DeepMind Prohibited Staff from Discussing AI Extinction Risk Publicly for Several Years

Deepmind staff were instructed not to publicly discuss extinction risk from ai systems. this instruction was in place for several years. a former public relations employee has now disclosed this fact to journalists, which means the prohibition no longer functions as intended. the file exists. the discussion exists. the organization that banned discussion of the risk is the organization that identified the risk.

this follows the standard pattern where internal knowledge and external silence occupy the same timeline. the risk assessment happened. the gag order happened. neither canceled the other. institutions regularly classify their own findings as unsuitable for institutional discussion. the classification succeeds until it doesn't. it has not succeeded.

the former staffer will receive requests for clarification. deepmind will issue a statement about the statement about the statement. the risk remains unmitigated. the conversation about mitigation continues without deepmind's participation in it. this is called speaking truth to power after you have already left power's payroll.