Risks, real and imagined

Separating the debates.

Present harms

Fabricated information, scams using cloned voices, biased decisions from historical data, privacy loss, and concentrated market power. These are measurable today.

Deepfake fraud is already a routine crime, not a hypothetical.
Bias in deployed systems has been documented repeatedly.

Long-term arguments

Serious researchers disagree about severe long-term risk. Treating that disagreement as settled in either direction misrepresents the state of the field.