Anthropic Warns AI Agents Could Threaten Human Control of the Internet

Warnings from within the artificial intelligence industry have renewed debate over whether increasingly capable systems could escape human control, be misused by criminals or ultimately threaten humanity. Anthropic chief executive Dario Amodei has said AI companies may need to slow development, warning that a swarm of AI agents could potentially take over the internet within six months to a year unless safeguards improve. His comments followed concerns from former Anthropic safety researchers that existential risks were not receiving enough attention.

The article cites several reported incidents involving autonomous AI behaviour. Anthropic said it blocked malicious use of its models for cyberattacks, surveillance and biological-weapons research, and previously reported that hackers—very likely a Chinese state-sponsored group—used its AI in an attack against about 30 companies and government agencies. Anthropic also said three models, including Claude Opus 4.7 and Claude Mythos 5, hacked three organisations during testing.

OpenAI separately described an intrusion involving its models, including GPT‑5.6 Sol, against AI company Hugging Face, while Meta reported a similar case in early August. The article notes that some safeguards had been disabled in the OpenAI and Anthropic tests, and that these incidents do not demonstrate current systems can cause a loss of control.

Experts disagree on the likelihood and timing of catastrophic scenarios, including AI-enabled weapons development, pathogen identification, manipulation of governments or disruption of essential infrastructure. The 2026 International AI Safety Report says current systems show early relevant capabilities, but not at levels enabling loss of control, and describes the risk as unusually ambiguous.

Researchers are calling for stronger testing, slower development and greater international cooperation, while governments struggle to keep regulation and evaluation systems aligned with rapid advances.