Could this be solved by digital signatures on web content? (Or, a way to store those)

Yes, pretty easily. In general actively faking old data, and especially faking it to convince the public is not what should be our main concern. Any given archive site can try that once, with high risk of quickly being caught.

A content addressed network, yes.

How do you obtain the key to check the signature?

Keybase would have been a good solution, essentially a GUI for a pgp keyserver but with social proofs like "the owner of this account also controls the domain washingtonpost.com (as evidenced by publishing the fingerprint as a TXT record)". Too bad about the zoom acquisition, development stalled since then.

I also don't know exactly what bag of bytes you would be checking the signature against, the raw normalized text of an story? Maybe if news articles were distributed as PDFs this would be a solved problem, but news sites actually don't want their content to be portable, they want you to sign in to their domain to view anything.

I designed it back in 2017. Called it the “permissionless timestamping network”.

https://intercoin.org/technology.pdf

You’d have to store the actual content. The network would just store the hash.

It’s just a Merkle DAG. The innovation is all about forcing nodes to timestamp everything if they timestamp anything.

Blockchains are overkill for this. Blockchains are for when content that was timestamped changes. Not when it just accumulates.

Also, I wrote this back in 2017 as an aspirational roadmap: https://github.com/Qbix/architecture/wiki/Internet-2.0

If you know you might want to prove something's authenticity later, post the SHA256 somewhere now. E.G. Multiple social media sites. Large and trusted web archives. Cryptocurrency blockchains. The last one's a stronger proof, with lots of money making sure it stays immutable.

Or hash everything you produce/consume. Then hash the list of hashes, and post that.

Or alternatively, counter forgeries by capturing more data. For web, but all data in general. Sensor RAWs to prove an image isn't "AI". Browser and network stack RAM dumps to prove a website's authenticity. Etc. There's what, a couple dozen accelerometers, GPS's, LIDARs etc. on the latest iPhones?

I agree. Archives shutting down creates a huge risk for what I call "historical context attacks" in the future, similar to what you're describing.

It's about generating false historical context to support falsified digital artifacts.

I wrote an article about it for Fast Company in 2020:

https://www.fastcompany.com/90549441/how-to-prevent-deepfake...