Jamie McClelland: AI Hacking the Planet

A colleague asked me if we should move all our money to our pillow cases after
reading the latest AI editorial from Thomas
Friedman.
The article reads like a press release from Anthropic, repeating the claim that
their latest AI model is so good at finding software vulnerabilities that it is
a danger to the world.

I think I now know what it’s like to be a doctor who is forced to watch Gray’s
Anatomy.

By now every journalist should be able to recognize the AI publicity playbook:

Step 1: Start with a wildly unsubstantiated claim about how dangerous your
product is:

AI will cause human extinction before we have a chance to colonize mars
(remember that one? Even Kim Stanley Robinson, author of perhaps the most
compelling science fiction on colonizing mars calls bull
shit
on it).

AI will eliminate all of our jobs (this one was extremely effective at
providing cover for software companies laying off staff but it has quickly
dawned on people that the companies that did this are living in chaos not
humming along happily with functional robots)

AI will discover massive software vulnerabilities allowing bad actors to “hack
pretty much every major software system in the world”. (Did Friedman pull that
directly from Anthropic’s press release or was that his contribution?)

Step 2: To help stave off human collapse, only release the new version to a
vetted group of software companies and developers, preferably ones with big
social media followings

Step 3: Wait for the limited release developers to spew unbridled
enthusiasm and shocking examples that seem to suggest this new AI produce is
truly unbelievable

Step 4: Watch stock prices and valuations soar

Step 5: Release to the world, and experience a steady stream of mockery as
people discover how wrong you are

Step 6: Start over

Even if Friedman missed the text book example of the playbook, I have to ask:
if you think bad actors compromising software resulting in massive loss of
private data, major outages and wasted resources needs to be reported on, then
where have you been for the last 10 years? This literally happens on a daily
basis due to the
fundamentally flawed way capitalism has been writing software even before the
invention of AI. A small part of me wonders – maybe AI writing software is not
so bad, because how could it be any worse than it is now?

Also, let’s keep in mind that AI’s super ability at finding vulnerable software
depends on having access to the software’s source code, which most companies
keep locked up tight. That means the owners of the software can use AI to find
vulnerabilities and fix them but bad actors can’t.

Oh, but wait, what if a company is so incompetent that they accidentally
release their proprietary software to the
Internet?

Surely that would allow AI bots to discover their vulnerabilities and destroy
the company right? I’m not sure if anyone has discovered world ending
vulnerabilities in Anthropic’s Claude code since it was accidentally released,
but it is fun to watch people mock
software that is clearly
written by AI (and spoiler alert, it seems way worse that software written
now).

Well… we probably should all be keeping our money in a pillow case anyway.