1
00:00:03,030 --> 00:00:06,180
Not all generative AI is created equal.

2
00:00:06,180 --> 00:00:09,090
In fact, if data security
or privacy-related concerns

3
00:00:09,090 --> 00:00:10,950
are holding your organization back,

4
00:00:10,950 --> 00:00:12,780
today I'll show you how the combination

5
00:00:12,780 --> 00:00:16,410
of Microsoft 365 Copilot and
the data security controls

6
00:00:16,410 --> 00:00:19,560
in Microsoft Purview provide
an enterprise-ready platform

7
00:00:19,560 --> 00:00:21,480
for GenAI in your organization.

8
00:00:21,480 --> 00:00:24,660
This way, GenAI is seamlessly
integrated into your workflow

9
00:00:24,660 --> 00:00:26,647
across familiar apps and experiences,

10
00:00:26,647 --> 00:00:29,040
all backed by unmatched data security

11
00:00:29,040 --> 00:00:32,820
and visibility to minimize data
risk and prevent data loss.

12
00:00:32,820 --> 00:00:35,550
First, let's level set
on a few Copilot security

13
00:00:35,550 --> 00:00:37,200
and privacy basics.

14
00:00:37,200 --> 00:00:39,240
Whether you're using the free Copilot Chat

15
00:00:39,240 --> 00:00:41,640
that's included with Microsoft 365

16
00:00:41,640 --> 00:00:44,430
or have a Microsoft 365 Copilot license,

17
00:00:44,430 --> 00:00:46,800
they both honor your
existing access permissions

18
00:00:46,800 --> 00:00:49,590
to work information in
SharePoint and OneDrive,

19
00:00:49,590 --> 00:00:51,960
your Teams meetings and your email,

20
00:00:51,960 --> 00:00:53,820
meaning generated AI responses

21
00:00:53,820 --> 00:00:56,763
can only be based on information
that you have access to.

22
00:00:57,600 --> 00:00:59,880
Importantly, after you submit a prompt,

23
00:00:59,880 --> 00:01:02,310
Copilot will retrieve relevant index data

24
00:01:02,310 --> 00:01:03,720
to generate a response.

25
00:01:03,720 --> 00:01:04,890
The data only stays

26
00:01:04,890 --> 00:01:08,100
within your Microsoft 365
service trust boundary

27
00:01:08,100 --> 00:01:09,660
and doesn't move out of it.

28
00:01:09,660 --> 00:01:11,130
Even when the data is presented

29
00:01:11,130 --> 00:01:13,950
to the large language models
to generate a response,

30
00:01:13,950 --> 00:01:15,900
information is kept separate to the model,

31
00:01:15,900 --> 00:01:17,430
and is not used to train it.

32
00:01:17,430 --> 00:01:19,500
This is in contrast to consumer apps,

33
00:01:19,500 --> 00:01:22,050
especially the free ones,
which are often designed

34
00:01:22,050 --> 00:01:23,490
to collect training data.

35
00:01:23,490 --> 00:01:25,200
As users upload files into them

36
00:01:25,200 --> 00:01:27,000
or paste content into their prompts,

37
00:01:27,000 --> 00:01:30,180
including sensitive data,
the data is now duplicated

38
00:01:30,180 --> 00:01:31,920
and stored in a location outside

39
00:01:31,920 --> 00:01:35,130
of your Microsoft 365
service trust boundary,

40
00:01:35,130 --> 00:01:36,900
removing any file access controls

41
00:01:36,900 --> 00:01:39,420
or classifications you've
applied in the process,

42
00:01:39,420 --> 00:01:41,640
placing your data at greater risk.

43
00:01:41,640 --> 00:01:43,500
And beyond being stored there for indexing

44
00:01:43,500 --> 00:01:45,120
or reasoning, it can be used

45
00:01:45,120 --> 00:01:47,520
to retrain the underlying model.

46
00:01:47,520 --> 00:01:49,710
Next, adding to the
foundational protections

47
00:01:49,710 --> 00:01:51,990
of Microsoft 365 Copilot,

48
00:01:51,990 --> 00:01:54,780
Microsoft Purview has
activity logging built in

49
00:01:54,780 --> 00:01:55,920
and helps you to discover

50
00:01:55,920 --> 00:01:57,720
and protect sensitive data

51
00:01:57,720 --> 00:01:59,310
where you get visibility into current

52
00:01:59,310 --> 00:02:01,410
and potential risks, such as the use

53
00:02:01,410 --> 00:02:05,430
of unprotected sensitive
data in Copilot interactions,

54
00:02:05,430 --> 00:02:06,842
classify and secure data

55
00:02:06,842 --> 00:02:08,940
where information protection helps you

56
00:02:08,940 --> 00:02:11,730
to automatically classify,
and apply sensitivity labels

57
00:02:11,730 --> 00:02:15,210
to data, ensuring it remains
protected even when it's used

58
00:02:15,210 --> 00:02:18,941
with Copilot, and detect
and mitigate insider risks

59
00:02:18,941 --> 00:02:20,220
where you can be alerted

60
00:02:20,220 --> 00:02:22,890
to employee activities with
Copilot that pose a risk

61
00:02:22,890 --> 00:02:24,780
to your data, and much more.

62
00:02:24,780 --> 00:02:26,010
Over the next few minutes,

63
00:02:26,010 --> 00:02:28,800
I'll focus on Purview
capabilities to get ahead of

64
00:02:28,800 --> 00:02:31,320
and prevent data loss and insider risks.

65
00:02:31,320 --> 00:02:34,140
We'll start in Data
Security Posture Management

66
00:02:34,140 --> 00:02:36,390
or DSPM for AI for short.

67
00:02:36,390 --> 00:02:39,270
DSPM for AI is the one place to get a rich

68
00:02:39,270 --> 00:02:42,750
and prioritized bird's eye view
on how Copilot is being used

69
00:02:42,750 --> 00:02:44,310
inside your organization

70
00:02:44,310 --> 00:02:46,740
and discover corresponding risks,

71
00:02:46,740 --> 00:02:48,000
along with recommendations

72
00:02:48,000 --> 00:02:49,710
to improve your data security posture

73
00:02:49,710 --> 00:02:52,110
that you can implement
right from the solution.

74
00:02:52,110 --> 00:02:54,780
Importantly, this is where
you'll find detailed dashboards

75
00:02:54,780 --> 00:02:58,890
for Microsoft 365 Copilot
usage, including agents.

76
00:02:58,890 --> 00:03:01,500
Then in Activity Explorer, we make it easy

77
00:03:01,500 --> 00:03:04,156
to see recent activities
with AI interactions

78
00:03:04,156 --> 00:03:06,210
that include sensitive information types,

79
00:03:06,210 --> 00:03:09,240
like credit cards, ID
numbers or bank accounts.

80
00:03:09,240 --> 00:03:11,760
And you can drill into each
activity to see details,

81
00:03:11,760 --> 00:03:15,030
as well as the prompt and
response text generated.

82
00:03:15,030 --> 00:03:17,040
One tip here, if you are seeing a lot

83
00:03:17,040 --> 00:03:18,930
of sensitive information exposed,

84
00:03:18,930 --> 00:03:21,450
it points to an information
oversharing issue

85
00:03:21,450 --> 00:03:22,590
where people have access

86
00:03:22,590 --> 00:03:25,680
to more information than
necessary to do their job.

87
00:03:25,680 --> 00:03:27,750
If you find yourself in this situation,

88
00:03:27,750 --> 00:03:30,630
I recommend you also check out
our recent show on the topic

89
00:03:30,630 --> 00:03:33,990
at aka.ms/OversharingMechanics

90
00:03:33,990 --> 00:03:36,300
where I dive into the
specific things you should do

91
00:03:36,300 --> 00:03:38,880
to assess your Microsoft 365 environment

92
00:03:38,880 --> 00:03:40,440
for potential oversharing risks

93
00:03:40,440 --> 00:03:43,320
to ensure the right people can
access the right information

94
00:03:43,320 --> 00:03:44,790
when using Copilot.

95
00:03:44,790 --> 00:03:48,420
Ultimately, DSPM for AI gives
you the visibility you need

96
00:03:48,420 --> 00:03:50,100
to establish a data security baseline

97
00:03:50,100 --> 00:03:52,440
for Copilot usage in your organization,

98
00:03:52,440 --> 00:03:55,740
and helps you put in place
preventative measures right away.

99
00:03:55,740 --> 00:03:58,560
In fact, without leaving DSPM for AI

100
00:03:58,560 --> 00:04:00,330
on the recommendations page,

101
00:04:00,330 --> 00:04:02,910
you'll find the policies
we advise everyone to use

102
00:04:02,910 --> 00:04:05,321
to improve data security, such as this one

103
00:04:05,321 --> 00:04:07,950
for detecting potentially
risky interactions

104
00:04:07,950 --> 00:04:11,730
using insider risk management
and other recommendations,

105
00:04:11,730 --> 00:04:15,030
like this one to detect
potentially unethical behavior

106
00:04:15,030 --> 00:04:18,060
using communication
compliance policies and more.

107
00:04:18,060 --> 00:04:19,260
From there, you can dive in

108
00:04:19,260 --> 00:04:21,810
to Microsoft Purview's
best-in-class solutions

109
00:04:21,810 --> 00:04:23,400
for more granular insights,

110
00:04:23,400 --> 00:04:26,520
and to configure specific
policies and protections.

111
00:04:26,520 --> 00:04:28,740
I'll start with information protection.

112
00:04:28,740 --> 00:04:30,810
You can manage data security controls

113
00:04:30,810 --> 00:04:33,510
with Microsoft 365 Copilot in scope

114
00:04:33,510 --> 00:04:35,700
with the information protection policies,

115
00:04:35,700 --> 00:04:39,300
and the sensitivity labels
that you have in use today.

116
00:04:39,300 --> 00:04:42,960
In fact, by default, any
Copilot response using content

117
00:04:42,960 --> 00:04:44,306
with sensitivity labels

118
00:04:44,306 --> 00:04:46,890
will automatically inherit
the highest priority label

119
00:04:46,890 --> 00:04:48,630
for the referenced content.

120
00:04:48,630 --> 00:04:51,360
And using data loss prevention policies,

121
00:04:51,360 --> 00:04:54,300
you can prevent Copilot
from processing any content

122
00:04:54,300 --> 00:04:57,240
that has a specific
sensitivity label applied.

123
00:04:57,240 --> 00:05:00,690
This way, even if users
have access to those files,

124
00:05:00,690 --> 00:05:03,330
Copilot will effectively
ignore this content

125
00:05:03,330 --> 00:05:06,330
as it retrieves relevant
information from Microsoft Graph

126
00:05:06,330 --> 00:05:08,100
used to generate responses.

127
00:05:08,100 --> 00:05:09,840
Insider risk management helps you

128
00:05:09,840 --> 00:05:12,840
to catch data risk based
on trending activities

129
00:05:12,840 --> 00:05:14,610
of people on your network

130
00:05:14,610 --> 00:05:18,150
using established user risk
indicators and thresholds,

131
00:05:18,150 --> 00:05:20,730
and then uses policies
to prevent accidental

132
00:05:20,730 --> 00:05:24,180
or intentional data misuse
as they interact with Copilot

133
00:05:24,180 --> 00:05:26,100
where you can easily create policies

134
00:05:26,100 --> 00:05:28,200
based on quick policy templates,

135
00:05:28,200 --> 00:05:29,670
like this one looking

136
00:05:29,670 --> 00:05:32,700
for high-risk data leak
patterns from insiders.

137
00:05:32,700 --> 00:05:36,330
By default, this quick policy
will scope all users in groups

138
00:05:36,330 --> 00:05:39,810
with a defined triggering
event of data exfiltration,

139
00:05:39,810 --> 00:05:43,680
along with activity indicators,
including external sharing,

140
00:05:43,680 --> 00:05:45,690
bulk downloads, label downgrades,

141
00:05:45,690 --> 00:05:48,540
and label removal in
addition to other activities

142
00:05:48,540 --> 00:05:51,000
that indicate a high risk of data theft.

143
00:05:51,000 --> 00:05:52,500
And it doesn't stop there.

144
00:05:52,500 --> 00:05:55,170
As individuals perform
more risky activities,

145
00:05:55,170 --> 00:05:58,200
those can add up to elevate
that user's risk level.

146
00:05:58,200 --> 00:06:01,830
Here, instead of manually
adjusting data security policies,

147
00:06:01,830 --> 00:06:03,870
using Adaptive Protection controls,

148
00:06:03,870 --> 00:06:05,790
you can also limit Copilot use

149
00:06:05,790 --> 00:06:08,250
depending on a user's dynamic risk level,

150
00:06:08,250 --> 00:06:10,140
for example, when a user exceeds

151
00:06:10,140 --> 00:06:12,420
your defined risk condition thresholds

152
00:06:12,420 --> 00:06:15,750
to reach an elevated risk
level, as you can see here.

153
00:06:15,750 --> 00:06:19,350
Using Conditional Access
policies in Microsoft Entra,

154
00:06:19,350 --> 00:06:22,170
in this case based on
authentication context,

155
00:06:22,170 --> 00:06:24,330
as well as the condition for insider risk

156
00:06:24,330 --> 00:06:27,150
that you set in Microsoft
Purview, you can choose

157
00:06:27,150 --> 00:06:30,060
to block their permission when
attempting to access sites

158
00:06:30,060 --> 00:06:32,160
with a specific sensitivity label.

159
00:06:32,160 --> 00:06:34,560
That way, even if a user is granted access

160
00:06:34,560 --> 00:06:36,990
to a SharePoint site resource by an owner,

161
00:06:36,990 --> 00:06:38,190
their access will be blocked

162
00:06:38,190 --> 00:06:40,920
by the Conditional Access policy you set.

163
00:06:40,920 --> 00:06:42,270
Again, this is important

164
00:06:42,270 --> 00:06:45,300
because Copilot honors the
user's existing permissions

165
00:06:45,300 --> 00:06:46,950
to work with information.

166
00:06:46,950 --> 00:06:49,230
This way, Copilot will
not return information

167
00:06:49,230 --> 00:06:51,600
that they do not have access to.

168
00:06:51,600 --> 00:06:53,220
Next, Communication Compliance

169
00:06:53,220 --> 00:06:55,950
is a related insider risk
solution that can act

170
00:06:55,950 --> 00:06:58,800
on potentially inappropriate
Copilot interactions.

171
00:06:58,800 --> 00:07:01,230
In fact, there are specific policy options

172
00:07:01,230 --> 00:07:03,690
for Microsoft 365 Copilot interactions

173
00:07:03,690 --> 00:07:06,720
in communication compliance
where you can flag jailbreak

174
00:07:06,720 --> 00:07:08,280
or prompt injection attempts

175
00:07:08,280 --> 00:07:10,350
using Prompt Shields classifiers.

176
00:07:10,350 --> 00:07:12,240
Communication compliance can be set

177
00:07:12,240 --> 00:07:14,040
to alert reviewers of that activity

178
00:07:14,040 --> 00:07:16,440
so they can easily discover policy matches

179
00:07:16,440 --> 00:07:18,390
and take corresponding actions.

180
00:07:18,390 --> 00:07:19,980
For example, if a person tries

181
00:07:19,980 --> 00:07:22,320
to use Copilot in an inappropriate way,

182
00:07:22,320 --> 00:07:24,780
like trying to get it to
work around its instructions

183
00:07:24,780 --> 00:07:27,420
to generate content
that Copilot shouldn't,

184
00:07:27,420 --> 00:07:29,070
it will report on that activity,

185
00:07:29,070 --> 00:07:29,903
and you'll also be able

186
00:07:29,903 --> 00:07:31,980
to see the response informing the user

187
00:07:31,980 --> 00:07:33,690
that their activity was blocked.

188
00:07:33,690 --> 00:07:35,670
Once you have the controls
you want in place,

189
00:07:35,670 --> 00:07:38,970
it's a good idea to keep
going back to DSPM for AI

190
00:07:38,970 --> 00:07:40,680
so you can see where Copilot usage

191
00:07:40,680 --> 00:07:42,840
is matching your data security policies.

192
00:07:42,840 --> 00:07:46,260
Sensitive interactions per
AI app shows you interactions

193
00:07:46,260 --> 00:07:48,630
based on sensitive information types.

194
00:07:48,630 --> 00:07:52,260
Top unethical AI interactions
surfaces insights based

195
00:07:52,260 --> 00:07:55,230
on the communication compliance
controls you've defined.

196
00:07:55,230 --> 00:07:59,220
Top sensitivity labels referenced
in Microsoft 365 Copilot

197
00:07:59,220 --> 00:08:00,750
reports on the labels you've created,

198
00:08:00,750 --> 00:08:02,460
and applied to reference content.

199
00:08:02,460 --> 00:08:04,380
And you can see Copilot
interactions mapped

200
00:08:04,380 --> 00:08:06,510
to insider risk severity levels.

201
00:08:06,510 --> 00:08:09,180
Then digging into these reports
shows you a filtered view

202
00:08:09,180 --> 00:08:11,250
of activities in Activity Explorer

203
00:08:11,250 --> 00:08:13,770
with time-based trends
and details for each.

204
00:08:13,770 --> 00:08:16,590
Additionally, because all
Copilot interactions are logged,

205
00:08:16,590 --> 00:08:19,230
like other Microsoft 365 activities

206
00:08:19,230 --> 00:08:22,219
in email, Microsoft Teams,
SharePoint and OneDrive,

207
00:08:22,219 --> 00:08:23,250
you can now use

208
00:08:23,250 --> 00:08:25,980
the new data security
investigation solution.

209
00:08:25,980 --> 00:08:29,130
This uses AI to quickly reason
over thousands of items,

210
00:08:29,130 --> 00:08:31,014
including Copilot Chat interactions

211
00:08:31,014 --> 00:08:33,900
to help you investigate the
potential cause of risks

212
00:08:33,900 --> 00:08:36,570
for known data leaks in similar incidents.

213
00:08:36,570 --> 00:08:39,090
So that's how Microsoft 365 Copilot,

214
00:08:39,090 --> 00:08:40,380
along with Microsoft Purview,

215
00:08:40,380 --> 00:08:42,300
provides comprehensive controls

216
00:08:42,300 --> 00:08:44,820
to help protect your data, minimize risk,

217
00:08:44,820 --> 00:08:47,430
and quickly identify Copilot
interactions that could lead

218
00:08:47,430 --> 00:08:50,430
to compromise so you can
take corrective actions.

219
00:08:50,430 --> 00:08:53,190
No other AI solution has this level

220
00:08:53,190 --> 00:08:54,630
of protection and control.

221
00:08:54,630 --> 00:08:58,893
To learn more, check out
aka.ms/M365CopilotwithPurview.

222
00:09:00,660 --> 00:09:03,248
Keep watching Microsoft
Mechanics for the latest updates

223
00:09:03,248 --> 00:09:05,248
and thanks for watching.