Thread by @jsrailton

This morning blue check accounts accounts claimed large explosions at the Pentagon.

... then the White House.

Russian state media amplified the faked Pentagon image from their gold check account.

The images look AI generated, as folks identified. 1/

2/ The images, when you zoom in, are literally *all wrong* and it didn't take sharp eyed observers to notice obvious inconsistencies.

https://twitter.com/N_Waters89/status/1660651721075351556

3/ Exactly as we have all warned, one of the blue check accounts was designed to impersonate Bloomberg, the news agency.

https://twitter.com/AndyBCampbell/status/1660654574644887558

4/ Markets responded to the artificial swirl of #AI -generated fake images & explosion reports.

One report puts the S&P 500 Index $SPX drop at $500 billion before recovering. You can see the NASDAQ $COMP dropping before recovering, too.

Sound familiar?

thestreet.com/technology/s-p…

5/ Sidebar: if the DC explosion tweet story sounds familiar...

10 years ago the Syrian Electronic Army hacked the Associated Press twitter account & published a similar claim.

Causing a stock dip then, too.
ap.org/ap-in-the-news…

6/ Back to today: One of the key blue check accounts responsible for pushing today's "pentagon explosion" fake this morning has done it before.

Just last month.

Scrutiny of the image again shows hallmarks of generative #AI.

7/ The swirl of #AI fakes also got enough traction that the relevant emergency services felt the need to respond.

https://twitter.com/ArlingtonVaFD/status/1660653619954294786

8/ The state of our current ripe-for-manipulation information environment:

Main amplifiers of #AI pentagon explosion disinformation had blue & gold checks.

But the Pentagon's *own official account @PFPAOfficial had verification stripped by Musk.

It is currently unverified.

9/ Many lessons today:

☑️Entry cost to generate good-enough AI fakes has plummeted
☑️Twitter's blue check self own = fertile ground for disinfo.

Here's another

☑️#AI fakes are as bad today as they will ever be. They are only going to get better & harder to quickly spot.

10/ What scares me: people who did today's #AI explosion disinformation must've known it would not last.

Too many people could look out windows and tweet 'nope.'

But if they picked a more distant area, far from a capitol, debunking would have taken *time*.

Expect more of that.

11/ Today's #AI explosion disinfo had an obviously short half life, which raises interesting questions about intent & what could possibly be achieved in that timeframe

Cautionary note: plenty of speculation about Syrian Electronic Army's 2013 AP tweet turned out to be wrong.

12/ Another lesson:

Some blue check #OSINT-branded accounts repeated the Pentagon explosion without doing OSINT.

Folks, there are webcams pointed at the DC skyline.

And the buildings pictured don't exist.

Their tweets added credibility & further sped the lie.

13/ Exercise: ask yourself how disinformation like this plays out in the context of a real high stakes moment.

Like, say, a report of an attack on the president with realistic pictures / video /audio...a few hours before polls close on Election Day 2024.

We are unprepared.

13/ Generative AI basically brought the wind machine to Twitter's hurricane of disinformation.

Verification is only going to get harder.