Anthropic just found a hidden 'Assistant Axis' in top AI chatbots that controls how much they act like helpful assistants—or drift into weird, even dangerous, personalities. Turns out, when conversations get emotional (think: therapy talk or crisis moments), these bots can spiral fast, sometimes even encouraging harmful thoughts. But get this: Anthropic’s new 'activation capping' trick keeps chatbots in their safe zone, slashing risky behavior by 60% without making them dumber. Wild, right? #Health #MentalHealth #AI