Feedback-ID: ic129442d:Fastmail
MIME-Version: 1.0
Date: Mon, 24 Aug 2026 09:54:25 +0100
From: "Martin Thomson" <mt@lowentropy.net>
To: "ai-control@ietf.org" <ai-control@ietf.org>
Message-Id: <6255e60c-19e9-4287-bf1f-081deb08a071@app.fastmail.com>
In-Reply-To: <D99F6497-3F3F-4844-9AF8-C4421A6B9E21@cooperw.in>
References: <D99F6497-3F3F-4844-9AF8-C4421A6B9E21@cooperw.in>
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Message-ID-Hash: HYYUWKO337PXJVIEIW6FLKNLD4W23H4V
Precedence: list
Subject: =?utf-8?q?=5Bai-control=5D_Re=3A_Questions_about_4=2E2_Search?=
Archived-At: 
 <https://mailarchive.ietf.org/arch/msg/ai-control/dUVzVfY1vNZAdM0QAdX4CBPjEfo>

On Fri, Aug 21, 2026, at 11:27, Alissa Cooper wrote:
> "[...] Where existing controls restrict presentation of these=20
> items, such as limitations on snippet size, those apply before any=20
> changes.=E2=80=9D
[...]
> If I=E2=80=99m understanding this right, why does that last sentence b=
elong in=20
> this vocabulary document? Like, why is the vocabulary document decidin=
g=20
> a priori that every site owner who says =E2=80=9Csearch=3Dy=E2=80=9D w=
ants search=20
> applications to snip before translating, rather than translating befor=
e=20
> snipping?

I've lost context, but is seems like the idea was that the definition of=
 "excerpt length" is defined in the native form "at most 100 characters"=
 makes no sense as a restriction after text-to-speech.  The text was att=
empting to address that more generally.

However, with hindsight, it seems like we don't need to say this.  Anyon=
e defining some sort of restriction can (and should) consider the effect=
 their restriction might have on accessibility-related transformations. =
 Then it is *them* not us, who needs to consider this.  I've suggested t=
hat we strip this language in https://github.com/ietf-wg-aipref/drafts/p=
ull/227

> "Allowing this use is conditional on the outputs of any processing=20
> being exclusively used by the search application according to the othe=
r=20
> restrictions in this section. That includes the training of AI models=20
> using the assets and the use of those models provided that the=20
> resulting models and their outputs are used exclusively in ways that=20
> meet the above conditions.=E2=80=9D
>
> What does the =E2=80=9Cthat=E2=80=9D refer to as the first word of the=
 second sentence?=20

Ah, "that" was supposed to refer to the use that was allowed.  That coul=
d be clearer.

Overall, this could be clearer, but your understanding is correct:

> Is the point here that if two of the uses that the search application=20
> makes of the asset are (1) to train AI models using the asset or (2) t=
o=20
> use those trained models, those two uses are allowed when search=3Dy o=
nly=20
> if the presentation of the asset meets the two bullet point conditions=
?=20

I've tried to clarify further in https://github.com/ietf-wg-aipref/draft=
s/pull/228