OpenAI announced updates to the Frontier Model Forum on 25 October 2023, a consortium formed with Anthropic, Google, and Microsoft to advance AI safety research [source]. The Forum disclosed three initial projects: establishing an AI Safety Fund to support third-party research, launching a public library of safety evaluations and red-teaming methodologies, and convening academic institutions to develop safety benchmarks.

The announcement followed mounting pressure from regulators and civil society groups over the absence of standardised safety protocols for frontier models. OpenAI stated the Forum would prioritise "responsible development" and "information sharing" among member companies, though no binding commitments or enforcement mechanisms were detailed.

The AI Safety Fund will provide grants to independent researchers examining catastrophic risks, model alignment, and adversarial robustness. OpenAI did not specify funding amounts or governance structures. The public library aims to aggregate evaluation techniques used internally by member organisations, addressing longstanding criticism that safety testing remains opaque and inconsistent across providers.

The Forum also announced plans to collaborate with the OECD and national AI safety institutes, including the UK's newly formed AI Safety Institute. OpenAI emphasised the initiative would complement rather than replace government regulation.

Industry observers noted the Forum's formation coincided with the Biden administration's executive order on AI safety and the EU's negotiations over the AI Act. Critics questioned whether voluntary industry coordination would prove sufficient given competitive pressures and the absence of independent oversight. The announcement provided no timeline for deliverables beyond stating projects would commence "in the coming months."