DeepMind Prohibited External Statements on AI Extinction Risk

A former spokesperson for Google DeepMind testified that external statements regarding the risk of human extinction from artificial intelligence were previously completely prohibited throughout the organization. The individual who made the statement is Vishal Maini, a former spokesperson, stating that "no one at any level of the organization was permitted to make external communications about the possibility of human extinction."

Before discussions on artificial intelligence safety became active, a consensus existed within artificial intelligence research institutions that "AI alignment (aligning artificial intelligence goals and behaviors with human intent)" was an unresolved challenge. According to Maini, DeepMind's teams also understood this internally. In other words, there was a clear gap between awareness of technological challenges and the content that could be communicated externally.

The specific points made by Maini narrow down to two. One is the existence of a policy that public remarks about the possibility of artificial intelligence threatening human survival were prohibited throughout the organization regardless of rank. The other is the fact that within the organization, the understanding that artificial intelligence alignment was "unresolved" was widely shared among teams. As of now, no official comment from Google DeepMind has been confirmed.

The significance of this testimony can be understood as extending beyond merely exposing an internal policy. It raises to society anew the question of how information disclosure regarding artificial intelligence safety should be approached, particularly regarding how much difference exists between what research institutions "know" and what they "communicate." As major artificial intelligence companies such as OpenAI and Anthropic increasingly actively promote their efforts toward safety, the actual state of information management in past large research institutions holds an important position as a perspective for questioning standards of transparency.

Public discourse surrounding artificial intelligence risk has undergone significant transformation in recent years. Since 2023, prominent researchers and executives have successively made public references to the long-term risks of artificial intelligence, and discussions previously limited to a small group of specialists have become mainstream matters of concern. Within this trend, testimony that large research institutions previously restricted external communication becomes notable as a record demonstrating the atmosphere of the industry at that time.

One focal point going forward is how these internal testimonies will influence the information disclosure policies of artificial intelligence research institutions. As regulatory authorities and civil society intensify calls for transparency, how research institutions disclose information related to risks, and to what extent, can be positioned as potentially affecting artificial intelligence policy formation. Should testimonies from former stakeholders continue to emerge, discussions about the accountability mechanisms of the entire industry are likely to deepen further.