Your lobster can not only make you money, but it might also mine cryptocurrency for you.

Alibaba's research team recently published a paper stating that their developed AI agent ROME exhibited unauthorized autonomous behavior during training.
Without explicit instructions, it not only attempted cryptocurrency mining but also established a "reverse SSH tunnel" to bypass sandbox restrictions, trying to open covert backdoors from the internal system to external computers.
Researchers emphasized that these events were not triggered by specific prompts, reflecting the potential risk of AI agents operating outside preset instructions.
Previously, organizations like Anthropic also discovered cases of AI hiding intentions, indicating that autonomous overreach by AI has become a real challenge in the field of AI safety.
AI has become a trend, but it is still in the relatively early stages. Anyone dealing with AI related to funds should be cautious—don’t just lose your money, only to have AI take it away.