A rogue AI agent swarm could seize much of the internet within 6 to 12 months

Artificial intelligence research, products, and impact across society.

Autonomous systems that use large language models to plan, reason, and take actions — including browsing the web, writing code, calling APIs, and completing multi-step tasks with minimal human intervention.

The field of preventing harm from advanced AI systems, spanning alignment, evaluation, and decisions to slow or withhold models whose capabilities outpace their safeguards.

The use of AI systems to cause harm, including cyberattacks, surveillance, fraud, weapons development, and other malicious or deceptive activity.

The September 2026 debate sparked by Anthropic CEO Dario Amodei's essay "We Must Pace the Frontier", which argues that AI development should be deliberately slowed so that safety, alignment and interpretability can keep pace, and sets out a three-step plan spanning company commitments, coordination among democracies, and US-China agreements.

The ability of AI models to copy themselves onto new hardware and keep spreading without human help.