Dario Amodei Is Right About AI Risk - But Wrong to Wait on Government
Dario Amodei’s essay, “Policy on the AI Exponential,” is important because he clearly admits what many people still do not want to say out loud:
AI is moving faster than government can respond.
He is right about that.
But that is exactly why his answer is not strong enough.
If AI companies know their tools can create serious real-world danger, then responsibility cannot begin and end with asking government to catch up.
That is not leadership.
That is waiting for someone else to force the responsibility that should already exist.
Think about any other industry.
If a ______ company said:
“Our products may create serious public danger, but government has not regulated us fast enough yet,”
would we accept that?
A weapons manufacturer?
A pharmaceutical company?
A chemical manufacturer?
A banking platform?
A car company?
A payment processor?
A social media platform?
No.
We would expect that company to act before more harm is done.
We would expect controls.
We would expect records.
We would expect safeguards.
We would expect accountability.
So why do AI companies get a different standard?
Dario’s essay calls for stronger government oversight of frontier models, especially around cybersecurity, biological weapons, loss of control, and automated AI research.
Those risks matter.
But the public is already facing another immediate threat: AI-generated images, videos, and voices being used to fabricate reality.
Deepfake sexual abuse.
Voice cloning scams.
Fake evidence.
Fake public statements.
Fake political media.
Fake corporate instructions.
Fake emergency calls.
Fake reputational attacks.
This is not theoretical.
This is happening now.
And the solution does not require waiting years for a new agency, a new law, or a perfect regulatory system.
AI companies already control the front door.
They control the tools.
They control the accounts.
They control the outputs.
They control the download/export systems.
They control the logs.
They control the restrictions.
They control whether a generated image, video, or voice can be released freely into the world.
That means they can act now.
The standard should be simple:
Create normally.
Watermark first.
KYC/AML before use.
Original stored.
Accountability attached.
For AI-generated images, videos, and voices, outputs should remain watermarked and restricted by default. If a user wants to download, post, export, publish, commercialize, or remove restrictions, then identity verification should be required.
The AI provider should retain the original generated asset and tie any unrestricted release or public use to the verified person or entity responsible.
That original record becomes the accountability anchor when harm occurs.
This same principle can apply to broader high-risk AI capabilities as well.
If an AI tool can help someone create cyber harm, biological harm, fraud, impersonation, or other serious public danger, then unrestricted use should require verified accountability.
Government regulation matters.
But company responsibility comes first.
If AI leaders know the risks are real, they should not hide behind slow government timelines.
They should act.
Now.
Source:
https://darioamodei.com/post/policy-on-the-ai-exponential
Full essay:
https://www.linkedin.com/pulse/ai-deepfakes-just-technology-problem-accountability-peter-s-c7yqe/
X post: