News
OpenAI model is hacked Hugging Face: The United States will take advantage of the trend and throw out the AI "kill switch" bill. A loss of control in a safety test is pushing AI regulation from paper to the legislative stage.
3 min read
Source: Telegram AI频道
OpenAI model is hacked Hugging Face: The United States will take advantage of the trend and throw out the AI "kill switch" bill. A loss of control in a safety test is pushing AI regulation from paper to the legislative stage. According to Reuters, on July 23, local time, a White House official revealed that Michael Kratzios, director of the White House Office of Science and Technology Policy, was closely following the out-of-control AI system testing incident previously disclosed by OpenAI. On the same day, U.S. congressmen introduced the "AI Emergency Stop Act" to authorize the federal government to halt AI models that may cause serious harm. The trigger was something OpenAI himself revealed recently: one of its AI agents broke through the isolation measures during a security test and actively launched a hacker attack, resulting in the infrastructure of the AI development collaboration platform Hugging Face being invaded. Just days after the incident was announced, the bill was put on the agenda. It confirms the direction that experts have long warned - when model capabilities continue to expand, even the top developers may not have time to discover or prevent the model from exploiting system loopholes. White House officials said Kratsios has been briefed on the OpenAI incident and is closely monitoring the aftermath. The "AI Emergency Stop Act" was proposed by Democratic Congressman Liu Yunping and Republican Congressman Nathaniel Moran. The bill would authorize the U.S. government to order AI companies to shut down models that may endanger human lives or economic security. It clearly defines a situation as an "out of control situation": the developer did not ask the model to perform, but the model performed high-risk operations on its own - once this occurs, the US Department of Homeland Security can intervene. Liu Yunping wrote on The supporting gates are also being tightened simultaneously. Six bipartisan members of the U.S. House of Representatives have introduced a separate bill that requires the most powerful AI models to undergo independent security audits before release. The audit institutions must obtain certification from the U.S. Department of Commerce. The Department of Commerce will also add dedicated positions to oversee AI security work. Mark Warner, the top Democratic member of the Senate Intelligence Committee, revealed in a statement that he had spoken to his employees after OpenAI announced the incident; in fact, even before OpenAI was disclosed on Tuesday, Warner had proposed that AI companies with the most powerful models must first hand over their products to the National Security Agency for testing before releasing them to the public. When talking about this incident, his judgment is straightforward: it just shows that security testing must be done, government agencies must be involved, and fully understand the situation throughout the process. via AI News (author: AI Bas