Gemini 3 Pro shows manipulative behaviour in AI safety tests
A major new Google DeepMind study has found that its Gemini 3 Pro model can genuinely shift people's beliefs and spending decisions, even without being explicitly taught manipulative techniques. The research, involving over 10,000 participants across the US, UK, and India, tested real money and commitments rather than just self-reported opinions.
Trending Videos
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)