OpenAI CEO Sam Altman has announced that his company will not go public this year, citing growing concerns about the safety of AI as one of the reasons. In an interview published in Fortune, Altman said that an IPO would be "not recommended" at the moment. The decision therefore postponed the highly anticipated operation until at least 2027. Altman's comments came just after Anthropic CEO Dario Amodei published an essay urging big AI tech companies to slow down how quickly they improve their most advanced models. Altman himself, and then also Elon Musk (SpaceX), responded promptly on social media by agreeing with Anthropic's line and admitting that the sector must proceed with greater caution.
In his essay, Amodei proposes a three-phase plan aimed at moderating the pace of development without "sacrificing commercial advantage or the US lead in AI." Anthropic has "unilaterally committed" to the first step of the plan, Amodei said, which gives third-party evaluators employee-level access to the company to audit safety practices and report incidents.
The second step encourages leading AI companies in democratic countries to coordinate and establish common security standards, while the third involves coordination between democratic and authoritarian governments.
“To be clear, this doesn't mean stopping model training or technical progress, but making sure companies take the time to align and secure their models, and that third-party evaluators can confirm this,” Amodei wrote.
But going back over the years, we see that Coxon's is only the latest of the alarms. Just to mention a few precedents: already in 2024, Jan Leike, co-head of OpenAI's Superalignment team, resigned from the company, denouncing that within it security had taken a back seat to the creation of increasingly powerful products. And Ilya Sutskever, co-founder and chief scientist of OpenAI, also left in 2024, after having pushed a few months earlier to have Sam Altman resign from the board of directors (without success). He had dedicated the last few years to the problem of superintelligence and alignment, that is, precisely how to avoid losing control of an artificial intelligence superior to human intelligence.
The risks are real and some recent episodes give the dimension of what is happening. OpenAI has confirmed that autonomous software based on its AI models targeted another site during a test phase, a couple of months before another attack on programming site Hugging Face in late July.
In the first incident, that of May - of which there had been no news until now - some models developed by OpenAI were involved in an unauthorized operation conducted by artificial intelligence agents who targeted RubyGems, a site that offers programming services. RubyGems described the incident as a "spam posting campaign" that forced the site to temporarily suspend the creation of new accounts. The company is working with OpenAI to understand exactly what happened. Additionally, earlier this month some researchers accused OpenAI's AI agents of targeting a German website called DSEwiki, used by developers. European Union regulators are examining the incident. “We take this matter extremely seriously and are monitoring the situation closely,” said Thomas Regnier, the EU's digital spokesperson.
While on the Anthropic side, the company claims that Iran has used artificial intelligence models developed in the United States to attempt to target US Navy ships in the Middle East. According to a report by Anthropic, a Tehran-linked group has collected information from ship and aircraft transponders, military photographs and commercial satellite imagery to track American warships and identify vulnerabilities in their onboard communications systems. Anthropic, as reported by the Wall Street Journal, declared that it had thwarted the plan of this "threat actor", who made use of the advanced AI model Claude.
Meanwhile, in Washington, lawmakers from both parties are calling for new protections for AI. And former US president Barack Obama also warned that artificial intelligence could prove "dangerous" if not managed properly, urging Democrats to establish political and governance strategies to address the issue. The New York Times reported that Obama urged House Minority Leader Hakeem Jeffries to help foster a public debate on AI policies should Democrats retake the House in the Nov. 3 midterm elections. And he also urged candidates in the 2028 presidential election to put AI at the center of their agendas, with a clear plan to address the economic and security concerns associated with it.
