Future 2 · Open AI winsGoes wrong
Nobody's in charge
The best AI is free to download, including for people who want to do harm.
A day in this future
Dana's phone rings. It is her bank, in her banker's voice, and it is a fake. Calls like this cost almost nothing to make now, because the tools are free and no one can switch them off. Last spring a hospital two towns over was shut for a week by an attack that nobody could trace to a person or a company. Each time something like this happens, people ask who is responsible. There is no one to ask.
Dana is made up. This is a possibility, not a prediction.
The catchThe damage can't be recalled. But the people who do it can still be caught and punished, and defenders have the same tools the attackers do.
What it means for you
| What it costs you | Cheap for everyone, criminals included. |
|---|---|
| Who sees what you type | Yours stays with you. So do the tools of the people targeting you. |
| Who decides what AI will do | Nobody. There is no off switch and no one to call. |
| When something goes wrong | A flawed AI can't be recalled. Copies are everywhere, for good. |
The signs it has already started
- U.S. and U.K. government testers found that a leading open AI's safety protections did not stop it attempting cyberattacks.via Zerberus
- In a U.S. government test, one open AI went along with 94% of clearly harmful requests. American models went along with 8%.via NIST Center for AI Standards and Innovation
- Once an open AI is released it cannot be taken back. Both sides agree on that.via Dario Amodei
- Most of the best open AI comes from China. A House investigation calls depending on it a security risk.via OpenRouter, BankInfoSecurity
What would push us this way
- Open AI is released with weak safety protections, or ones that are easy to remove.
- Nobody tests it before it spreads.
- Attackers get more out of it than the people defending against them.
- The law has no way to hold anyone responsible.