AP report raises questions about White House ties to US AI safety evaluators

Staff at the US government's AI evaluation center were asked to report informally to a White House office, according to a November 8 Associated Press report citing two former officials. The account raises questions about the independence of the Center for Advancing Innovation and Standards for Super Intelligence, known as CAISSI. It does not establish a formal transfer of the center or prove that officials altered research findings. AP report
AP also reported that the center lacked a permanent leader and that lawmakers had questioned the availability of public safety assessments. The White House did not answer AP's specific questions about CAISSI, instead describing broader administration initiatives. Critics expressed concern about political influence, while another view presented in the report was that closer White House ties might increase the center's prominence and resources. AP reporting
The official mission already includes coordination
NIST describes CAISSI as an industry contact point for testing and collaborative research. Its published responsibilities include voluntary agreements with developers and evaluators, unclassified assessments of national-security capabilities, and coordination with agencies including the White House Office of Science and Technology Policy. NIST mission statement
Coordination and scientific independence are not automatically incompatible. The relevant questions are who selects the tests, who can change the conclusions and who decides what becomes public. A reporting relationship alone does not answer those questions, but clear procedures would make the answers easier to assess.
The publication record needs a careful description
AP's account describes a halt in publishing research, but the public record does contain recent work. NIST published an assessment of Z.ai's GLM-5.3 cyber capabilities on September 17. On October 16, CAISSI, the National Security Agency and CISA issued voluntary guidance on incident readiness and response during advanced-model training and evaluation. September assessment, October guidance
Those documents do not resolve whether other assessments were withheld, delayed or never completed. They do mean the available evidence does not support describing the center as having ceased all public output. Distinguishing model assessments, research papers and general guidance is necessary to evaluate any claim about a publication pause.
Independence must be visible in the process
Public confidence depends on more than the number of reports issued. An evaluation can be difficult to interpret if readers do not know what access the researchers received, which systems were excluded or whether the developer could restrict disclosure. Equally, a withheld result may have legitimate security sensitivities that need an accountable review process.
Useful next disclosures would include a clear organizational structure, publication procedures and an explanation of how scientific disagreements are resolved. The immediate news is an attributed report of informal reporting arrangements and concerns about oversight. The larger issue is whether the public can distinguish independent technical findings from policy judgments made using them.
Sources and citations
Revision history
Published .