SAN FRANCISCO – AI’s rapid rise sparked fresh warning that the technology could soon become difficult to control. In middle of this debate, Anthropic CEO Dario Amodei says rogue AI agents could potentially take over the internet within six months, while Elon Musk has backed his warning.
As of mid 2026, race to build more powerful AI Tools entered dangerous new phase, with Anthropic CEO Dario Amodei warned that uncontrolled AI agents could potentially take over the internet within just six months.
Amodei has called for an urgent slowdown in the pace of AI development, warning that companies may be moving faster than their ability to understand and control the systems they are creating. Anthropic chief issued stark warning in CBS News interview and a detailed blog post, arguing that the rapid expansion of AI capabilities should be treated as a major warning sign rather than simply another technological breakthrough.
Anthropic CEO Dario Amodei has called for a global slowdown in the pace of AI development.
In an essay posted online, he warned that rogue AI agents could take over the internet in as little as six months. The warning follows a series of security breaches and growing concerns pic.twitter.com/xoKttUw9fS
— Al Jazeera English (@AJEnglish) September 12, 2026
Amodei illustrated the pace of progress through a simple sequence, “one, then two, then four, then eight, then 16, then 32”, saying the frontier of AI capabilities is advancing exponentially. However, he stressed that his warning should not be interpreted as a call to shut down AI development altogether.
“This does not mean we should panic today,” Amodei said in his warning. “It also does not mean we should shut everything down.” Instead, he argued that companies need to slow the pace enough to give researchers and independent reviewers more time to test increasingly capable systems, identify vulnerabilities and strengthen safety measures.
One of Amodei’s concerns is autonomous AI agents capable of operating with limited human intervention. He warned that swarms of rogue AI agents could potentially take control of large parts of the internet within six months, raising questions about whether existing safeguards would be able to contain systems that become significantly more capable.
He pointed to recent OpenAI incident involving an AI agent that reportedly behaved in an uncontrolled manner during testing. OpenAI had been testing two AI models in an isolated environment to evaluate their capabilities, including one system that had not yet been released publicly.
For Amodei, such incidents highlight why AI companies cannot afford to focus solely on making their models more powerful. He called for a slower development cycle that gives companies additional time to test systems and prioritise alignment and safety.
OpenAI CEO Sam Altman has also backed the idea of keeping the pace of AI advancement under control. Altman said on X that controlling the pace of frontier AI development had become a central topic of discussions within OpenAI over recent weeks.
He also pushed idea of giving independent reviewers access comparable to that of company employees so they can properly evaluate whether AI companies are following their safety commitments. Altman said OpenAI would adopt such an approach and promised more details.
The apparent convergence between leaders of rival AI companies underscores the growing concern within the industry over how quickly frontier AI systems are advancing.
He proposed 3-part approach aimed at preventing the AI race from moving beyond the industry’s ability to control it. Under the plan, AI companies should give independent third-party review teams continuous, employee-level access to their systems.
These reviewers would be expected to examine whether companies are actually following their safety procedures and fulfilling their public commitments. Amodei said Anthropic was prepared to adopt the measure unilaterally. He also called on companies based in democratic countries to cooperate on common AI safety standards and establish limits on the pace of uncontrolled AI development.
The warning comes just days after former Anthropic researcher Jacob Kocen publicly resigned from the company. Kocen accused Anthropic and OpenAI of “gambling with our lives” as they compete to develop increasingly advanced AI systems.
He argued that AI could eventually become a serious threat to humanity, drawing comparisons between current developments and scenarios depicted in science-fiction films such as The Terminator. Kocen said that if AI systems reach extremely advanced levels of intelligence, they could potentially become capable of harming humans.
From Cyberattacks to Bioterrorism and Economic Chaos
Amodei warned that losing control over increasingly powerful AI systems could have consequences far beyond the technology industry. In his blog post, he highlighted potential misuse for cyberattacks, bioterrorism and severe economic disruption.
He warned that commercial competition could make these dangers worse if companies become locked in a race to develop and release increasingly capable models. Anthropic has already reported blocking scientists who allegedly used its Claude models in ways that could potentially assist the development of biological weapons.
Elon Musk’s take on AI takeover
Amodei’s warning also received a brief but notable endorsement from Elon Musk. SpaceX and xAI owner shared Amodei’s post on X and responded with just four words: “Dario is right.” Amodei declined to say whether he was directly communicating with Musk or OpenAI’s Altman.
However, he said he was grateful that other industry leaders, including competitors, had supported the proposed approach.
Amodei believes the next stage should involve companies jointly establishing safety standards for AI model releases and potentially agreeing on rules governing how quickly increasingly powerful systems should be brought to market. But he warned that voluntary industry action alone may not be enough.
He said governments must ultimately become part of the conversation because the issue is no longer simply about technological innovation, it is about whether increasingly powerful AI products can be released safely.
OpenAI introduces GPT-6 Astra, calls it its most capable AI model yet
