Newsletters
OpenAI and Anthropic are intensifying warnings about the potential risks posed by advanced artificial intelligence, even as both companies continue releasing increasingly powerful models and compete for investment ahead of anticipated stock market debuts.
The two San Francisco-based AI developers have recently focused on the possibility of "recursive self-improvement," a scenario in which AI systems could design and train future generations of AI with minimal human involvement.
Anthropic described the prospect as plausible but uncertain last week, while OpenAI went further, predicting Monday that such a milestone could arrive by March 2028.
OpenAI has framed the development as a major technological objective, arguing that AI systems capable of conducting research autonomously could dramatically accelerate scientific discovery. The company has also promoted a vision of personal AI assistants for every individual, describing the technology as potentially more transformative than inventions such as electricity or the automobile.
Like previous transformative technologies, OpenAI argues that AI can be managed safely through the creation of appropriate safeguards, comparable to the development of seatbelts, driver's licenses and speed limits.
Anthropic has similarly highlighted AI's potential to drive advances in medicine, technology and economic growth. However, the company has adopted a more cautious tone, warning openly about the possibility of a "loss of control" over increasingly capable systems.
"You want the option to be able to take your foot off the gas and put your foot on the brake," Anthropic co-founder Jack Clark said last week.
Both companies have suggested that international coordination may be necessary to manage the development of frontier AI systems. The proposals would potentially involve major competitors, including Chinese AI firms and technology companies such as Google.
Anthropic last week proposed a global coordination framework that could temporarily suspend frontier AI development, provided that no participant uses the pause to gain a competitive advantage. Observers have widely viewed that condition as difficult to achieve under current market dynamics.
OpenAI on Monday also floated the idea of a multilateral organization that could "slow frontier development if necessary."
The concept of international oversight of advanced AI predates the latest proposals.
Since 2023, OpenAI has advocated for an international authority modeled on the International Atomic Energy Agency, the United Nations body responsible for nuclear oversight. The company has argued such an institution could help regulate the pace of AI capability growth.
At the same time, competition within the sector remains intense. OpenAI and Anthropic are currently releasing new models at a rapid pace, with updates arriving roughly every four to six weeks.
Anthropic also revised its internal guidelines in February, creating separate categories for behaviors the company refuses to engage in itself and practices it believes the broader AI industry should avoid, even if it is not willing to make unilateral commitments.
Set as preferred source
Set as preferred source
1 min read
1 min read
