SB 1047 101: What You Need to Know About California’s Groundbreaking AI Safety Bill
As artificial intelligence continues to evolve, becoming more powerful and integrated into our daily lives, the risks associated with advanced AI models are growing just as rapidly. California’s SB 1047, also known as the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, seeks to address these risks and ensure that developers of the most powerful AI systems are held accountable for the safety and security of their technologies. This post will walk you through what SB 1047 is, what it does, and why it is so crucial to our global safety and economy.
What is SB 1047?
SB 1047 is a bill designed to protect society from the potential catastrophic risks posed by frontier AI models—these are the kinds of AI systems that have the power to influence critical sectors like infrastructure, the economy, and even global security. The bill aims to regulate developers who create AI models using vast amounts of computing power or financial resources. Specifically, it targets AI models built using over $100 million in computing power or those altered with over $10 million in fine-tuning.
The goal is simple: to ensure that these powerful AI models do not cause mass harm, such as cyberattacks on critical infrastructure or even unintended use in weapons systems.
Key Provisions of SB 1047
1. Safety Protocols: SB 1047 requires developers of these large AI models to create and follow a Safety and Security Protocol (SSP). This includes measures for cybersecurity, testing procedures, and safeguards to prevent harmful capabilities from emerging in the AI systems.
2. Annual Audits: Developers must undergo third-party audits to ensure that their SSPs are effective and followed rigorously. A redacted version of the SSP must also be made public to maintain transparency.
3. Kill Switch: One of the more discussed features of the bill is the requirement for a "kill switch", allowing developers to shut down any AI model they control if it presents an imminent risk of harm. This provision ensures that emergency situations can be managed before they spiral out of control.
4. Whistleblower Protections: To encourage accountability, the bill includes protections for whistleblowers who expose violations of the law.
5. Know-Your-Customer Requirements: The bill introduces export control measures for AI training resources, ensuring that cloud computing used for AI development is tracked and regulated.
6. A Narrow Focus on Large AI Developers: Importantly, SB 1047 only applies to the largest, most resource-heavy AI developers. Small businesses, startups, and academic institutions are exempt unless they are using enormous amounts of computing power. This means that innovation in smaller enterprises will not be stifled by the bill’s regulations.
Addressing Misrepresentations
Like any significant legislation, SB 1047 has sparked a heated debate, with several misrepresentations clouding the conversation. Some critics argue that developers would be required to guarantee their models' safety with "full certainty," which is virtually impossible. However, the bill only requires “reasonable care”, a standard already used in many industries and backed by centuries of legal precedent.
Another common misrepresentation involves the penalty of perjury for developers who submit inaccurate safety documents. Critics claimed this could lead to jail time for developers making honest mistakes. However, SB 1047 was amended to replace perjury with civil liability, meaning developers can only be penalized if they knowingly provide false information.
The kill switch provision has also raised concerns, especially in the open-source AI community. However, the bill only requires a kill switch for models still controlled by developers, not those that have been publicly released or altered beyond their control.
Why is SB 1047 crucial?
The risks posed by AI are no longer just theoretical. As AI models grow more powerful, they could potentially cause unprecedented harm, such as cyberattacks that shut down critical infrastructure or systems that could be misused as weapons. SB 1047 ensures that these frontier AI models are developed responsibly and with safeguards in place to protect society from risks we can’t fully predict today.
This bill is not about stifling innovation—it’s about protecting our global economy and communities from the unintended consequences of AI. By regulating only the largest developers, it leaves space for smaller innovators while ensuring that those with the most powerful technologies are held accountable.
The Clock is Ticking
Governor Gavin Newsom has until Monday, September 30th 2024, to either sign or veto SB 1047. His decision will be critical not just for Californians but for the entire world. As AI becomes more integrated into global systems, the need for regulations like SB 1047 becomes increasingly urgent. The bill could set a precedent for AI safety legislation worldwide, ensuring that as AI grows, it does so in a way that is responsible and secure.
The question now is: Will Governor Newsom sign on Monday to protect us, or will he put our lives at risk by accepting millions from Big Tech? Only time will tell. The World is watching.
Let’s continue the conversation about responsible AI and push for the kind of regulation that protects both innovation and public safety.
#DemandEthicalAI #SignSB1047 #ProtectOurFuture
