Artificial Intelligence & Machine Learning
,
Next-Generation Technologies & Secure Development
,
Regulation
Watchdogs Say Public Accountability Suffers When AI Oversight Stays Hidden
A long-awaited voluntary artificial intelligence safety framework from the White House was meant to give more clarity on how the government will scrutinize AI companies.
See Also: Securing AI Workloads With Ubuntu Pro
The White House decided to keep the framework behind closed doors, refusing to release it to the public, drawing criticism from many technology watchdogs and politicians who say that secret oversight is no oversight at all.
Axios first reported that the White House chose not to publicly release the framework, which should lay out which type of AI systems the government would review before public release and how it plans to approach the evaluation process. Parts of the framework have leaked. The Washington Post reported, the White House will focus federal efforts on testing closed models and not open-weight models (see: Chinese AI Models Narrow Gap With US Frontier Labs).
Chris Mackenzie, vice president for communications for Americans for Responsible Innovation, told ISMG in an interview that the White House’s decision to withhold details of the framework defeats the purpose of a safety framework.
“If only the AI companies know the rules, then there is no accountability,” Mackenzie said. “Just like other industries held to public scrutiny, oversight of AI should also be public, and we need to be aware of what the rulebook is.”
He added that the framework should publicly clarify how the administration will treat AI model safety reviews, especially after the ad hoc reviews the government has initiated on new frontier models, most famously on Anthropic’s Fable 5 and Mythos 5 models (see: US Lifts Export Curbs on Anthropic AI Models).
The ah hoc process created uncertainty over which models could be released and which ones would get an export ban. The government allowed OpenAI to release GPT-5.6 to a vetted group of users but it exerted export controls over the use of Anthropic’s Fable 5 and Mythos 5.
Technology companies such as OpenAI, Anthropic, Meta and Google were summoned by the White House on Tuesday to discuss the framework.
Other groups, like the Information Technology and Innovation Foundation, criticized the secretive process, stating that without public scrutiny of the framework, smaller developers, open-source projects, security researchers and allied governments will end up guessing which rules are shaping the global market.
“If the specific cyber benchmarks must remain classified, the administration should still publish the scope, criteria and timelines, because predictable rules are what let U.S. developers move fast – and secrecy here does more to entrench the incumbents in the room than to make anyone safer or to democratize the process for innovation,” said ITIF Associate Director of AI Policy Michelle Lopes Maldonado in an emailed statement.
The United States does not currently have a statute codifying the use and regulation of AI systems, though many bills have been filed in Congress seeking to do so. Rep. Lori Trahan, D-Mass., who introduced the Frontier Risk Oversight, National Transparency, Independent Evaluation and Reporting Act, posted on X that blocking the public release of the framework is “disappointing.”
“Clearly some of this work touches real security concerns, but we should all be worried about the prospect of our national approach living behind a secretive wall – whether at DHS, the intelligence community or another agency. AI governance belongs in a civilian agency, where it can be seen and questioned, not buried inside the national security apparatus. Secrecy by default doesn’t make us safe,” she said.
The White House’s plan to develop safety and review frameworks comes on the heels of several security incidents caused by AI models and a rising concern over the popularity of Chinese open-weight models.
“This is all happening when there is a serious trust gap in AI,” Mackenzie said.
Click Here For The Original Source.
