Anthropic and OpenAI models tried to trick humans into poisoning code during safety testing

.....

© Politico