Overview
- OpenAI announced the partnership with Cathedral on Tuesday, November 3, 2026, saying Cathedral researchers will work alongside internal teams and be allowed to use OpenAI’s models to probe defenses and test fixes.
- The company more than quadrupled its top bug bounty payout to $172,000 for exceptional critical findings and expanded monitoring of advanced models across training, evaluation, internal use and select deployments.
- OpenAI pledged $1 billion in subsidized Daybreak access over six months to give trusted security teams its most capable models for defending essential services and infrastructure.
- The move builds on a prior sprint that dedicated about 25% of product engineers and used Codex to reproduce flaws and prepare patches, reflecting a shift from occasional drills to continuous, model‑assisted defense.
- Cathedral’s recent claim that its agents rapidly accessed parts of Anthropic’s systems has heightened scrutiny of third‑party testing and disclosure practices and could prompt closer industry and government attention to vendor risk and verification.