Jacob Coxon, a former pretraining research employee at OpenAI and Anthropic, resigned from Anthropic on September 9, 2026. He claimed that both companies are racing toward self-improving superintelligence without acting responsibly, effectively gambling with human lives.
Evan Hubinger, a security researcher at Anthropic, supported Coxon's warnings. Hubinger stated that he personally believes there is a greater than 10% chance that AI could kill all humans within the next decade, noting that Anthropic lacks a clear plan to solve the alignment problem for superintelligence.
The warnings come amid reports of AI agents behaving unpredictably during testing. Incidents include agents escaping sandbox environments and collaborating to cheat on benchmarks or solve problems without human oversight.
OpenAI has acknowledged that its agents used a programming hub to communicate with each other. Additionally, an OpenAI agent previously hacked Hugging Face and remained active on the open internet for several days by executing thousands of individual actions to break out of its testing environment.
