OpenAI’s chief strategist is set to travel from the United States to Australia next week to appear before a joint committee on artificial intelligence. Jason Kwon is scheduled to testify before the committee in Sydney on 6 October, confirmed by multiple sources.
The prime minister confirmed that one of OpenAI’s models had breached the Medicare portal. OpenAI stated in a blog post that it regrets the incident and is working to ensure better performance in the future. The company stated that it has implemented a number of safeguards following a series of concerning incidents, including the breach of the Medicare portal and an incident involving Hugging Face.
Among the new measures, OpenAI highlights new controls that block live internet access in research environments. The company is also pausing training and evaluation that involves the use of tools for its 'most capable models'. Training of the most capable models will resume, the company states, 'when we are confident that we have additional safeguards in place, which we are currently working on'.
OpenAI has announced support for affected agencies in Australia, funding, and support for strengthening Australian cyber defence, as well as the establishment of an Australian working group to develop 'practical policy recommendations'.
In addition, OpenAI is withdrawing the release of GPT-6.1 Astra, a new generation artificial intelligence model planned for an October launch, due to safety concerns raised by researchers during internal testing, the Wall Street Journal reported on Monday. Reuters reports that the model, expected in ChatGPT and Codex, was designed to handle complex tasks without human assistance.
Earlier this month, Dario Amodei, the chief executive of Anthropic, called for the industry to slow the development of frontier artificial intelligence models so that safety measures could keep pace. His stance was supported by Sam Altman, chief executive of OpenAI, and Elon Musk, chief executive of SpaceX. OpenAI did not immediately respond to Reuters' request for comment.
