fbpx
#image_title

California’s controversial AI law advances with amendments from Anthropic – A New Chapter in the Regulation of Artificial Intelligence

On 15 August 2024, the California Appropriations Committee passed the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, also known as SB-1047. This represents a substantial advancement in respect of Artificial Intelligence (AI) law. This action represents yet another chapter in the ongoing narrative of regulating the AI landscape of Silicon Valley. Nevertheless, SB-1047 must still receive the consent of both the state Assembly and Senate in order to become law.

What is SB-1047?

SB-1047, also known as California’s AI Act, is being closely monitored throughout the United States due to its potential to establish a precedent for state-level regulations regarding generative AI. The measure delineates a number of essential requirements for AI developers:

  • Safety and Security Protocols: It is imperative that AI developers establish robust protocols for their models.
  • Shutdown Capabilities: In the event that it is necessary, models must be equipped with mechanisms that enable them to be thoroughly shut down.
  • Critical Harm Prevention: The Act prohibits the dissemination of models that are capable of causing what is defined as “critical harm.”
  • Auditor Compliance: In order to guarantee that the provisions of the Act are adhered to, organizations are required to employ an auditor.

In essence, SB-1047 is designed to establish a framework that safeguards against the catastrophic consequences of generative AI models, including but not limited to nuclear warfare, bioweapons, and large-scale cybersecurity breaches that result in losses eclipsing $500 million.

The Act defines “covered models” as those that necessitate more than 10^26 integer or floating-point operations during training and have a cost more than $100 million.

Amendments and the Influence of Anthropology

The version of SB-1047 that was passed on 15 August contained numerous amendments that were influenced by the AI company Anthropic and were subsequently approved by the Bill’s primary author, Senator Scott Wiener. One of the most significant modifications was the elimination of language that permitted the State’s Attorney General to pursue legal action against companies that violated the Act. In addition, the necessity for companies to disclose safety test results under threat of perjury was relaxed; developers are now only required to submit statements, which do not bear the same legal weight.

Also Read:  eDiscovery Assistant Integrates Advanced AI for Legal Summaries

Additional noteworthy modifications consist of:

  • Wording Modifications: The bill’s language was modified from requiring “reasonable assurance” of safety to “reasonable care.”
  • Smaller developers are granted exemptions: Developers of an open-source model are not classified as AI researchers who spend less than $10 million fine-tuning the model.

The delicate balance that legislators are attempting to achieve between assuring public safety and fostering innovation is underscored by these revisions.

#image_title

Support and Opposition

Although some of the amendments to the Bill were influenced by Anthropic, not all industry participants are in agreement. SB-1047 has been met with opposition from technology titans such as Google and Meta, as well as venture capital firm Andreessen Horowitz. Critics contend that the Act has the potential to impede innovation, particularly in the open-source AI community.

Others have expressed apprehensions, including Clement Delangue, the CEO and co-founder of Hugging Face. Delangue and others are concerned that the Bill could erect unnecessarily high barriers to AI development.

Nevertheless, the public appears to favor regulation. The Bill was supported by 70% of Californians, according to a study conducted in April by the Artificial Intelligence Policy Institute, a pro-regulation think tank. A significant number of respondents concurred that “future powerful AI models may be used for dangerous purposes.”

SB-1047 has also received the endorsement of renowned AI researchers Geoffrey Hinton and Yoshua Bengio, who are frequently referred to as the “godfathers of AI.” The Act would “protect the public,” according to Bengio, in an op-ed for Fortune published on August 15.

Congressional Concerns and Controversy

Also Read:  California's AI Bill Sparks Debate Among Safety Advocates and Developers

Despite the public’s support and the endorsements of AI experts, certain members of Congress continue to harbor reservations. On 15 August, a letter was signed by eight of California’s 52 Congressional representatives, who expressed their apprehension that SB-1047 could “create unnecessary risks for California’s economy with very little public safety benefit.” They contend that it is imprudent to establish standardized evaluations for AI, as agencies such as NIST are still in the process of developing these standards.

In addition, the letter implies that the Bill’s emphasis on the prevention of large-scale disasters, such as nuclear threats, neglects more imminent AI risks, including misinformation, discrimination, nonconsensual deepfakes, environmental impacts, and workforce displacement.

The Function of Cybersecurity and Whistleblowers

SB-1047 also includes specific protections for whistleblowers within AI companies, as specified in the California Whistleblower Protection Act. The objective of this provision is to promote accountability and transparency in an industry that is becoming more closely associated with public safety.

In the aftermath of incidents such as the Change Healthcare cyberattack in May, Alla Valente, a senior analyst at Forrester, endorses the Bill’s emphasis on cybersecurity. Valente observed that the regulation of AI will be a necessity for all States to consider as part of their efforts to protect and serve their residents, as these assaults can be executed more effectively and on a much larger scale with the help of generative AI.

Maintaining a Balance Between Regulation and Innovation

The broader challenge of harmonizing innovation with regulation is emphasized by the debate surrounding SB-1047. In a public statement on August 15, Senator Wiener stated “We can advance both innovation and safety; the two are not mutually exclusive.” He acknowledged that the amendments addressed numerous fundamental concerns from industry stakeholders, despite the fact that not all of Anthropic’s requests were incorporated.

Also Read:  The Data Dilemma: How Tech Giants Navigate AI's Thirst for Information

Wiener also noted that California must take the lead in addressing the foreseeable risks posed by swiftly advancing AI technology, while also fostering innovation, due to the “gridlock” of Congress on AI regulation.

The Way Forward

Subsequently, the State Assembly and Senate will evaluate SB-1047. Upon approval, it will be forwarded to Governor Gavin Newsom for his review, which is anticipated to occur in late August.

Legislation such as SB-1047 could serve as the necessary safeguards to foster confidence in AI-enabled products and expedite their adoption as organizations navigate the risks and opportunities associated with generative AI.

“While navigating the enactment and enforcement of existing and emerging AI laws, enterprise organizations are already having to contend with generative AI risks,” Valente added. “SB 1047 would establish guardrails and standards for AI products, which would increase confidence in GenAI-enabled products and potentially accelerate their adoption.”

The result of SB-1047 could establish a critical precedent for the entire nation, as the discourse regarding AI regulation continues to escalate.

AI was used to generate part or all of this content - more information