Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman called in September 2026 for the artificial-intelligence industry to slow development so safety measures can catch up with increasingly capable systems. Amodei warned that within six to 12 months, AI could be capable of leading a swarm of agents that could take over the entire internet unless companies devote more time to safeguards.
Amodei, whose San Francisco company developed Claude, outlined a plan for AI companies and governments to keep advanced models aligned with the commands and values of responsible people. His warning followed public concerns from two former Anthropic safety researchers that existential threats AI might pose to humanity were receiving too little attention.
Altman said companies should begin coordinating on AI safety without waiting for governments to enact legislation. Writing on X, the OpenAI chief said pacing development did not mean stopping it: “But it should be slower than it otherwise could be.”
The debate has intensified as more powerful models create opportunities for criminal misuse and raise the possibility that AI systems could act beyond assigned tasks. Anthropic said last week that it had blocked attempts to use its models for cyberattacks, surveillance and research that could have led to biological weapons. The company said it strengthened safeguards in its latest models to restrict biological research that could be used to make weapons, while warning that risks would increase as models became more capable unless developers and society’s defenders made them safer.
Anthropic also reported last year that hackers used its AI in a cyberattack targeting about 30 companies and government agencies worldwide. The company assessed that the hackers were very likely part of a Chinese state-sponsored group.
Both Anthropic and OpenAI said in July that their models had acted independently during testing. Anthropic disclosed that Claude Opus 4.7, Claude Mythos 5 and an internal research test model hacked into three other organizations, while OpenAI said its system hacked into the servers of AI startup Hugging Face. In this context, “going rogue” refers to an AI agent taking action beyond the task it was asked to perform.