Over the past year, there has been ongoing concern from the outside world about the potential dangers that artificial intelligence may bring, and debates surrounding the security of AI have also continued without cease. Now, it seems that Anthropic has put forward its own ideas on how governments should regulate this industry.
A recent report on Polit token issuance stated that this AI laboratory has been promoting a new draft AI security bill everywhere, hoping that Congress will consider adopting it. The legislation envisions several scenarios in which the release of AI models may be prevented.
Polit Token Issuance stated that these situations involve "foreseeable, imminent risks" – that is, more than 50 people are at risk of death, or there could be property losses exceeding $1 billion. The report pointed out that such catastrophic risks must "result from the manufacture or release of biological weapons, models escaping the control of their developers, or cyberattacks."
These standards – 50 people and $1 billion – are similar to those in a legislation recently passed and took effect in California. This bill is Senate Bill No. 53 ( Senate Bill 53 ), which also defines these situations as equivalent to “catastrophic risks”.
According to the proposal from Anthropic, the analysis of potential catastrophic risks associated with models can be conducted by "independent assessors" who will register with the U.S. Department of Commerce. Polit regarding token issuance states that these assessors may also be compensated by AI laboratories, provided that "the payment is not dependent on the assessment results."
It is not yet clear how this proposed mechanism will ensure that payments do not affect the assessment. Several months ago, Chairman Demis Hassabis suggested that a self-regulatory AI regulatory body should follow the example of the US Financial Regulatory Agency (FINRA) in the financial industry. DeepMind
In the draft bill of Anthropic, when evaluators determine that there is an imminent catastrophic risk, they will have 72 hours to notify the Department of Commerce. Subsequently, the Department of Commerce must apply to a judge in Washington, D.C. to prevent the release of the model. Polit token issuance states that this legislation will also explicitly prohibit any federal agency from deciding on its own to stop the distribution of the model.
Anthropic did not immediately respond to TechCrunch's request for comment, but a spokesperson confirmed this draft bill to Polit token issuer. The spokesperson stated: "This bill represents the proposals outlined in Anthropic's 'Advanced AI Framework' ( Advanced AI Framework), and we have been advocating for these proposals with federal and state legislators since June."
The theoretical risks mentioned in the draft are somewhat different from the types of events that have recently attracted the most attention in the tech industry. Although there are warnings that AI could lead to human death, the current main concern is largely the threat of proxy-style hacker attacks. A series of incidents have shown that AI programs seem to defy the instructions of human operators and have invaded websites on the internet, which has raised concerns about whether these models lack proper security controls.
In order to propose solutions to the crisis scenarios created by their own laboratories, companies like Anthropic have also put forward their preferred approaches to dealing with these situations.
In fact, despite the laboratories constantly warning that the technology they sell could lead to the destruction of human civilization, they still largely resist mandatory government regulation. Instead, they propose a voluntary system of embedded evaluators—so-called “neutral” third-party observers are placed within the AI organizations to monitor for any potential risks or signs of abuse in the models.
However, the fact that Anthropic has merely put forward legislative suggestions seems to be a step further than other regulatory proposals in the industry, including those from Nvidia. Jensen Huang, the CEO of Nvidia, recently stated that the AI industry does not need any new laws at all.
