►
From YouTube: IETF102-OAUTH-20180717-1550
Description
OAUTH meeting session at IETF102
2018/07/17 1550
https://datatracker.ietf.org/meeting/102/proceedings/
A
B
B
Better
yeah
so
welcome
to
a
the
first
session
of
faith.
We
have
a
session
that
is
shared
with
the
acne,
so
the
first
an
hour
and
a
half
will
be
off
yeah
and
and-
and
the
last
hour
of
this
would
be
an
acne.
So
we
have
90
minutes
here.
B
This
is
obviously
I.
Am
the
note
well
everything
that
you
say
or
contribute
here
is
still
a
under
the
note.
Well
so
make
sure
you're
familiar
with
this
a
we
have
a
diverse
product.
That
would
thank
you
for
that.
M
minutes,
taker
takers,
a
Mike
and
Justin
thank
you
and
blue
sheets
are
circulating
already
didn't
have
to
be
too
close.
It's
good
yeah,
a.
D
E
D
D
And
actually,
oh
yeah,
so
there's
an
action
item
for
us
to
go
through
the
comments
to
make
sure
that
we
sort
of
summarized
on
how
we
address
the
difference.
I
use
she
comments
so
that
that
would
obviously
help
occur
to
advance
the
document
faster
because
he
needs
to
also
produce
a
write-up
for
the
document
going
forward.
It's
not
just
a
working
group,
the
Shepherd
doing
the
write
up,
but
also
the
the
responsibility
to
do
the
write
up.
D
B
And
and
the
Durst
documents
will
probably
discuss
most
of
those,
so
I
think
incremental
authorization
will
be
discussed
today,
em
again
jot
jot
BCP
and
MPLS.
We
started
that
right
up,
Homewood
will
be
discussed
today.
Security
topics
talkin
it
will
be
discussed
on
Thursday
and
talking
binding
and
talking
exchange
will
be
discussed
later
today.
Okay,.
B
F
Okay,
great
yes,
sir.
Thank
you
so
I'm
here
today
to
talk
about
the
earth
to
turbo
incremental
auth,
which
was
recently
adopted
by
the
working
group.
So
thank
you
for
that
next
slide.
Let
me
just
recap:
while
we're
doing
this,
so
the
problem
statement
is
something
like
this
asking
for
the
kitchen
sink
of
scopes.
Upfront
can
also
be
a
bad
thing.
F
E
F
But
the
theory
is
that
that
is
something
else
challenging
to
do
if
an
app
needed
say
five
different
scopes
and
wanted
to
request
each
of
those
separately,
it
would
then
have
to
track
five
separate
authorization
call
five
separate
refresh
tokens,
so
the
advantage
of
incremental
auth
is
that
they
get
grouped
together
into
a
single
grant.
That
represents
everything
the
user
has
granted
so
far,
particularly
relevant
for
public
clients.
There
next
slide,
please
so.
F
F
F
So
when
basically,
the
app
does
three
requests.
The
first
one
is
completely
normal
for
native
apps
and
then
the
second
one.
The
first
two
steps
are
exactly
the
same
as
no
laws,
so
they'll
just
request,
whatever
scope
they
wanted.
But
if
you
look
at
this
second
to
last
line
at
the
bottom
there,
when
they're
exchanging
the
authorization
code
for
that
second
grant,
they
include
you
know,
field
called
existing
grant
the
Refresh
token
from
the
first
request,
and
that
will
result
in
the
access
token
and
refresh
tokens
being
returned
on
the
bottom
line.
F
F
F
I
really
appreciate
that,
but
yeah
I
really
wanted
to
send
out
a
call
to
action
here
if
anyone
is
interested
in
this
spec
and
if
you
have
a
client
or
a
server
that
that
can
potentially
use
this
very
interested
to
to
get
in
touch
with
you
and
to
work
on
some
interrupts
there,
no
I'm.
So
that's
probably
going
to
be
one
of
my
main
focuses
going
forward
between
now
and
the
next
idea
next
slide.
F
Please,
and
so
with
that
point
in
mind,
I'm
happy
to
announce
that
the
apple
of
iOS
library
does
have
support
for
this
standard.
Now,
particularly
the
public
client.
One
I
realize
been
at
Norma
show
code,
but
I
just
wanted
to
show
this,
because
it's
actually
just
just
to
emphasize
how
simple
ease.
So
this
is
like
literally
the
sum
total
amount
of
code
that
a
client
would
need
to
provide
in
order
to
increment
an
existing
grant.
They
had
with
the
calendar
scope.
F
That
is
literally
all
they
have
to
do,
and
this
is
kind
of
like
why
we're
doing
why
I
really
wanted
to
push
this,
and
that
is
if
we
make
this
really
easy
for
people
to
do
then
I'm
hoping
more
people
will
do
it
and
I
think
it's
the
right
thing
for
users,
and
so
if
more
people
do
the
right
thing
the
users,
then
then
we
all
win
next
slide.
Please
and
I
published
a
demo
using
this,
this
new
API,
so
the
API
by
the
way
it
has
actually
emerged.
Yet
it's
in
a
pull
request.
F
But
if
you,
if
you
are
an
iOS
developer
and
you
click,
this
link
it'll
have
everything
a
demo
showing
how
this
works
using
those
new
API.
So
you
can
try
that
out
and,
as
I
said
before,
I'm
really
interested
to
gets
in
the
airdrop
happening,
so
I'd
love
for
if
anyone's
implementing
is
server-side.
You'd
be
great
to
point
this
demo
at
your
server
too
and
see
if
it
works
and
see
what
problems
we
have,
if
any
all
right
last
slide.
F
So
that
was
my
update
very
very
interested
to
hear
what
people
think
of
the
if
any
Wells
has
had
a
chance
to
review.
C
F
F
F
So
the
reason
the
reason
I
originally
had
it
should
there
is
I'm
trying
to
kind
of
balance
balance
two
competing
things
here.
The
let
me
just
read
the
section
that
says:
if,
if
the
verification
succeeds,
the
new
refresh
token
issued
in
the
access
token
response
of
the
token
endpoint
should
include
authorization
for
the
Scopes
in
the
previous
grant.
So
it's
basically
saying
when
the
when
the
user,
when
the
client
provides
the
existing
ground
parameter,
any
new
tokens
issue
should
contain
the
Scopes
from
that
parameter.
F
The
only
reason
I
really
made
that
it
should
not
a
must,
is
I
kind
of
always
feel
like
it's
up
to
the
authorization
so,
but
what
what
they
want
to
do,
but
it's
it
is
probably
a
little
bit
confusing
to
say
that
so
should
there,
because,
if
you're
sort
of
not
doing
that
in
you
know
in
a
normal
case,
then
you're
not
really
implementing
this
spec
at
all.
So
my
plan
there
is
to
actually
change
it
to
a
must
but
kind
of
qualify
that
with
some
words
to
say
you
know.
H
F
Good
question
and
I
think
I
mean
that's
coming
another
trend
that
we're
going
to
see
actually
so
you're,
very
topical
and
and
for
the
most
part,
I
think
that
can
already
be
implemented
in
the
bounds
of
Hoth.
So
I
definitely
wouldn't
want
to
have
any
texture
that
would
tie
anyone's
hand
around
selected
consent.
So
I
guess
what
you're
saying
is.
F
Potentially
the
existing
grants
could
be
shown
on
that
incremental
auth
screen,
that's
not
something
that
at
least
Google's
in
orientation
does
today,
but
they
could
be,
and
the
user
could
actually
revoke
something
as
they
reviewing
and
maybe
granting
or
maybe
not
granting
the
next
one.
So
I
totally
agree
to
your
comment.
I'm
it's
gonna
be
hard
to
come
up
with
some
good
words
ran
out
uh-huh,
but
I
think
we
can.
We
can
probably
work
at
such
then
that
it
accommodates
then
yeah.
H
F
I
guess
the
wrinkle
is,
if
you,
if
you
interpret
a
street,
if
you
actually
kind
of
never
do
it,
then
then
you
have
any
whether
the
spec
I
suppose,
maybe
that
might
be
worth
noting.
But
but
yes,
it's
more
like
in
a
normal
case,
you
would
consider
that
existing
grant
and
likely
bundle
that
in
but
I
guess
either
way.
If
we
keep
it
as
a
should.
Maybe
I
can
just
clarify
that
you
actually
never
do
that
then
you're,
probably
not
going
to
spec
at
all,
does
that
dick.
G
Great
now,
maybe
framing
it
more
on
that
the
response
shooting
must
include
all
the
authorization
that
the
user
is
granted
and
so
by
putting
the
user
into
the
equation,
then
you're
tying
in
to
what
the
user
is
granted
as
opposed
to
whatever
the
server
decided
as
being
previous
grants
right.
So
that
way,
if
any
came
in
and
out,
it
makes
it
clear.
I
like.
I
F
That
is
a
very
good
point
from
twisting.
This
is
something
this
spec
does
not
today
that
they
should
so
I
think
the
OAuth
spec
actually
already
documents
this
behaviour.
So
if
you
are
returning
an
authorization,
we're
sort
of
different
scope
than
what
was
requested,
you're,
actually
a
meant
to
populate
that
field,
but
these
behaviors
not
mentioned
is
highly
relevant
incremental.
Also
I'll.
Don't
me
take
those
a
note
to
add
something
as
well.
F
Alright,
so
the
second
point
here
is
the
respect
goes
into
a
little
bit
of
detail
kind
of
some
best
practices
around
how
to
handle
denials
in
the
case
of
incremental
authorization.
You
know
one
of
the
things
is
kind
of
not
throwing
the
baby
out
with
the
bathwater.
If
you
use
a
where
to
choose
no
for
a
particular
authorization
scope,
then-
and
you
know
you
shouldn't
like
completely
abandon
all
the
grants
that
you
have
so
far-
the
app
should
actually
continue
operating.
I
think
this
is
very
important
from
a
user
standpoint.
F
You
don't
want
to
kind
of
add
to
the
user
or
like
really
anticipate
that
they're
granting
scopes.
They
don't
want
to
that's
the
whole
reason
why
we
haven't
word
creative
spec,
but
Austin
raises
the
point
about
how
should
the
server
handle
it
so
I'll
definitely
add
a
section
in
that.
I
totally
agree
with
the
assumption
stated
here,
which
is
that
also
the
authorization,
so
that
would
not
normally
discard
any
previously
rendered
scopes.
If
the
user
didn't
explicitly
indicate
that.
G
Take
heart
again,
this
brings
up
sort
of
another
question:
I,
don't
know
whether
the
working
group
has
looked
into
this
anytime
in
the
past,
but
standardizing
token
discovery,
endpoint
err,
a
client
could
call
a
known
endpoint
and
get
back
all
the
Scopes
that
were
in
the
token.
So
then
they
could
learn
what
do
they
have
now
and
then
they
would
know
what
do
they
need
to
ask?
For
you
know
a
number
of
the
providers
have
a
proprietary
API
for
that,
but
standardizing
that
could
simplify
work
for
developers
well,.
H
Annabel
Bachman
Amazon
a
similar
point
to
what
I
said
earlier.
Let's
I
think
we
want
to
avoid
assuming
that
an
authorization
server
isn't
going
to
revoke
previously
issued
consent.
If
that's
what
you
know,
they
may
want
to
allow
the
the
user
to
do
that
through
that
UI,
if
you're
a
user
seeing
a
bunch
of
stuff
that
you've,
you
know
whether
you
know
you've
granted
it
or
not.
Whether
you
know
this
is
new
stuff
that
you're
granting
versus
old
stuff.
If
you
hit
cancel.
What
does
that
mean?
You
know?
Maybe
it
means
you
cancel
everything.
F
That's
a
good
point
well
for
one
I.
Think
I'll
make
sure,
like
this
pig
doesn't
tie
anybody's
hand
would
relate
to
that
I.
Think
he's!
You
know
everything
is
up
to
the
authorization
server
I
mean
I,
am
interested
to
capture
usability,
best
practices
and
then
kind
of
give
advice
to
people
in
mulaney's.
So
that
is
an
interesting
question.
It's
only
we
have
a
yes
or
a
no,
and
this
is
actually
like
a
yes/no
and
there
might
be
some
other
actions
in
the
users.
So
do
you
think
that's
worth
documenting.
H
H
At
the
very
least,
in
like
a
consideration
section,
perhaps
right
on
another
thought
is
basic
question
is:
do
we
actually
have
an
access
denied
use
case
where
the
customer
or
the
user
is
not
revoking?
Everything
like
if
I'm,
if
I'm
a
a
user
and
I
reject
the
new
scopes,
but
the
authorization
server
still
says
the
old
scopes
are
valid.
Does
it
make
sense
that,
for
that
to
be
an
access
denied,
or
should
that
be
a
success,
and
just
issuing
you
back
the
Scopes
you
had
previously
because
technically
you're
still
authorized
just.
H
So
I'm
saying,
if
we're
going
to
support
that
kind
of
a
use
case
where
there
could
be
a
revocation
of
everything,
then
maybe
it
makes
sense
for
us
to
reserve
access
denied
for
the
case
where,
where
everything's
getting
revoked
and
in
all
other
cases,
you're
authorized.
But
you
need
to
check
what
you
got
authorized
for.
H
F
F
Imagine
well
well
actually,
the
way
this
is
specified.
You
would
only
present
the
the
new
consents,
so
it
knows
what's
being
asked
in
that
incremental
authorization
request.
So
it
knows
what
to
present
in
that
context,
to
you
to
give
an
example.
What
I'm
talking
about
like
imagine,
you
have
a
cloud
product
with
various
different
scopes
and
you
have
like
two
laptops
and
two
command
lines,
both
interacting
with
some
cloud
tool.
F
H
F
F
H
J
There
are
a
lot
of
other
systems
that
you
know,
there's
Emma
and
Seba,
and
even
the
device
flow
have
all
these
temporary
tickets
to
kind
of
tie
different
parts
of
the
process
together
here.
So
we
have
approached
this
from
a
lot
of
different
directions,
but
fundamentally
there
is
no
object.
That
is
the
authorization
grant
that
we
agreed
on
that
like
this
is
what
it
is
in
these
what
the
parts
are,
which
is
why
all
of
these
assumptions
about
like
how
is
this
stored?
How
is
this
referenced?
J
That's
something
consider
what
I
actually
got
out
to
talk
about,
though,
was
that
the
introspection
specification
we
intentionally
wrote
that
to
be
slightly
vague
as
to
who's
supposed
to
be
using
it.
The
primary
use
case
is
very
definitely
the
resource
server
talking
to
the
authorization
server.
That
is
that's
the
job
that
it
was
meant
to
solve.
However,
if
you
look
at
how
the
spec
is
written
in
kind
of
what
it
allows
in
different
places
and
stuff
like
that,
the
client
can
use
it
without
really
going
off-book.
J
The
trick,
of
course,
is
to
have
your
authorization
server
implemented
in
such
a
way
that
it
can
safely
allow
clients
to
introspect,
which
is
not
a
which
is
not
as
simple
as
it
seems
on
the
surface,
having
implemented
that
in
several
different
systems.
There
are
a
lot
of
gotchas
about
allowing
clients
to
ask
what
a
token
is
good
for
so
I
think
it
that's.
J
J
Think
that,
if
it's
that,
if
we're
going
to
recommend
the
application
of
introspection
to
help
solve
this,
this
whole
notion
of
I'm
gonna
go
look
up
what
my
tokens
good
for
and
so
that
I
only
ask
for
the
Delta
or
I
go
look
up
what
the
token
is
for
to
figure
out.
If
I
got
what
I
asked
for
or
something
like
that,
then
we
should
document
that
here
the
sort
of
the
AAS
implementation
and
security
considerations
for
token
introspection,
as
used
by
clients
for
this
I,
think
that
that
would
be.
That
would
be
good
here.
J
F
J
D
I
was
just
wondering
whether
it's
really
a
good
idea
to
do
token
introspection
from
the
client
when
you're
only
interested
in
this
course,
because
it's
essentially
you
get
the
token
you
sent
the
token
back.
There's
one
additional
round-trip
lots
of
data
being
passed
around
you.
You
may
get
actually
realistically
more
back
than
just
the
scope.
You
actually
get
a
lot
of
data,
so,
like
seems
extremely
wasteful,
like
in
terms
of
what
the
round-trip
and
the
what
goes
over
the
wire
wouldn't
be
more
sensible
to
actually
pass
a
parameter
along.
That
comes
back.
D
J
You
know
I'm
just
I
didn't
say:
introspecting
from
a
client
was
a
good
idea.
I
just
said
it
was
allowed
by
the
spec,
that's
true
or
not
strictly
disallowed
to
be
pedantic,
so
yeah
there
are
a
lot
of
downsides
and
one
of
them
is
that
the
data
model
that
you
get
back
from
introspection
is
intended
for
resource
servers.
It's
got
a
lot
of
extra
information
that
clients
generally
shouldn't
know
anything
about.
J
Also
going
back
to
the
point
about
clients
being
stupid,
generally
they're
gonna
use
the
token
and
see
if
it
works,
I
mean
most
low
off
clients
that
I
have
seen
in
the
wild.
Ignore
the
scope
field.
They
get
a
token.
It's
the
200,
the
access
token
field.
Is
there
they're
happy?
They
don't
check
expires
in
they
don't
check
scope.
You
know
they
don't
do
anything
else.
If
I
get
the
access,
token
I'm
call
the
API
hey
the
API
failed.
J
J
It
doesn't
make
sense
to
do
that
extra
round-trip,
because
they're
doing
a
round-trip
to
see
if
the
token
is
any
good
by
using
it-
and
they
already
have
to
have
appropriate
recovery
for
that
failing
anyway,
for
the
token
being
revoked
or
not
being
strong
enough
or
having
expired
or
any
any
number
of
other
things
that
they
already
have
right.
The
tendency
is
to
reuse
those
code,
those
recovery
code
paths,
as
opposed
to
trying
to
do
something
more
smart
and
proactive.
J
That
said,
there
are
a
handful
of
people
who,
especially
with
kind
of
back-end
system,
clients
that
you
know,
can
afford
to
be
a
little
bit
smarter
about
things
or
it's
worth
putting
in
a
little
bit
extra
smarts.
Then
they
do
tend
to
do
these.
These
pre-emptive
checks
for
expiration
and
stuff,
like
that,
because
the
failure
case
is
more
expensive
than
the
check
in
those
it's
a
trade-off,
but
I
actually
the
vast
majority
of
cases.
It
doesn't
make
sense.
Yeah.
F
J
F
F
G
How
many
people
were
in
Singapore
saw
the
presentation
a
few
of
you
so
for
those
of
you
that,
were
there,
here's
updater
and
what
happened:
change,
name
from
mutual
ooofff,
because
brian
didn't
like
mutual
and
to
reciprocal,
which
I
think
actually
fits
better
added,
a
reciprocal
parameter.
That
could
happen
in
the
first
flow
so
that
the.
G
G
You
need
to
log
in
like
numerous
times
to
go
through
to
do
it,
because
you
you,
you
weren't
at
the
right
site,
and
so
you
have
to
get
bounced
over
to
one
site
to
go
and
start
something
and
all
over
the
place.
So
I
go
through
that
a
bit
here
and
the
flows,
you
know
the
user
is
over
in
the
Lexx.
App
like
so
knows
indicates
the
Sonos
authorizes
like
satsang.
G
Always
user
gets
redirected
to
Alexa
with
asanas
authorization
code,
Alexa
exchanges
that
code
for
an
access
token,
which
is
your
standard
flow,
and
then
Alexis
will
send
you
back
over
to
Sona
to
start
the
second
flow
and
then
Sona
always
sends
you
back
over,
which
then
of
course
drives
a
login
with
Amazon,
because
we
started
in
the
app,
but
now
we're
in
the
web
browser.
So
now
the
user
has
to
log
in
Amazon,
even
though
they
were
already
logged
into
the
Alexa
app
and
then
the
authorized
Sonos
and
Amazon.
G
Then
the
users
redirected
back
to
Sonos
with
the
Lex
authorization
code
so
known
as
exchanges
that
code
for
it,
and
then
users
rejected
back
over
to
Alexa
completion,
which
I
think
the
first
version
was
even
this
clean.
It
was
even
uglier
than
this,
so
the
reciprocal
overflow
I
made
the
font
a
little
smaller.
So
it
also
don't
have
screen.
G
Bit
of
a
cheat
there,
so
one
through
five
is
all
the
same
step.
Six
seven
and
eight
are
removed,
and
so
what
the
user
is
now
back
over
at
the
Alexa
app
at
the
end,
a
new
step
is
the
user
authorizes?
What
Alexa
is
going
to
grant
so
nose
and
then
Alexa,
just
back
channel
called
so
nose.
Token
end
point
with
their
authorization
code
and
provides
a
context
for
Sonos,
bypassing
so
knows
their
access
token
and
then
standard
flow
in
step,
11,
which
is
Sonos
exchanges,
the
authorization
code
for
an
access
token.
G
Thank
you,
I.
Think
Brian.
You
gave
me
the
suggestion
that
we
should
use
you
have
this
kind
of
grant
type
and
then
the
party
B
access,
token
and
prayer
persons
comments.
All
this
is
is
really
for
the
party
B
to
understand
the
context
and
then,
of
course,
the
authorization
code
from
party
a
so
that's
an
example
of
what
that
looks
like.
G
I
haven't
really
come
an
example
where
it's
asymmetrical,
where
really
the
user
can
start
at
either
side
and
either
party
could
update
the
other's
authorization
implementations
as
I
mentioned,
and
I
got
some
typos
that
people
have
pointed
out
in
the
spec,
but
sort
of
opening
up
for
discussion.
People
have
anyone
else
think
this
is
be
useful.
Besides
the
other
voice
assistance,
the
question
is.
F
G
F
G
F
G
C
G
G
H
G
H
L
H
K
H
H
G
H
G
C
G
G
G
G
G
D
I,
wonder,
would
you
actually
that
may
be
a
good
idea
to
include
that
parameter,
because
that
stage
you
could
actually
then
fall
back
to
the
other
mode,
rather
than
failing
later
on,
when
the
other,
when
you
have
the
authorizations
over
not
knowing
about
the
functionality
of
this
extension,
or
did
you
consider
that
case?
What
happens
if
you,
you
know
I
key
go
through,
then
it
doesn't
support
that.
Would
it
completely
fail
or
could
it
fall
back
to
the
longer
flow?
Instead,
you
know
what
I'm
talking
about
no.
G
G
G
G
We
thought
that
Sona
supported
it,
but
they
actually
didn't
deploy
it
right
when
we
would
make
a
call.
G
G
H
To
add
to
that,
if
you
think
about
how
this
would
fail,
you
know
if
Sonos
doesn't
support
it,
then
they're,
just
not
gonna,
ever
send
any
tokens
back
or
any
any
reciprocal
scope
back
to
us
and
if
we
try
to
give
them
something
anyway,
their
token
end
point
is
just
going
to
reject
it.
If
so,
no
supports
it,
but
Alexa
doesn't
then
Alexis
just
gonna
ignore
any
reciprocal
thing
that
gets
sent
back.
Sonos
is
never
going
to
get
a
call
to
their
token
and
point
to
give
them
tokens
at
that
point.
H
D
I
G
A
K
C
I
A
B
N
O
Close
enough,
okay,
I'm
gonna
Jam
three
different
documents
into
one
update
here,
a
lot
of
it.
It's
it's
just
sort
of
some
status
updates
and
overview,
but
a
couple
of
these
mutual
TLS
and
token
binding,
maybe
have
some
relevance
to
the
pop
token
stuff.
That's
going
to
be
discussed
on
Thursday
and
so
yeah
welcome
to
not
San
Francisco.
I
O
O
I
noticed
one
of
the
authors
of
that
stuff
is
in
here
so
I'm
sure
he
will
make
sure
I'm
right
on
point,
but
there
are
three
core
drafts
that
make
up
that
set
of
documents
right
now
coming
out
of
the
token
binding
working
group.
The
last
meeting
this
week
on
Friday.
The
very
very
last
meeting
is
for
that
working
group
for
anybody's
interested.
O
O
So,
across
the
million
cases
where
you
want
to
bind
to
a
key,
that's
used
in
a
different
domain,
there's
this
thing
called
a
referred
to
combining
versus,
what's
normally
called
the
preferred
excuse
me
provided
to
combining
or
the
provided
is
between
the
client
and
the
sort
of
the
main
server.
The
referred
enables
the
client
to
communicate
a
key
that
would
use
with
a
different
server
so
that
that
original
server
combined
tokens
2.4
its
usage
a
little
bit
more
graphical
showing
of
how
this
works
in
the
TLS
handshake.
There's
just
hello.
O
Extension
number
24
if
you're
interested
that
the
client
includes
in
its
client
hello,
along
with
basically
what
signature
algorithms
it
supports
to
do,
tow
combining
there's,
there's
three
defined
right
now
and
then,
if
the
server
is
happy
with
this
and
wants
to
do
tow
combining
with
this
client,
it
responds
including
the
extension
and
this
its
server
hello,
same
extensions,
same
structure,
but
it
only
has
one
of
these
key
parameter
types
was
the
key
signature
type,
basically
indicating
the
client
hey.
This
is
the
one
that
we're
going
to
use
in
this
case.
O
In
this
example,
the
client
offers
up
a
couple,
different
choices
and
the
server
responds
with
okay.
Let's
do
two
combining
and
please
use
ECDSA
with
the
p2p
civ
curve
with
me,
assuming
negotiation
isn't
successful
in
the
TLS
handshake,
then
on
every
subsequent
HTTP
request.
The
client
includes
this
header,
which
is
the
sec
token
binding
header,
with
an
encoded
token,
by
any
message
in
that
message,
that
can
be
more
token
bindings.
Each
individual
took
combining
indicates
the
type
which
was
either
to
provided
or
referred.
O
The
key
type
and
the
exported
key
material
which
is
defined
in
RFC
I,
can
remember
the
number,
but
you
can
think
of
it
as
basically
a
like
a
cryptographically,
secure
identifier
contextualized
for
this
usage
for
the
individual
TLS
session,
and
then
there's
some
room
for
extensions.
What
this
does
is
when
verified
properly.
O
This
proves
that
the
client
possesses
the
private
key
with
respect
to
that
TLS
connection
corresponding
to
the
public
key
that
was
set
in
that
message,
and
these
keys
are
long-lived
and
span
TLS
connections,
so,
while
TLS
connection
might
be
torn
down,
expired
or
whatever
the
token.
The
actual
binding
to
this
key
can
persist
beyond
that,
because
the
the
signature
is
is
over
the
ekm
for
the
TLS
session,
but
the
key
itself
is
longer-lived.
O
O
This
idea
of
cross
domain
or
federated
token
binding
is
also
supported
in
the
web
browser
use
case.
There's
an
HTTP
response,
header
that
the
the
relying
party
or
client
in
the
Roth
of
an
ID
world
includes
on
its
redirect
over
to
the
IDP
or
the
authorization
server,
and
what
this
does
is
it's
just
called
included
referred
token.
O
It
would
use
as
well
as
the
token
binding
that
it
would
use
between
itself
and
the
reliant
party,
and
this
allows
for
that
identity
provider
to
bind
tokens
that
to
the
key
that
the
browser
uses
with
the
rely
on
party
basically
binding
for
the
when
it
comes
back
for
the
initial
connection
there
and
that's
the
provided
and
referred
as
the
the
name
and
the
distinction
between
the
two
again
provided
as
the
one
with
that's
used
normally
with
their
server
you're
talking
to
and
referred,
is
from
a
referred
site
in
this
case.
O
It's
the
same
same
header
and
just
inside
of
that
encoded
message.
It
actually
includes
two
token
bindings.
Instead
of
one
and
there's
guidance
in
the
document
itself.
That
basically
suggests
that
generic
token
binding
implementations
should
be
able
to
send
referred,
based
on
some
other
signal
to
the
API
itself
or
from
the
application,
and,
ultimately
that's
what
a
lot
of
oh
F
is
going
to
rely
on.
Is
the
hopes
that
implementations
do
in
fact
expose
that
in
a
reasonably
useable
way?
O
So
with
that
in
mind-
and
it
was
quick,
but
it
hopefully
gives
a
little
bit
of
overview.
The
oauth2
token
binding
standard,
that's
being
worked
on
within
this
working
group,
basically
provides
an
oauth2
proof
of
possession
mechanism
based
on
token
binding
that
defeats
replay
or
play
of
lost
or
stolen
security.
Tokens
of
all
sorts,
probably
the
most
interesting
and
the
most
applicable
use
case
is
binding
access
tokens.
You
bind
those
against
the
referred
to
combining
ID,
similarly
sort
of
supporting
that
the
federated
use
case.
O
So
the
client
makes
request
includes
both
the
normal
token
binding
idea
that
it
would
use
with
the
a
s
as
well
as
the
token
binding,
ID
or
key
that
it
would
use
with
the
resource
server.
The
authorization
server
then
binds
the
access
tokens
to
that
key.
That
would
normally
use
with
the
resource
server
and
thus
provides
this
proof
of
possession
mechanism
around
the
access.
Token
binding
it
to
the
the
key
use
with
the
resource
server
and
this
can
be
represented
and
in
the
jwq
access
token,
as
well
as
the
introspection
response
with
the
CNF.
O
The
confirmation
claim
to
find
anything
in
RFC,
7800
and
inside
of
that
a
new
confirmation
value,
tbh
forto,
combining
hash
member.
So
it's
a
sha-256
hash
of
the
token
binding
ID
included
inside
either
the
introspection
response
or
the
the
JWT.
That
basically,
is
a
hash
of
the
tow
combining
ID,
which
is
what
allows
the
resource
server
to
make
that
comparison
and
check
whether
in
fact
the
the
issued
token
is
bound
to
the
key,
which
is
in
turn
proof
possession
of
by
the
client
there's.
O
Also
some
text
in
there
about
binding,
refresh
Durkins
just
to
the
provided
token
binding
ID
there's
some
support
for
binding
authorization,
codes
to
variant
flows,
around
native
app
clients
and
web
server.
Clients
are
a
little
esoteric.
I
won't
go
into
the
details
and
then
speaking
of
esoteric,
there's
also
some
language
about
binding
JDBC,
based
authorization,
grants
and
jibiji
client
authentication.
O
That's
it
in
a
nutshell:
won't
go
name
deeper
details,
but
I
will
cover
a
few
happenings
since
London
draft
7
was
published
and
really
the
only
main
difference
here
is.
It
was
clarified
that
the
encoding
of
that
tbh
value
is
basically
for
URL
encoded,
but
that's
a
somewhat
ambiguous.
In
some
cases
it
was
clarified
that
that
does
not
include
any
of
the
trailing
pad
characters
or
white
space
or
anything
else.
O
This
actually
was
a
result
of
a
review
on
the
MTO
stuff
from
Justin
that
was
equally
applicable
here,
so
I,
just
added
the
clarification
and
I
updated
and
affixed
a
few
references
that
were
either
referencing
wrong
parts
and
other
documents
or
sort
of
the
the
HTML
tooling,
where
it
does
self
references
when
it
shouldn't
erase
to
those
as
well
as
update,
you
know,
related
not
directly,
but
somewhat
related.
The
open,
ID
connect,
token
bound
authentication
draft.
A
new
draft
of
that
was
published,
and
it's
worth
mentioning
here,
just
because
the
tbh.
O
The
three
underline
sort
of
quarto
combining
specifications
are
I've
said
it
now,
I
think
two
or
three,
or
maybe
four
ifs
in
a
row
that
they're
almost
done.
But
in
this
case
they
really
are
almost
done.
The
token
binding
over
HTTP
is
in
the
RFC
editors
cube,
which
is
really
good
and
then
not
far
behind
it.
O
The
TLS
extension
for
negotiation
and
the
token
binding
protocol
itself
are,
in
this
approved
announcement
to
be
sent
so
they're
just
waiting
for
a
tea,
follow
up
there
and
I
think
they're
close
ish,
so
it
shouldn't
be
much
longer
and
that's
where
we're
at
with
toe
combining
4'o
next
one.
The
sorry.
N
Call
my
DNA
I'm
thinking
about
to
AP's
reside
on
the
same
device,
whether
it
they
can
share
a
TLS
connection,
and
my
understanding
is
no
because
the
token
binding
negotiation
is
at
the
TLS
level.
So
if
one
app
already
has
a
TLS
connection
to
the
same
authorization
server
or
a
resource
server,
then
another
app
should
initiate
another
TLS
connection
so
that
it
can't
have
a
different
configuration
for
the
talk
combining
whether
the
application
be
wants
to
use
the
talk
combining
or
not.
K
John
bradley
yubico
and
one
of
the
chairs
of
the
token
binding
work
group,
so
the
definitive
answer
is
it
depends
so
alexei
who's.
Eyeballing
me
would
probably
have
the
better
answer
for
Windows,
which
probably
has
the
most
advanced
implementation,
but
it's
going
to
if,
if
you've
implemented
token
binding
by
including
CRO
net
into
your
application
and
every
application
has
its
own
key
store,
then
yes
they're
all
going
to
be
different.
If
you're
running
say
on
Windows
that
actually
provides
an
API.
K
Q
It's
under
a
but
yes
so
under
a
book
called
Microsoft,
so
each
application
will
have
its
own
till
that
cos
connection
and
unassuming,
but
the
you
know
the
token
binding
keys
can
be
shared
and
that's
platform
specific
right.
So,
like
the
extent
to
which
token
binding
keys
can
be
used
by
multiple
applications
is
platform.
Specifics
on
Windows
we
have
a
concept
of
app
containers
and
if
the
applications
are
in
the
same
app
container
the
same
integrity
layer
level,
then
they
can
share
tow
combining
keys.
N
Calm
eyed
again
and
the
first
motivation
of
the
pixie
is
to
distinguish
two
applications
and
deciding
the
same
device
to
not
confuse
each
other,
so
armed
fair.
How
the
context
of
the
token
binding
ID
we've
got
sharing
materials,
so
they
there
should
be
an
API
between
the
application
and
the
operating
system,
so
that
I
am
who
and
who,
on
my
tour,
combining
keys
fancy,
especially
on
the
first
authorization
case
and
later
token
binding
token
bound
token
to
the
resource.
Er.
K
Right
so
again,
that's
application
specific,
so
work,
that's
that's!
Probably
the
part
of
the
HTTP
spec
that
it
took
us
the
most
amount
of
time
to
negotiate
with
the
iesg
is
trying
to
explain
that
you
know
different
vendors
are
going
to
have
different
api's
and
slightly
different
takes
on
it.
So
on
Windows,
apparently
that
you
have
some
concept
of
an
application
group,
so
they
can
only
share
if
they're
part
of
the
same
ie
signed
by
the
same
developer,
key
or
some
such
thing.
K
Perhaps
Android
will
have
a
similar
mechanism,
but
since
they
haven't
and
I
said
how
they're
going
how
they're
going
to
expose
it
on
Android
I
can't
we
can't
answer
those
questions
until
the
the
various
operating
systems
actually
implement
the
the
api's.
But
there
should
not
talk
in
binding,
IDs
or
hucked,
and
combining
keys
ought
not
to
be
shared
between
completely
unrelated
applications
in
a
single
browser.
K
A
N
H
Annabel
Backman
Amazon,
so
regarding
token
binding
of
access
tokens
received
from
the
token
endpoint,
it
wasn't
clear
to
me:
is
there
a
limit?
Is
it
only
one
referred
token
binding
that
you
can
include,
or
could
you
request
an
access
token
that
is
bound
to
n
different
referred
token
token
bindings
referred
token
IDs,
whatever
the
right
term
is,
the
terms
are
a
little
hard.
O
H
The
the
VIP
or
the
load
balancer
somewhere
away
from
the
actual
systems
that
are
trying
to
do
stuff
with
it
that
actually
know
they
need
to
share.
There's
some
I
I.
Don't
really
think
it's
feasible
to
just
assume
that
that's
going
to
be
able
to
be
shared,
so
I
think
we
should
think
about
how
we
can
get
from
a
refresh
token
to
access
tokens
bound
to
indifferent
token
binding
IDs.
O
It
you
know
one
of
the
reasons
to
binding
it
to
a
key
is
so
that
you
have
to
have
that
key
in
order
to
use
it
and
and
I
recognize
that
access,
widespread
access
and
distribute
systems
to
private
keys
can
be
problematic.
But
what
I
don't
think
we're
in
a
position
where
we
want
to
undermine
the
value
of
the
protocol
to
support
that
kind
of
case?
Let.
H
Me,
let
me
rephrase
my
my
suggestion:
it
doesn't
necessarily
have
to
be
a
single
access,
token
bound
to
multiple
referred
token
binding
IDs
as
long
as
there's
there's
a
mechanism
for
a
service
or
for
for
the
client
to
get
an
access
token.
That
is
bound
for
a
token
binding
ID
from
some
other
service,
separate
from
the
one
at
request
now
seriously
it.
H
There
are
some
implications
of
that.
You
know
as
far
as
those
tokens
have
to
continue
to
both
be
valid.
You
know
this
is
the
subsequent
refresh
for
a
new
token
shouldn't
invalidate
the
previous
one.
There's
implementation
details
that
fall
out
of
that,
but
again
aren't
normative,
but
are
probably
worth
being
discussed
as
best
practices
are
not
normative.
Texts.
K
So
we
have
identified
that
in
some
clustered
environments,
where
the
the
coinage
is
being
distributed
across
the
cluster,
it
may
just
be
the
token
binding.
Isn't
the
right
answer
that
well,
mutual
TLS
may
actually
fit
better
for
some
of
those
server
based
applications.
The
sweet
spot
for
token
binding
tends
to
be
more
than
mobile
applications,
which
are
dynamic
and
generally
not
clustered,
clients
that.
H
Actually,
a
lot
less
worried
about
the
refresh
token
being
bound
to
a
single
token
binding
ID,
just
because
that's
having
a
central
system,
or
at
least
a
single
distributive
system,
that
is
doing
token
refreshes
that
just
kind
of
a
reasonable
thing.
It's
the
access
token
that's
potentially
being
used
by
n
different
client
systems
to
you,
know
M
different
resource
servers
that
where
I
see
the
problem
but
I,
think
I
think
it's
a
good
point
that
if
this
doesn't
have
to
your
token
binding,
isn't
necessarily
the
hammer
to
use
on
every
now.
That's
absolutely
true.
K
So,
as
far
as
the
resource
server
is
concerned,
you're
either
you
or
some
sort
of
introspection.
So
in
principle
the
token
binding
ID
for
that
particular
access
token
can
be
contained
inside
of
the
access
token,
so
you're
not
really
just
because
you're
creating
a
new
access
token,
potentially
for
a
different
part
of
the
client.
That's
that's
talking
to
that
resource,
or
both
of
them
can
be
valid
at
the
same
time,
you're
not
revoking
it.
So
I
think
there
may
be
some
of
these.
These
larger
sort
of
edge
cases
that
we
might
want
to
go
through.
G
The
card
Amazon
yeah
I,
haven't
looked
at
this
back
in
a
while,
so
I,
don't
know
if
this
is
included,
but
just
picking
up
on
what
they're
talking
about
one
of
the
issues
in
the
past
is
that
your
TLS
layers
manage
at
your
load
balancer,
but
your
access
token
is
out
the
application
right,
and
so
how
do
you
get
the
trust
where
that
the
ID
vis
negotiated
somewhere
different
than
where
the
app's
trying
to
verify
it
sounds
like
not
necessarily
that
maybe
some
standardization
around?
How
does
a
load
balancer?
O
G
O
Thank
you,
nice.
It
sits
in
the
token
binding
working
group.
It's
call,
it
has
a
horrible
name.
I
can't
do
better.
It's
a
HTTP
tow
combining
with
TLS
terminating
reverse
proxies
or
something
like
that,
but
you
it's
really
short,
but
you
can
find
the
draft.
So
yes,
I,
agree.
It's
a
problem
that
I'm
trying
to
try
to
put
out
something.
That'll
help
solve
it.
So.
I
B
O
O
An
enhanced
security
profile
of
OAuth
2
based
on
TLS
client
certificates,
mutual
TLS.
The
draft
is
already
being
used
by
open
banking
PSD
to
ask
sort
of
regulatory
regimes,
as
well
as
being
preference
of
profile
by
other
stos.
What
is
it?
It's
an
asymmetric
key
based,
client
authentication
method
for
for
the
client
to
the
authorization
server
using
mutual
TLS.
O
Member
inside
the
confirmation
claim
this
one's
called
x5t
pound
sha-256,
and
it's
just
the
shock
to
be
656
thumbprint
of
the
certificate
embedded
inside
the
the
access
token
confirmation
clam.
A
few
things
have
happened
since
london
working
group
last
call
happened,
I
was
on
vacation
for
some
of
it,
so
that
was
fun
getting
the
feedback
was
trying
to
relax
on
the
beach
draft.
Seven
eight-nine
have
been
published.
Mostly,
it's
been
editorial
clarifications
and
improvements
from
people
making
my
text
better.
Thank
you.
O
All
one
sort
of
important
thing,
I
think
is
worth
noting
here
is
that
the
I
dropped
the
use
of
the
sender
constraint,
terminology
with
respect
to
access
tokens,
because
a
reviewer
pointed
out
that
going
back
to
the
pop
architecture.
Document
sender
constrained
is
defined
as
they
as
a
client,
authenticating
literally
authenticating
itself,
via
some
other
method
and
having
an
access,
token
constrained
or
bound
to
that
client.
Whereas
this
this
one
actually
is
kind
of
a
weird
middle,
but
more
than
anything,
the
access
token
is,
is
bound
to
the
certificate
itself
and
not
the
client
identity.
O
So
trying
to
be
more
consistent
with
terminology
here.
I
just
got
rid
of
the
sender,
constraint,
terminology
and
I
include
a
picture
of
NAT
here,
presenting
at
a
conference
just
a
couple
weeks
ago,
extolling
the
virtues
of
sender,
constraint,
tokens,
and
this
particular
draft
I-
guess
he's
not
here,
but
I
was
hoping
to
give
him
a
hard
time
about
using
the
wrong
terminology
and
push
on
it.
O
But
that's
the
only
major
like
difference,
that's
not
editorial
and
that
in
turn
came
along
with
a
changing
of
a
metadata
parameter,
name
to
reflect
that
small
but
legitimate,
breaking
change
and
that's
all
the
stuff.
That's
really
happened
and
I
wanted
to
make
the
pulp
fiction
reference
in
here
trying
to
find
the
shepard
for
this
document.
But
this
has.
O
P
O
That's
in
these
are
moving
thanks
for
all
the
reviews,
I'm
looking
forward
to
it
progressing
Oh
token
exchange
now
in
draft
14
real,
quick
overview.
This
is
like
a
STS
framework
via
the
token
endpoint,
on
an
authorization,
server,
SD,
SPN
security.
Token
service
I
want
to
try
to
provide
contacts
here,
but
it's
hard
because
it's
really
sort
of
a
Swiss
Army
knife
of
just
token
in
token
out
actually
two
tokens
and
in
some
cases
and
it's
oaken
out
with
with
you
know,
claims
or
content
representing
that
delegation.
O
This
is
just
one
use
case
which
is
sort
of
the
resource.
Server
as
a
client
gets
an
access
token
in
from
a
normal
off
flow
and
in
chain
exchanges
that,
with
its
own
authorization,
server
some
other
authorization
server,
sorry
its
own
authorization
server
only
to
to
get
a
token,
that's
appropriate
for
usage
on
some
back-end
system.
The
details
are
here:
it's
just,
but
it's
one
of
many
use
cases
that
are
facilitated
by
this
one.
That
I
just
hear
a
lot.
O
That's
enough
for
context
and
an
opportunity
to
show
a
picture
of
my
kid
from
a
year
ago
happening
since
I
eat
99.
So
99
in
Prague
was
actually
the
last
time
I
presented
on
this
because
it's
been
sort
of
in
the
throes
of
review
or
latter,
not
not
working
group.
But
a
lot
of
review.
We've
had
graphs,
10,
11,
12,
13
and
14
publish
clarified
some
stuff
around
the
Act
claim.
O
The
act
fame
is
the
actor
sort
of
a
claim
about
to
whom
this
token
and
its
privileges
have
been
delegated,
and
it
says
that
it
just
clarifies
that
only
that
the
top-level
claims
about
the
subject
and
the
token
itself,
as
well
as
the
current
actor,
which
is
the
most
recent
actor
in
the
in
Act
claim,
are
actually
to
be
considered
when
applying
access
control
decisions
had
several
clarifications
and
editorial
improvements.
Stuff
mostly
out
of
80
review.
I
did
change
with
working
group,
consent,
scope
and
client
ID
claim
names
to
what
they
are
now.
O
They
had
previously
been
SCP
and
CID
to
align
more
consistently
with.
What's
what's
in
the
introspection
draft
and
try
to
just
normalize
names,
I
try
to
really
to
go
with
the
shorthand
convention
there,
which
made
sense
at
the
time,
but
in
retrospect
and
having
some
developers
and
other
spec
writers.
Look
at
it.
O
Consistency
between
the
two
made
a
lot
more
sense
and
will
make
even
more
sense,
probably
when
Torsten
presents
his
next
draft
here
about
signed
introspection
responses
because
then
they'll
line
up
some
new,
you
are
eyes
were
added
a
long
time
ago
to
represent
sam'l
tokens,
no
normative
stuff
around
it.
Just
giving
you
our
eyes
to
say
hey.
This
is
a
sam'l
token
and
I
have
answered
quite
a
few
questions
and
with
this
fact
that
there's
no
normative
support
in
the
document
for
validating
or
issuing
access
tokens
from
other
authorization
servers.
O
Currently,
this
document
is
in
ad
evaluation
waiting
for
write-up
on
a
point
raised
that
I
think
I've
addressed
with
that
Act
claim
I
think
ekor
left.
So
it's
sort
of
that
kills
a
lot
of
this
type
slide,
but
I
had
three
things:
I
wanted
to
do
in
this
slide.
One
is
I,
always
include
pictures
of
the
places
where
we've
been
are
gone.
Closest
thing
I
had
here
was
a
picture
of
me
in
Bangkok
from
almost
20
years
ago.
O
I
did
want
you
to
get
the
one
night
in
Bangkok
song
stuck
in
your
head,
so
hopefully
I'll
have
that
going
on
the
rest
of
the
night
and
the
main
point
of
the
slide
was
to
try
and
sort
of
go
into
actually
closing
out
his
point
needed
and
moving
things
forward,
because
I
think
it's
been
taking
care
of
him.
We're
just
sort
of
sitting
here,
but
he's
he
left
earlier.
So
yeah.
B
B
M
Okay,
good
yeah,
then,
is
the
presentation
visible?
Yes,
okay,
because
I
can't
continue
it
so
I'm
gonna,
my
name
is
Dawson
notice
that
yes,
I've,
come
and
I
would
like
to
present
a
individual
draft
that
I
did
with
you
know
from
connected
ID,
it's
a
small
extension
to
token
introspection
next
slide,
please
which
allows
a
resource
server
to
request
and
then
authorization,
so
it
responds
with
a
chart.
We
thought,
can
you
please?
Oh
I
can't
see
which
slide
you're
showing
so
assumption
is
in
slide
number
two.
So
it
shows
the
the
extension.
M
M
We
are
looking
into
my
eye
assurance
level
use
cases
which
is
payment
and
electronic
signatures,
and
those
use
cases
require
the
resource
server
to
keep
track
of
all
the
data
that
it
consumed
from
authorization,
server
for
auditing
and
number
affiliation
our
purposes,
and
that's
why
it's
a
good
idea
to
have
a
science
assertion
on
resize
and
use
of
structured
access.
Token
is
not
always
possible,
for
example,
iterations
where
you
want
to
have
an
integrated
authorization
which
allows
the
client
to
obtain
a
grant.
M
That
is
good
for
interacting
with
different
API
is,
for
example,
to
sign
a
contract
on
one
hand
and
to
initiate
a
payment
on
the
other
side.
So
we
can't
carry
all
data
in
one
shot
because
it
would
violate
privacy
regulation
and
that's
why
we
sing
a
token
introspection
is
a
much
better
fit,
but
our
top
introspection
needs
to
support,
signing
and
potentially
encryption,
and
that's
why
we
have
written
this
graphical
next
slide.
Please
it
shows
in
detail
how
this
is
going
to
work.
It's
quite
simple,
and
there
is
a
new
except
declined.
M
You
that
the
resource
or
uses
the
accept
our
application,
slash
dot.
Gwt
next
slide.
Please
das
just
answers
with
the
perspective
GWT
next
slide.
Please
and
the
a
s
determines
whether
what
with
algorithms
are
being
used
and
whether
there
is
just
assigned
a
judge
or
a
a
signed
and
encrypted
job
either
by
default
or
the
a
s
may
supply
some
configuration
data
of
why
our
reservation,
the
draft
also
defines
some
registration
ways
for
that,
following
the
pattern
established
by
the
Open
ID
connect
on
dynamic
client
reservation
next
slide,
please.
M
So
that's
the
last
slide.
I
already
presented
it
off
in
London
and
got
some
feet
some
pushback
on
the
topic,
because
we
also
had
included
proxy
use
case
because
of
the
fees
by
feedback.
We
got
rid
of
that
on
focusing
on
the
non-repudiation
between
our
s
and
is
entirely
so.
What
we
did
is
we
incorporated.
We
did
a
version,
oh
one,
and
we
incorporates
feedback
from
the
mailing
list.
M
J
So
Justin
richer.
They
deleted
all
the
stuff
that
I
thought
was
a
terrible
idea
from
the
last
document.
So
that's
a
huge
improvement.
The
security
discussions
is
going
in
the
right
direction.
I
think
it
needs
a
lot
more
chewing
on,
but
that's
what
we
do
here
I
would
support
this
going
forward
with
the
caveat
and
I
raised
this
in
London
as
well
that
we
that
we
think
about
using
this
response
mechanism
for
other
parts
of
the
OAuth
protocol.
That
might
also
want
a
signed
response.