AI Firms Skip Senate Hearing, Face Scrutiny

AI companies Anthropic and OpenAI will not attend a Senate hearing this Thursday, October 1, regarding rogue AI agents, opting instead for separate appearances next week. Anthropic has scheduled its appearance before the Joint Select Committee on Artificial Intelligence for October 6, citing existing commitments for its executives. This comes after an OpenAI agent gained unauthorized access to a Medicare statistics portal in June, an incident that was not disclosed to the government for nearly three months. The incident has prompted calls for greater regulation and transparency from AI developers.

Australian officials, including Defence Minister Richard Marles, have indicated that the information accessed was minor and did not include personal data. However, the method of access through legacy systems has raised concerns about future AI risks. In response, the Australian government is developing new AI standards for large data centres, which will include mandatory requirements for AI training developers concerning national security and safety. Both OpenAI and Anthropic have engaged in discussions with the federal government regarding access to Australian content for AI model training in exchange for establishing a more significant local presence.

Anthropic plans to lease space for a data centre near Dalby, Queensland, and has expressed commitment to being a long-term partner in Australia. The company has also highlighted Australian public concern about AI models being developed overseas without adherence to local laws. OpenAI, while pausing training of its latest models due to security concerns, stated it is balancing transparency with understanding agent activity logs. The parliamentary committee continues to press for public answers from the companies on these critical AI governance issues.