Secret AI Safety Rules Raise Concerns About Government Transparency and Potential Conflicts of Interest
Protect Democracy warned that the undisclosed framework for evaluating artificial intelligence systems could create opportunities for corruption. The organization said officials from the Office of the Director of National Cyber Affairs, the Office of Science and Technology Policy, the Treasury Department, and the Department of Commerce are supporting the rapid deployment of AI companies favored by President Donald Trump.
Protect Democracy also pointed to the administration’s reported retaliation against AI companies accused of being “woke.” The group cited a recent ruling that found Trump’s decision to blacklist Anthropic was unlawful, arguing that the case raises additional questions about whether political considerations could influence the government’s AI review process.
As AI capabilities expand and cybersecurity threats grow, Protect Democracy said the public should not be expected to blindly trust the administration to select and rigorously test leading AI models. The organization referenced recent AI security incidents, including the reported hacking of an OpenAI model on Hugging Face, as evidence of the need for strong and transparent oversight.
The group also warned that the public may not know whether future vulnerabilities result from incomplete government reviews, outdated or ineffective testing methods, missed security checks, or a failure to follow the latest AI safety and risk-mitigation practices developed by experts.
“We have a right to know what’s in the framework,” Protect Democracy said, joining broader calls for federal agencies to disclose how they assess and approve the deployment of advanced AI models.
In a post on X last month, U.S. Rep. Greg Casar, D-Texas, accused President Trump of taking “millions of dollars from AI billionaires” while “completely failing to protect us from the dangers of AI.”
“Now, in response to extremely dangerous AI cybersecurity issues, we’re saying we’ve launched a ‘voluntary’ review that no one has ever seen before,” Casar wrote. “They’re falling asleep at the wheel. They’re too busy making money to protect our jobs and our national security.”
More recently, California state Sen. Josh Becker asked the court to grant Protect Democracy’s request. In a declaration, Becker supported the organization’s appeal and highlighted California’s efforts to establish independent AI safety standards.
Becker pointed to SB 813, legislation that would create a process for an independent organization to establish baseline safety standards for artificial intelligence. Unlike the Trump administration’s approach, Becker said California’s proposal would make the benchmarks, standards, and methodologies used to evaluate AI risks available to the public.
Becker added that the bill has already been shaped by public feedback and that one provision was withdrawn following public opposition. Supporters say the example demonstrates why transparency and public participation should be central to AI regulation and federal oversight.
Source: arstechnica.com