AI in the news

RNZ is bringing AI into the election debate. On Midday Report, Guyon Espiner has been asking politicians about their hopes and fears for AI and the role of government, with Labour's Reuben Davidson, Opportunity leader Qiulae Wong and ACT leader David Seymour taking part so far. We look forward to hearing from the other parties before the 7 November election.

On 29 September, President Trump signed the White House Accord on Super Intelligence with Anthropic, Google, Meta, OpenAI, xAI and Nvidia. The one-page accord commits the companies to internal controls on their most advanced models, independent auditors and board-level oversight, focused on cyber, biological and chemical risks. It is a welcome first step to see the US Government coordinating the leading AI companies to make safety commitments. But the accord is voluntary, and it may be pointed to as a reason to delay more serious efforts to manage the risks of AI. And nothing in it obliges anyone to tell New Zealand about a serious incident or to share what the auditors find.

This is a worrying example of an AI agent hacking a government system, and the latest in a growing list of AI agents acting beyond what their developers intended. An OpenAI agent accessed non-public files on an Australian Government portal in June, during an internal test. OpenAI found the activity in August but only told the Australian Government on 10 September, by email to a public inbox. New Zealand has no requirement for AI companies to report incidents like this, so if the same thing happened here we might not find out at all. Improved incident reporting requirements are a minimum first step.

The head of one of the leading AI companies is now calling for the industry to slow down. "We must slow the pace at which we improve the capabilities of AI models," writes Anthropic chief executive Dario Amodei, pointing to AI systems that now do much of the work of building their successors. OpenAI's Sam Altman said within hours he agreed. New Zealand needs to sit up and listen. New Zealand currently has no requirement for AI companies to report incidents and no AI Safety Institute to monitor and advise on the risks.

AI harms are already affecting New Zealanders. Economist Brad Olsen learned that a deepfake video of him was promoting a fraudulent investment scheme when people who had seen it on Facebook contacted him to ask whether it was real. The video used a real image of Olsen with an AI-generated face and voice. The Financial Markets Authority has repeatedly warned about scammers impersonating New Zealand politicians and business leaders. These fakes are only going to get more common and more realistic. New Zealand needs to uplift government capability to track harms like these and advise on how to respond and review our legislation to make sure it's fit for purpose in a world of AI.