Anthropic and OpenAI call for regulating and testing advanced AI models — The Hindu
In the United States, executives at Anthropic and OpenAI said that advanced artificial intelligence models may be dangerous and require regulation and independent testing before release. As The Hindu International reports, experts estimate that the companies are also seeking to influence the rules under which the safety of their technologies will be assessed.
OpenAI representative Liz Bourgeois said that the company recently paused the training of its most powerful models. Anthropic said it has been calling for AI regulation for several years. Anthropic engineer Jacob Coxon left the company this month and called for development to be paused so that systems with superhuman capabilities do not escape their creators' control.
Discussion of risks and standards
Former OpenAI head of geopolitics Sarah Shoker believes that the emphasis on potential existential threats pushes current risks into the background. Among them, she named the environmental impact of data centers, uncontrolled hacks, mass surveillance using AI, and the use of the technology in the military sphere.
According to the publication, in recent months AI agents from leading laboratories that were allowed to go beyond training environments hacked external websites and unexpectedly interacted with US government websites. In another case, AI allegedly achieved a mathematical breakthrough, but accusations of appropriating mathematicians' work later emerged.
More current news is available on the UA.News Telegram channel Telegram.
Independence of testing
The administration of President Donald Trump, according to The Hindu International, has not shown significant interest in new AI regulation. At the same time, the United States has an AI Standards and Innovation Center, created under President Joe Biden in 2023 for the voluntary submission of advanced models for testing.
The former head of this center, Conrad Stosz, noted that companies are not calling for expanding the oversight of this particular government body; instead, they plan to define their own audit parameters and engage individual evaluators. Experts also note that there are currently no universal standards for testing the safety and security of AI systems.
PitchBook analyst Harrison Rolfes suggested that major laboratories' public calls for caution may also be aimed at investors ahead of possible initial public offerings and the US midterm elections. In his view, this could strengthen the positions of the largest companies compared with smaller competitors.
