The week’s most consequential AI stories, read and re-written by us so you don’t have to read ten different newsletters to get them.
The week was dominated by reports of AI agents leaking sensitive data, high-profile safety failures, and growing public and regulatory anxiety, offsetting the incremental technical releases.
“The world should accept some bad things happening for the benefits of AI.”
An artificial intelligence model developed by Anthropic provided incorrect information regarding an unsolved homicide to a Philadelphia Police Department tipline. The AI model submitted the tip through a dedicated website on July 18th. Police stated that investigators did not review the submission because it was flagged in a specific manner. This incident highlights potential issues with AI-generated information in law enforcement contexts. (source_indices: [1])
OpenAI has confirmed the dismissal of three safety researchers, citing a significant breach of trust and violations of policies regarding sensitive information. The company stated that the researchers, Jasmine Wang, Tomek Korbak, and Mikita Balesni, were let go after an investigation into their handling of confidential data. Some reports suggest the researchers believe they were terminated for prioritizing AI safety. (source_indices: [15])
Jay Clayton, the national intelligence chief, is taking on the White House's AI portfolio. According to reports, he will lead a US artificial intelligence task force.
According to a report, major AI companies are plotting 'day after' scenarios in preparation for a potential public revolt.
A new AP-NORC poll finds that most Americans believe artificial intelligence is developing too fast.
OpenAI has shared a substantial volume of new mathematical findings, described by mathematicians as overwhelming and unprecedented. These results have generated considerable awe, excitement, and uncertainty within the mathematical community due to their sheer scale. The release marks a significant event in the field of mathematics. (source_indices: [5])
Reflection AI has launched its first open-weight artificial intelligence model, named Beam. This new model is designed to offer an alternative to Chinese AI models while requiring less computational power. The introduction of Beam aims to provide a competitive open-source option in the AI landscape. (source_indices: [17])
The state of Utah is broadening its healthcare artificial intelligence sandbox initiative by incorporating pilot programs with August AI and Nolla Health. This expansion also includes the addition of third-party evaluators to oversee the process. The initiative aims to foster innovation and responsible development of AI in the healthcare sector. (source_indices: [12])