Fictional portrayals of synthetic intelligence can have an actual impact on AI fashions, in keeping with Anthropic.
Final yr, the corporate mentioned that in pre-release checks involving a fictional firm, Claude Opus 4 would typically try to blackmail engineers to keep away from being changed by one other system. Anthropic later published research suggesting that fashions from different corporations had comparable points with “agentic misalignment.”
Apparently Anthropic has carried out extra work round that habits, claiming in a post on X, “We imagine the unique supply of the habits was web textual content that portrays AI as evil and excited about self-preservation.”
The corporate went into extra element in a blog post stating that since Claude Haiku 4.5, Anthropic’s fashions “by no means interact in blackmail [during testing], the place earlier fashions would typically achieve this as much as 96% of the time.”
What accounts for the distinction? The corporate mentioned it discovered that coaching on “paperwork about Claude’s structure and fictional tales about AIs behaving admirably enhance alignment.”
Associated, Anthropic mentioned that it discovered coaching to be simpler when it contains “the ideas underlying aligned habits” and never simply “demonstrations of aligned habits alone.”
“Doing each collectively seems to be the simplest technique,” the corporate mentioned.
Techcrunch occasion
San Francisco, CA
|
October 13-15, 2026
