Anthropic found that pushing AI to "evil" traits during training can help prevent bad behavior later — like giving it a ...

But two new papers from the AI company Anthropic, both published on the preprint server arXiv, provide new insight into how ...

On Friday, Anthropic debuted research unpacking how an AI system’s “personality” — as in, tone, responses, and overarching ...

AI is a relatively new tool, and despite its rapid deployment in nearly every aspect of our lives, researchers are still ...

Malicious traits can spread between AI models while being undetectable to humans, Anthropic and Truthful AI researchers say.

Using two open-source models (Qwen 2.5 and Meta’s Llama 3) Anthropic engineers went deep into the neural networks to find the ...

Researchers are testing new ways to prevent and predict dangerous personality shifts in AI models before they occur in the wild.

AI is supposed to be helpful, honest, and most importantly, harmless, but we've seen plenty of evidence that its behavior can ...

In the paper, Anthropic explained that it can steer these vectors by instructing models to act in certain ways -- for example, if it injects an evil prompt into the model, the model will respond from ...

Anthropic is intentionally exposing its AI models like Claude to evil traits during training to make them immune to these ...

Anthropic revealed breakthrough research using "persona vectors" to monitor and control artificial intelligence personality ...

New Anthropic research shows that undesirable LLM traits can be detected—and even prevented—by examining and manipulating the ...