AI News Roundup – Trump administration limits use of Anthropic’s new Mythos-type Fable model, Apple announces new AI features for Siri, OpenAI considers major ChatGPT changes amid IPO, 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.

• The Trump administration has restricted the use of some of Anthropic’s newest AI models by foreigners on national security grounds, according to The Wall Street Journal. This past week, Anthropic released its latest AI model, Claude Fable 5, a limited-capability variant of the company’s Mythos cybersecurity model. Fable 5 will allow users to query Anthropic’s advanced Mythos cybersecurity model, which raised concerns after its release earlier this year over its potential to be used for malicious hacking and was previously limited in its release due to these risks (as this AI Roundup has covered). However, Fable 5 will shut down requests regarding sensitive issues (such as bioweapons or exploiting software vulnerabilities) and direct users to Anthropic’s older Opus 4.8 chatbot. The restrictions on Fable 5 have already angered users and AI developers, who claim that the company’s safeguards were too restrictive and that it was not readily clear what users were permitted to ask. One user reported that the model refused to answer a basic biology question regarding mitochondria. In response to these concerns, the company said that it “made the wrong tradeoff and we apologize for not getting the balance right,” and that restrictions on biology-related queries were “necessary to be overly conservative” to avoid “highly risky” research, and that the company would work to reduce unnecessary obstructions in the model. Some AI developers also claimed that the model’s outputs related to AI development were purposely degraded, likely to combat model “distillation” that allows AI models (often low-cost, open-source models) to gain knowledge and capabilities from other models. These developers claim that these restrictions were evidence of anticompetitive behavior from Anthropic, though the company cited national security reasons as a justification. Shortly after Fable 5’s release, the Trump administration announced that it would place export controls on the new model, effectively prohibiting use of the product outside of the U.S. In a blog post addressing the government’s action, the company said that it was suspending access to Fable 5 by all users regardless of nationality “to ensure compliance” with the directive, though Anthropic disputes the severity of the risks given as justification for the government’s action. The move was reportedly prompted by concerns raised with White House officials by Amazon’s CEO Andy Jassy, who claimed that Amazon teams had used Fable 5 to gain information on cyberattacks that was supposedly off-limits to users due to Anthropic’s claimed restrictions on the model.

• Bloomberg reports on new AI features coming to Apple’s Siri voice assistant app. The company held its annual Worldwide Developers Conference (WWDC) this past week, and announced a long-awaited overhaul of its Siri voice assistant to integrate AI features. Siri, first introduced in the iPhone 4S in 2011, has long been a core component of Apple’s operating systems, but was seen as falling behind in the AI race as customers leaped to AI chatbots such as OpenAI’s ChatGPT or Anthropic’s Claude. Apple claims the new Siri, called Siri AI, will be smarter and more reliable than past iterations of the assistant, and will be able to seamlessly integrate into the rest of the Apple ecosystem. Siri AI will include capabilities to search for concert information, play music, and set reminders, as well as perform certain tasks on a user’s behalf. The new system also includes AI-powered photo editing and generation capabilities. Siri AI runs on Apple Foundation Models, which were developed by the company based on Google’s Gemini AI models (as this AI Roundup covered this past January). The company also announced other AI features across its products, including the ability for AI to organize tabs in the Safari browser, suggest photos to share in iMessage conversations, and the ability to create shortcuts with natural language prompts, as well as revamped parental controls for devices owned by children. The Siri AI features are currently available as a beta for developers, but are expected to be released to the public later this year.

• Major changes may be coming to OpenAI’s ChatGPT amid the company’s looming IPO, according to The Financial Times. The company is reportedly planning to transform ChatGPT, currently a chatbot app, into a “superapp” that also integrates coding tools and AI agents, with the belief that more features will generate more revenue for the company. As this AI Roundup covered in March, OpenAI has shifted its goals towards attracting business customers to compete with its rival Anthropic and as the company faces pressure to increase revenue and chart a path to profitability ahead of the company’s planned initial public offering (IPO) later this year. Indeed, the company confidentially filed its S-1 form with the U.S. Securities and Exchange Commission as part of the IPO process last week, following Anthropic’s similar filing the week before and just a few days before Elon Musk’s SpaceX raised $75 billion in its IPO, making Musk the world’s first trillionaire. Competition in the AI space has grown with demand, though business customers have begun to balk at the high prices AI labs are charging for their products. The Wall Street Journal also reported this past week that OpenAI executives are considering major price cuts for its AI products to undercut Anthropic due to pushback from the lucrative business customers OpenAI seeks to poach from its rival. ChatGPT’s transformation into a “superapp” would be a part of this strategy, as the new app would direct users towards using OpenAI’s coding tool Codex and integrations with other companies such as Canva and Booking.com. The ChatGPT changes are expected to slowly be rolled out over the company’s apps and websites in the coming weeks.

• The New York Times reports on a new AI application aimed at helping doctors answer complex medical questions. Over half of physicians in the U.S. are reportedly users of OpenEvidence, a new AI chatbot system designed for medical professionals. The system is trained exclusively on medical journals and other research data, and uses that information to respond to doctors’ questions, which often include patient symptoms and characteristics and a request for help with a diagnosis. OpenEvidence then responds with a summary of likely diagnoses and provides links to research that informs the recommendations it makes. OpenEvidence is part of a new wave of AI applications in the medical field. Early applications included tools to help physicians with documenting patient visits and completing burdensome paperwork, often called “AI scribes.” Over a quarter of U.S. physicians now use such systems, according to a recent report from the American Medical Association. Newer AI apps, like OpenEvidence, are aimed at assisting with substantive medical work, including diagnosis and treatment of patients. OpenEvidence is available to government-verified physicians as a free app download, though the product is supported by advertisements (often from pharmaceutical companies) that are displayed while OpenEvidence generates an answer. Some healthcare institutions have been skeptical of the tool, as doctors often use it on their own without employer oversight, though some hospitals (such as Mount Sinai in New York City) are planning to allow OpenEvidence to link to patients’ electronic health records, though only after assurances and testing to protect patient privacy and safety. The app is reportedly compliant with federal privacy laws surrounding health information, but one Mount Sinai doctor told The Times that they were “not going to just throw a patient’s data over the wall to a private company.” Despite these concerns, OpenEvidence remains incredibly popular among physicians and has the potential to help under-resourced hospitals that do not have access to specialist expertise when addressing thorny medical questions.

• The Japanese government is planning to set up an AI-powered autonomous bioscience lab to aid drug research, according to Nikkei Asia. The country’s Ministry of Education, Culture, Sports, Science and Technology is planning to appropriate around 8 billion yen (about $50 million USD) over two years for the effort, which would construct a lab operated by robots that would perform biological and chemical experiments. An AI system would orchestrate the lab by directing the robots’ movements and actions and analyzing the results of the experiments. Researchers will be able to access the lab’s results, which will be available through an online service. A similar project is underway at Carnegie Mellon University in the U.S., amid a push to improve research efficiency. The Carnegie Mellon project is estimated to increase productivity by up to seven times and possibly double annual research paper output. The Japanese project will be directed by the Institute for Molecular Science, an inter-university biotechnology consortium, as well as the Institute of Science Tokyo and the National Institute of Informatics, the latter of which is a Japanese AI institute. The experiments are expected to focus on genetics, proteins, and brain activity, with an ultimate goal of developing new drugs and bioscience materials. After the lab is completed and operational, the government ministry will evaluate the results and consider deployment of similar automated labs in other scientific fields.