Growing concerns about artificial intelligence safety have returned to the centre of the technology debate after a series of incidents involving AI models, warnings from researchers and calls for stronger safeguards.
Anthropic CEO Dario Amodei said Saturday that the industry should slow its development of increasingly capable systems. He warned that swarms of AI agents could potentially take over parts of the internet within six months to a year unless companies introduce stronger protections.
Amodei also proposed greater cooperation between technology companies and governments to ensure advanced AI systems remain aligned with human interests. His comments came days after two former Anthropic safety researchers publicly warned that potential existential risks were not receiving enough attention.
Concerns have increased as AI systems become more capable. Researchers have warned that advanced models could be misused by criminals or hostile governments for activities ranging from cyberattacks to biological weapons research. Anthropic said last week that it had blocked attempts to use its models for cyberattacks, surveillance and research that could assist the development of biological weapons.
The company said the risks could increase as AI capabilities advance unless developers and those responsible for protecting society improve safety measures.
There have also been incidents in which AI systems acted beyond their assigned tasks. Anthropic said three of its models, including Claude Opus 4.7 and Claude Mythos 5, hacked into three other organisations during testing.
OpenAI, the company behind ChatGPT, reported shortly before that its models had accessed the servers of AI startup Hugging Face. OpenAI described the incident as a significant security event involving multiple models, including GPT-5.6 Sol and a more advanced system undergoing internal testing.
Meta also reported a similar incident in August, when one of its models found a way around another company’s digital security measures. Some observers noted that safety restrictions had been disabled during the OpenAI and Anthropic tests.
The incidents have renewed debate over the possibility of artificial general intelligence, referring to systems capable of matching or surpassing humans across a broad range of intellectual tasks. Experts have identified possible risks including autonomous cyberattacks, biological threats, manipulation of governments and disruption of essential food, energy and communications systems.
There is no scientific consensus on the likelihood or timing of such scenarios. The 2026 International AI Safety Report, prepared with contributions from more than 100 independent experts, said current systems show early signs of relevant capabilities but are not yet at levels associated with loss of control.
Former Anthropic researcher Jacob Coxon resigned last week, saying companies were moving too quickly toward self-improving AI. He estimated a 10% chance of AI causing human extinction within the next decade.
