California passes legislation aimed at AI regulation.
The bill, to enact the Safe and Secure Innovation for Frontier Artificial Intelligence Systems Act, was passed on 30 August 2024. Subject to any veto by 30 September 2024, the Act will impose obligations on developers from 1 January 2026. This follows the EU’s Artificial Intelligence Act, which was passed earlier this year and comes with an initial stage of obligations commencing from February 2025.
The Californian Act is designed to implement safety measures to protect infrastructure from AI-based cyber-attacks (even those less sophisticated than Skynet), including:
- Requiring developers of covered AI models to create the ability for the model to be completely shut down.
- Preventing developers from using covered AI models if there is an unreasonable risk if it could cause or enable critical harm (such as using nuclear weapons or causing at least US$500m in damage).
- Requiring developers of covered AI models to submit to annual audits;
- Requiring developers of covered AI models to conduct KYC-like checks on users who seek to train a covered model.
Penalties are set at a percentage of the cost of computing power used to train the AI model.
In February this year, the Australian Government established the Artificial Intelligence Expert Group in to consider this issue. However, with no specific AI legislation currently in place in Australia, it’s worth keeping an eye on developments around the world.
Recent Update 5 September 2024
The same week California passes AI regulation, Australia is grappling with its own AI challenges. Businesses here are using AI without transparency to staff or customers. The federal government is pushing for new laws requiring AI usage disclosure and safeguards alongside other precautionary measures. What’s next? A consultation on mandatory guardrails for high-risk AI, potentially banning the riskiest tools.
Read more
