Stop Unicode now
Then we can replace ASCII with BSCII, which includes £.
We can also add back, þ, ð, and æ
ASCII smuggling isn't just an AI security risk
Fraudsters have found a new use for ASCII smuggling, typically used to hide malicious prompts intended for AI models, in an old-school attack method: email phishing. Microsoft uncovered a massive phishing campaign using invisible Unicode tag characters that peaked at more than 2.37 million messages in late February, remained …
-
-
Saturday 5th September 2026 09:50 GMT Bebu sa Ware
Re: BSCII
ASCII is/was a seven bit encoding; BSCII given the opportunism, expediency and general corner cutting in 'B' must be a two bit code. ;) Three characters plus an escape char (unnecessary for politicians as they rarely say anything that is worth two bits let alone needing one bit.)
A suboptimal (it would be wouldn't it) Huffman encoding.
If you were pining for thorn and eth an Iceland keyboard and locale, might serve. ;)
-
-
Sunday 6th September 2026 08:59 GMT Dan 55
Re: BSCII
I don't know, spending all this money reinventing something which already exists.
If you want to push the boat out, bit 7 is for reverse colours.
-
-
-
Friday 4th September 2026 22:52 GMT ChoHag
Oh if only someone had warned us about this in 2014 https://www.unicode.org/reports/tr36/tr36-15.html !
(ISTR reading about it much sooner than that but I am too lazy to look for the history of the Unicode TRs (see also 39) for this throw-away comment to a problem we've known about for well over a decade that the AI folk have decided to reanimate)
(Also it's not an ASCII anything; if we were only using ASCII this problem would not occur, not even with a NRCS)
-
Sunday 6th September 2026 17:06 GMT the spectacularly refined chap
(Also it's not an ASCII anything; if we were only using ASCII this problem would not occur, not even with a NRCS)
No ASCII had its own set of very similar flaws, perhaps not so much with GUIs but certainly with text consoles or anything that emulates them.
Consider for instance
via car^H^H^H^Hgra
-
-
Friday 4th September 2026 23:24 GMT Tron
Well...
You would have assumed that this sort of thing was being spotted from day 1 on even basic e-mail filters.
As for prompt injections, AI should never be able to do anything without user consent on a system. Otherwise it is just a hacking platform that should be outlawed. If you do not consent, you are fine. If you do, you are playing Russian roulette.
-
Friday 4th September 2026 23:49 GMT Pascal Monett
Re: Well...
The only type of consent that has ever been baked into anything Internet is auto-acceptance.
Anything else requires a coder to manage the situation, and coders still haven't come around to understanding that.
Most don't really have the choice anyway because manglement tells them that acceptance should be done by default (looking at you, Redmond, but you're hardly alone in this).
-
-
Saturday 5th September 2026 08:11 GMT Richard 12
Where's the news?
This has been standard spam practice for over a decade, possibly two decades.
The presence of any non-printing Unicode used to significantly raise the spam score. Heck, non-printing codepoints within a word used to raise the score to infinity and either bounce or immediately erase it.
-
Saturday 5th September 2026 11:28 GMT just4this
old is new
“to split financial-lure words in phishing emails in an apparent attempt to evade keyword matching”
It was a common trick to post things like….
Why don’t you fu[b][/b]ck off?
on forums to bypass keyword matching decades ago. This new technique doesn’t seem too dissimilar to me.
-
Wednesday 9th September 2026 17:26 GMT Anonymous Coward
Not already filtering out these?
Shouldn't any basic spam filter already catch this? Seems like you'd want to run the mail through the filter twice:
First time as-is
Then remove all nonprinting characters, formatting, etc. and replace looks-like-English-but-isn't letters (like the Cyrillic "а") with their English equivalents.
Then run the results through the spam filter again.
