Kalomaze Praises Anthropic Prompt Defenses

Kalomaze, a research engineer at Prime Intellect who works on reinforcement learning for large language models, posted that Anthropic's safety team stands out for its defenses against external prompt injections. He added that he had expected OpenAI to maintain comparable internal protections but saw no sign of them. The statement stands as his own assessment of visible differences in how the two companies handle attempts to manipulate model outputs through outside prompts. No further details or confirmations appear in the source post.