EXCLUSIVE: Behind AI industry's call for independent reviews, a fight over what gets made public
- OpenAI and Anthropic scrambling to maintain editorial control over 'independent' safety evaluations.
- Internal documents reveal labs want to gatekeep what researchers publish to protect trade secrets.
- Proposed 'safety' measures include embedding lab employees to watch evaluators and restricting cross-firm access.
- Critics warn the lack of standardized, truly independent oversight turns AI safety into a PR-managed charade.
Brief Summary
The tech titans behind the AI boom are engaged in a frantic behind-the-scenes power struggle to ensure that 'independent' safety evaluators don't accidentally let the cat out of the bag. While CEOs publicly pledge their commitment to transparency, private negotiations reveal a heavy-handed push to maintain editorial veto power over critical safety reports. Labs are citing intellectual property concerns as a justification to redact findings, essentially turning the gatekeepers of AI safety into subjects of the very companies they are supposed to be policing.
Why This Matters
You are being sold the idea that these powerful AI models are being stress-tested by neutral experts, but the reality is that the labs are holding the leash. Because there is no federal standard for these reviews, the companies building the technology are effectively writing their own report cards. If these labs succeed in controlling the narrative, any potential risks—whether they involve data security, rogue behavior, or systemic bias—could be scrubbed from public view before you ever hear about them, leaving you to trust the word of corporations that are prioritizing trillion-dollar valuations over full disclosure.