured transparency fell 17 points against stiffened criteria, with training data still the most opaque domain.
Regulations that force developers to disclose trade secrets to the public could violate the Constitution.
Novel constitutional constraint on the transparency agenda: because trade secrets are protected property, mandatory public disclosure of training-data and model details could trigger Fifth Amendment takings claims.
Each large developer must design, implement, and publicly post a company-wide safety and risk management plan describing how catastrophic risks are identified, assessed, and reduced.
The reference explainer for SB 53 implementation: how the first binding US frontier-transparency statute converts voluntary safety frameworks into enforceable publication, 15-day incident reporting, and whistleblower duties.
The EU GPAI Code of Practice commits Signatories to providing a description of intended model behavior (a 'model spec') in their Model Reports
Turns the Code of Practice's model-spec commitment into concrete disclosure requirements — publish intended behaviors relevant to systemic risk so the AI Office can audit against them — a template for spec-level transparency mandates.
All of the companies reviewed are racing toward AGI/superintelligence without presenting any explicit plans for controlling or aligning such smarter-than-human technology
Latest external scorecard of company disclosure and risk-management practice (Anthropic and OpenAI at C+, most rivals D or below) — evidence for how far voluntary transparency falls short of what mandatory floors would demand.