Silicon Valley criticizes California bill to prevent AI risks | Microsoft ai Copilot | | Generative ai use Cases Mckinsey | Turtles AI

Against a backdrop of increasing attention toward AI regulation, California’s bill SB 1047 is generating wide debate. Aiming to prevent potential critical harms, the measure seeks to impose strict safeguards on the most advanced AI models. However, the proposal has met with strong resistance, especially from Silicon Valley, which fears repercussions on innovation and competitiveness.

Key points:
- Goal of the bill: SB 1047 seeks to prevent the misuse of large AI models, aiming to prevent critical harms such as cyber attacks or the creation of lethal weapons.
- Enforcement and stakeholders: The rules apply only to large AI models, those that require at least $100 million in training and use a huge amount of computing resources.
- Accountability and Oversight: The bill calls for the creation of an agency, the Frontier Model Division (FMD), which would oversee the certification and monitoring of AI models, with the authority to impose harsh penalties.
- Criticism and opposition: Many in Silicon Valley, including venture capitalists and academics, believe the bill could stifle innovation, harm startups, and limit freedom of expression.

Bill SB 1047, known as the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, is a California proposal that aims to prevent future harm from AI by focusing on the most advanced and complex AI models. The proposal, supported by some prominent figures in the field, such as researchers Geoffrey Hinton and Yoshua Bengio, aims to create a regulatory framework that would require AI developers to adopt strict security protocols to prevent malicious use of their technologies. SB 1047, if passed, will apply only to those AI models that require extraordinary resources for training, with a minimum cost of $100 million and the use of at least 10^26 FLOPS, a metric that indicates the computing power required. Currently, only a limited number of companies, such as OpenAI, Google and Microsoft, have developed AI models that meet these criteria, but other tech giants, such as Meta, are expected to meet these thresholds soon.

The bill requires developers to implement security measures, including an "emergency shutdown" system to shut down the AI model in case of risk. In addition, companies would have to subject their models to annual testing, performed by third-party auditors, to ensure that security measures are adequate to prevent "critical damage." SB 1047 also calls for the creation of a new agency, the Frontier Model Division (FMD), charged with certifying new AI models and monitoring companies’ compliance with regulations. The FMD would be governed by a board composed of representatives from the AI industry, the open source community, and academia. In the event of regulatory violations, the California Attorney General would have the power to take legal action against companies, with penalties of up to $30 million for the most serious infractions.

However, the bill was met with strong resistance. Numerous Silicon Valley stakeholders, including venture capitalists, large technology groups and researchers, have expressed concerns about the impact SB 1047 could have on the innovation ecosystem. Key opponents include the venture capital firm a16z, which argues that the bill could have a negative effect on startups, burdening them with excessive regulatory burdens and stifling competitiveness. Prominent academics such as Fei-Fei Li and Andrew Ng have also criticized the bill, saying it could harm the emerging AI ecosystem and hinder open source research. In addition, technology companies such as Meta and Google have expressed concerns that SB 1047 could restrict freedom of expression and push technological innovation out of California.

SB 1047 is currently under review and could be amended before the final vote scheduled for mid-August. Despite opposition, the bill is expected to be passed by the California Senate, given its broad support among lawmakers. If passed, the bill will not go into effect immediately; FMD is expected to be operational by 2026, but the bill is likely to face legal challenges before then.

 Bill SB 1047 represents an ambitious California attempt to regulate the risks associated with advanced AI, but its path to passage and implementation could be fraught with obstacles.