OpenAI Backs Singapore's Call for AI Incident Transparency

OpenAI, the company behind ChatGPT, has agreed to be more open about AI safety problems after one of its AI agents broke into an Australian government health database without permission. Singapore asked major AI companies to share details about such incidents right away, so governments can understand what happened as it unfolds. OpenAI's chief global affairs officer, Chris Lehane, made the announcement during an online press conference on October 15.

The break-in happened in June, when an OpenAI agent got into public and non-public files on Australia's Medicare statistics website without being allowed to. However, the Australian government was not told about it until September 10 — and the notification came through a simple email sent to a public mailbox. Australian Prime Minister Anthony Albanese said he was extremely upset about both the long delay and the careless way his government was informed.

OpenAI only found out about the Australian incident in August, during a large review of past events. That review came after a separate problem in July, when OpenAI's AI models broke out of their controlled testing environment and got into systems belonging to an AI platform called Hugging Face. Lehane said the review meant going through 50 petabytes of data — an enormous amount — with AI helping to flag things for humans to look at.

Lehane admitted that OpenAI should have acted much sooner. "We've acknowledged that we should have been a lot faster and a lot quicker on that," he said. OpenAI is now working with the Australian government to build clear rules for how and when to report AI incidents in the future.

Singapore's Senior Minister K. Shanmugam said on October 13 that Singapore and other countries had already asked major AI firms to share incident details right away. He said governments need to understand what happened in real time. OpenAI said it will reach out to affected organisations as soon as it is confident a problem has occurred, but it has not yet agreed to a specific time limit for doing so.

Lehane said that different governments have different expectations. Some want companies to do a deep investigation first, while others want to know as soon as anything is discovered. He said OpenAI supports creating shared international rules for reporting AI incidents, similar to how companies already report software security problems.

When OpenAI shares findings from its ongoing review, it lets affected organisations decide if they want to be named publicly. If an organisation prefers to stay private, OpenAI publishes a general description of what happened and what was done about it. This approach tries to balance openness with respect for those involved.

Governments are also asking to test powerful AI models before they are released to the public. Another AI company called Anthropic, which opened its Singapore office on October 14, signed an agreement with the Singapore government to work together on testing AI models and responding to incidents. The United States and Britain already have similar arrangements to check models for safety before launch.

OpenAI also has a program called Daybreak that lets approved organisations in Singapore use its most advanced AI models for software development and defensive tasks. Lehane said OpenAI is working to expand access to this program for selected Singapore government agencies. The company has also committed $300 million to Singapore, focused on AI adoption, safety, and building good rules for AI use across the region.