AI News Roundup – EU AI Act comes into force, OpenAI unveils research tool, California State University system provides AI tools to all students and faculty, and more

To help you stay on top of the latest news, our AI practice group has compiled a roundup of the developments we are following.

• • Several provisions of the European Union’s sweeping AI regulations came into effect this past week, according to the Financial Times. Some AI applications in the block are now banned, including social scoring systems, the collection of biometric information that could be used for discrimination. The European Commission also published guidance on how the regulations should be implemented by companies, which are intended to “explain how the prohibitions will apply,” according to one official. Further restrictions, including those focused on regulating large AI models and those that present a high risk to users (such as those in healthcare applications), will take effect between now and 2027.

• • OpenAI has released a new AI tool focused on assisting researchers in gathering and compiling information, according to The New York Times. Deep Research, according to the company, is “[a]n agent that uses reasoning to synthesize large amounts of online information and complete multi-step research tasks for you.” A demonstration provided to members of Congress and other policymakers in Washington, D.C. this past week had the tool gathering information about Albert Einstein as a hypothetical nominee for U.S. Secretary of Energy. While the reports the tool generates include citations, it is still susceptible to “hallucination,” as many AI systems are. Deep Research is currently available to subscribers to ChatGPT Pro, OpenAI’s $200/month subscription.

• • The California State University (CSU) system has signed an agreement with several AI companies to provide free access to AI tools to all students, faculty, and staff, according to the Los Angeles Times. CSU, which comprises 23 campuses throughout the state, partnered with companies such as Microsoft, OpenAI, Meta and Nvidia to provide the tools, as well as help “identify AI skills needed in the California workforce and provide advice on how best to teach them,” as well as provide internship opportunities for CSU students. The initiative is believed to be among the largest efforts to integrate AI into higher education, and comes as concerns over the use of AI for academic cheating have swept academia. Despite these and other issues with AI implementation, CSU Chief Information Officer Ed Clark told the LA Times that “[r]ight now, AI is transforming every field, from academia to the workforce…. we need to ensure that our students graduate with the knowledge of how to use these tools.”

• • AI tools have been employed to aid in the discovery and identification of deep-sea creatures, according to Bloomberg. A submersible robot known as MiniROV has been used to capture video over 650 feet beneath the ocean surface, at first while under the control of an operator. This footage is then fed into an AI program that allows the robot to autonomously track undersea organisms. In the article, MiniROV followed a jellyfish for over five minutes before losing its trail. AI training is also being used to help identify such creatures – researchers use a game to allow players to attempt to identify animals and tag them as a known type or unknown, with players winning points for correct identifications. These results are then used to help train the AI classification system that powers MiniROV and similar systems. Researchers are excited to use AI in this manner, which one said, “could provide valuable insights into daily movement patterns, feeding behaviors and species interactions that are difficult to capture through traditional observation methods.”

• • Google’s DeepMind released the newest versions of its Gemini AI models this past week, according to Reuters. In a blog post, the company said that an updated version of Gemini 2.0 Flash, its “workhorse” AI model, is now available to all users of the Gemini app as well as to developers through an API. Also released was Gemini 2.0 Flash-Lite, a lighter model intended to balance quality with cost and speed considerations. Gemini 2.0 Flash-Lite is said to cost $0.019 per 1 million input tokens, in comparison to $0.075 per 1 million input tokens for OpenAI’s competitor model, and thus DeepMind is attempting to position itself in a market that has been greatly concerned about the cost of developing and running AI models in recent weeks. Such costs were thrust into the spotlight with the release of models from China’s DeepSeek that alleged comparable performance to more expensive models from U.S. companies at a fraction of the end cost as well as the training cost.