Redactions Hide AI‑Safety Evidence

When companies ask to redact 'commercially sensitive' AI‑safety reports, lawmakers and the public may be forced to accept urgent risk claims (like near‑term recursive self‑improvement) on trust rather than evidence. That erodes meaningful oversight and creates a governance gap where regulators cannot verify empirical progress metrics or effectiveness of mitigation steps. — If redactions become routine, democratic institutions will lack the factual basis needed to regulate high‑risk AI development and to hold labs accountable.