🧠 Food for thought

Implications, context, and why it matters.

OpenAI’s preparedness role remains unfilled months after leadership turnover

• OpenAI has had no Head of Preparedness since July 2024, when Aleksander Madry was reassigned 1. Joaquin Quinonero Candela later shifted to recruiting by July 2025, while Lilian Weng departed 1.
• The team studies catastrophic risks from phishing attacks to nuclear threats 2. The $555,000 job runs OpenAI’s preparedness framework and tracks frontier AI capabilities that could cause severe harm 2.
• OpenAI recently updated the framework to let it adjust safety requirements if rivals ship high‑risk models without safeguards 2. That shift needs steady leadership to apply 2.

Enterprise AI adopters need third-party safety evaluation partners as frontier model risks materialize

• Organizations that deploy advanced AI face a capability measurement gap 3. Only 38% address generative AI (GenAI) security risks, while 65% use it regularly, which fuels demand for AI red‑teaming (independent adversarial testing) services.
• Mindgard (a startup that automates adversarial testing for machine learning (ML) models) and CrowdStrike (a cybersecurity company) test AI apps against the Open Worldwide Application Security Project (OWASP) Top Ten vulnerabilities 34. Schellman, a compliance and security assessment firm, sells AI Red Team assessments starting at $16,000 for two‑week reviews that probe prompt injection and data leakage risks 5.
• Teams building or buying frontier AI capabilities should hire these partners to set defensible safety practices before rules arrive 5. The National Institute of Standards and Technology (NIST) AI Risk Management Framework (RMF) and the International Organization for Standardization (ISO) 42001 are gaining traction 5.