After industry pushback, the White House has clarified it is not pursuing a new, FDA-style bureaucracy for AI model approval. Instead, the administration is focusing on direct, ongoing collaboration with major AI labs to mitigate extreme risks before models are released, favoring a flexible partnership over rigid regulation.

The US government's delayed AI safety framework is being developed behind closed doors. By sharing physical drafts in briefings without allowing companies to keep them and keeping the benchmarking process classified, the White House is creating uncertainty and forcing companies to self-regulate in a vacuum.

Anthropic’s choice to subtly degrade answers for AI development queries, rather than openly refusing them, was a critical error. This lack of transparency confused users and damaged trust, proving that the method of implementing safety guardrails is as important as the policy itself.

The US government's new AI safety testing framework is secret, with details withheld even from uninvited AI companies. This approach prioritizes maximum flexibility for the government but creates 'minimum knowability' for the industry, hindering planning and fostering distrust.

The most powerful AIs may never be released publicly due to their dangerous capabilities. As they are used internally, they pose significant risks that current transparency laws, which focus on public models, do not cover.

The proposed White House framework for reviewing advanced AI models applies to closed-source systems from companies like OpenAI but exempts open-weight models from Meta and others. This creates a potential regulatory loophole, as open-weight models can be harder to control and monitor once released into the wild.

Instead of establishing clear regulations, the White House is intervening directly in AI rollouts, limiting access to new models like OpenAI's on a case-by-case basis due to national security. This high-touch approach gives the government immense control but creates uncertainty and is viewed by some safety advocates as a 'worst of both worlds' scenario.

Reporting AI risks only to a small government body is insufficient because it fails to create 'common knowledge.' Public disclosure allows a wide range of experts, including skeptics, to analyze the data and potentially change their minds publicly. This broad, society-wide conversation is necessary to build the consensus needed for costly or drastic policy interventions.

The White House is expanding its safety framework to include open models not just for risk mitigation, but to avoid creating a two-tiered system. Officials fear that excluding open models would signal they are not 'approved,' disincentivizing enterprise adoption and harming US open-source labs. The framework is thus becoming a de facto certification.