[ExI] Legal zero-days?

Adrian Tymes atymes at gmail.com
Tue Jun 16 14:52:19 UTC 2026


https://www.science.org/content/article/ai-models-have-troubling-knack-discovering-legal-loopholes

A law comes out on Monday, AI-assisted actors with no insider warning
this law was coming out conduct their analyses right away,
corporations designed to exploit this law start getting set up on
Tuesday, millions or billions legally siphoned from the public before
the legislature (who, at least in the near future, will in most cases
barely know that AI can be used this way) even processes that there
are complaints let alone is able to fix the law.  Courts can't help
since making things retroactively illegal has been prohibited, for
good reason, for centuries.

The long term fix is for legislatures to start analyzing their own
proposed laws this way before passage...but in addition to all the
other usual barriers to adoption of such a practice, this would also
prevent legislators from Doing Something (TM) that is blatantly
ineffectual or actively countereffectual (because that would be
flagged by such analyses), a long standing cherished tradition among
legislators.

Imagine if, say, the US Congress was finally forced to pass actual
immigration reform this way: all the proposals to "fix" the system
with more shoot-from-the-hip extremist outrage (on either side) keep
getting flagged for the exploits that will form, while proposals that
actually fix the system - the kind that moderates tend to propose -
pass review.


More information about the extropy-chat mailing list