WEBVTT

1
00:00:00.080 --> 00:00:02.340
In this section of the course, we are going

2
00:00:02.340 --> 00:00:04.890
to discuss artificial intelligence.

3
00:00:04.890 --> 00:00:07.470
The artificial intelligence section of the course

4
00:00:07.470 --> 00:00:11.160
focuses on Domain 1, Governance, Risk, and Compliance,

5
00:00:11.160 --> 00:00:14.010
as well as Domain 3, Security Engineering,

6
00:00:14.010 --> 00:00:18.810
specifically Objectives 1.5 and 3.6.

7
00:00:18.810 --> 00:00:21.600
Objective 1.5 states that you must be able

8
00:00:21.600 --> 00:00:24.720
to summarize the information security challenges

9
00:00:24.720 --> 00:00:27.870
associated with artificial intelligence adoption.

10
00:00:27.870 --> 00:00:31.380
Objective 3.6 states that given a scenario,

11
00:00:31.380 --> 00:00:35.250
you must be able to use automation to secure the enterprise.

12
00:00:35.250 --> 00:00:39.420
Artificial intelligence or AI has already revolutionized

13
00:00:39.420 --> 00:00:41.460
how we interact with technology.

14
00:00:41.460 --> 00:00:45.390
However, with these advancements come significant challenges

15
00:00:45.390 --> 00:00:47.790
such as the need for responsible use

16
00:00:47.790 --> 00:00:49.770
and ethical considerations.

17
00:00:49.770 --> 00:00:53.910
Additionally, as AI becomes more embedded in our lives,

18
00:00:53.910 --> 00:00:58.110
it introduces new risks, including potential vulnerabilities

19
00:00:58.110 --> 00:01:01.680
and the misuse of AI for malicious purposes.

20
00:01:01.680 --> 00:01:03.990
Understanding and addressing these issues

21
00:01:03.990 --> 00:01:07.650
and risks early is crucial to ensuring that AI

22
00:01:07.650 --> 00:01:10.050
continues to contribute positively

23
00:01:10.050 --> 00:01:12.390
to our day-to-day and beyond.

24
00:01:12.390 --> 00:01:14.130
As we go through this section,

25
00:01:14.130 --> 00:01:15.690
we will cover many topics

26
00:01:15.690 --> 00:01:17.550
related to artificial intelligence,

27
00:01:17.550 --> 00:01:20.700
including generative artificial intelligence,

28
00:01:20.700 --> 00:01:24.960
ethical governance considerations, legal and privacy risks,

29
00:01:24.960 --> 00:01:28.530
threats to the model, AI-enabled attacks,

30
00:01:28.530 --> 00:01:33.530
AI manipulation attacks, AI usage risks, and AI bots.

31
00:01:33.960 --> 00:01:35.580
First, we will look at

32
00:01:35.580 --> 00:01:38.910
generative artificial intelligence or AI.

33
00:01:38.910 --> 00:01:42.720
Generative AI refers to artificial intelligence systems

34
00:01:42.720 --> 00:01:46.200
that create new content such as text, images,

35
00:01:46.200 --> 00:01:47.850
code, or music.

36
00:01:47.850 --> 00:01:51.030
Generative AI learns patterns from existing data

37
00:01:51.030 --> 00:01:52.920
and generates outputs that resemble

38
00:01:52.920 --> 00:01:54.990
or extend the original data.

39
00:01:54.990 --> 00:01:59.700
Models like the generative pre-trained transformer or GPT

40
00:01:59.700 --> 00:02:01.470
are trained on large data sets

41
00:02:01.470 --> 00:02:05.100
to understand language, context and structure.

42
00:02:05.100 --> 00:02:07.320
Based on their amount of training data,

43
00:02:07.320 --> 00:02:11.730
GPTs can produce coherent and contextually-relevant content.

44
00:02:11.730 --> 00:02:15.210
Baseline generative AI concepts include code assist

45
00:02:15.210 --> 00:02:16.710
and documentation.

46
00:02:16.710 --> 00:02:19.260
Code assist helps developers by suggesting

47
00:02:19.260 --> 00:02:20.790
or completing code.

48
00:02:20.790 --> 00:02:24.810
Documentation automatically generates technical documents

49
00:02:24.810 --> 00:02:27.030
such as user guides, release notes,

50
00:02:27.030 --> 00:02:29.400
and other technical documentation.

51
00:02:29.400 --> 00:02:33.000
In practice, a developer might use generative AI

52
00:02:33.000 --> 00:02:37.020
to assist in writing code by receiving realtime suggestions

53
00:02:37.020 --> 00:02:39.780
or generating comprehensive documentation

54
00:02:39.780 --> 00:02:41.820
based on their code base.

55
00:02:41.820 --> 00:02:45.870
Next, we will explore ethical and governance considerations.

56
00:02:45.870 --> 00:02:48.420
Ethical and governance considerations are used

57
00:02:48.420 --> 00:02:50.730
to establish frameworks and guidelines

58
00:02:50.730 --> 00:02:54.420
to ensure AI is developed and used responsibly,

59
00:02:54.420 --> 00:02:56.640
and that it aligns with societal values

60
00:02:56.640 --> 00:02:58.620
and organizational principles.

61
00:02:58.620 --> 00:03:00.510
Ethical governance considerations

62
00:03:00.510 --> 00:03:03.120
include the ethical governance of AI

63
00:03:03.120 --> 00:03:06.510
and organizational policies on the use of AI.

64
00:03:06.510 --> 00:03:09.540
The ethical governance of AI refers to the structures

65
00:03:09.540 --> 00:03:11.790
and practices put in place

66
00:03:11.790 --> 00:03:14.850
to guide the ethical deployment of AI.

67
00:03:14.850 --> 00:03:16.770
Ethical governance further ensures

68
00:03:16.770 --> 00:03:19.770
that this technology is used in ways that are fair,

69
00:03:19.770 --> 00:03:22.230
transparent, and accountable.

70
00:03:22.230 --> 00:03:24.750
Organizational policies on the use of AI

71
00:03:24.750 --> 00:03:28.650
are specific rules and guidelines created by companies

72
00:03:28.650 --> 00:03:32.940
to govern how AI is implemented within their organization.

73
00:03:32.940 --> 00:03:37.050
Local organizational policies ensure that AI use aligns

74
00:03:37.050 --> 00:03:38.910
with both legal requirements

75
00:03:38.910 --> 00:03:42.000
and an organization's ethical standards.

76
00:03:42.000 --> 00:03:45.060
For example, an organization might implement

77
00:03:45.060 --> 00:03:49.020
an ethical governance framework that requires all AI models

78
00:03:49.020 --> 00:03:52.980
to undergo bias testing, and mandates transparency

79
00:03:52.980 --> 00:03:55.650
into the AI decision-making process.

80
00:03:55.650 --> 00:03:59.250
This requirement can be supported by internal policies

81
00:03:59.250 --> 00:04:01.170
that restrict the use of AI

82
00:04:01.170 --> 00:04:03.900
in areas where it could lead to discrimination

83
00:04:03.900 --> 00:04:05.820
or privacy violations.

84
00:04:05.820 --> 00:04:09.330
Then we will look at legal and privacy rules.

85
00:04:09.330 --> 00:04:13.440
Legal and privacy rules ensure that AI systems are designed

86
00:04:13.440 --> 00:04:16.890
and deployed in compliance with laws and regulations

87
00:04:16.890 --> 00:04:20.190
to protect individual rights and personal data.

88
00:04:20.190 --> 00:04:24.180
Legal and privacy implications include potential misuse

89
00:04:24.180 --> 00:04:28.230
and explainable versus non-explainable AI models.

90
00:04:28.230 --> 00:04:32.490
The potential misuse of AI arises when AI systems are used

91
00:04:32.490 --> 00:04:35.160
in ways that could lead to discrimination,

92
00:04:35.160 --> 00:04:38.880
privacy violations, or other harmful outcomes.

93
00:04:38.880 --> 00:04:42.990
This misuse is often exacerbated by the opacity

94
00:04:42.990 --> 00:04:44.790
of the models that are used.

95
00:04:44.790 --> 00:04:48.510
This infers that AI models may be explainable

96
00:04:48.510 --> 00:04:50.220
or non-explainable.

97
00:04:50.220 --> 00:04:53.310
Explainable models are those that provide clear,

98
00:04:53.310 --> 00:04:55.740
understandable reasons for their decisions.

99
00:04:55.740 --> 00:04:57.330
Non-explainable models

100
00:04:57.330 --> 00:04:59.490
like many deep learning algorithms

101
00:04:59.490 --> 00:05:02.400
can be difficult to interpret or understand.

102
00:05:02.400 --> 00:05:05.460
Non-explainable models make it difficult to determine

103
00:05:05.460 --> 00:05:08.220
why the AI made the decisions it did.

104
00:05:08.220 --> 00:05:11.400
These are often referred to as black box models.

105
00:05:11.400 --> 00:05:16.320
Non-explainable AI models raise challenges in legal contexts

106
00:05:16.320 --> 00:05:18.480
where transparency is required.

107
00:05:18.480 --> 00:05:21.510
For example, in a credit scoring application,

108
00:05:21.510 --> 00:05:24.720
using a non-explainable model could result in

109
00:05:24.720 --> 00:05:29.130
unfair loan denials that cannot be easily justified.

110
00:05:29.130 --> 00:05:32.400
These AI decisions could lead to legal challenges

111
00:05:32.400 --> 00:05:35.550
considering anti-discrimination laws.

112
00:05:35.550 --> 00:05:38.850
Next, we will explore threats to the model.

113
00:05:38.850 --> 00:05:42.300
Threats to the artificial intelligence or AI model

114
00:05:42.300 --> 00:05:44.970
are those that can compromise the integrity,

115
00:05:44.970 --> 00:05:48.990
security, or confidentiality of AI systems.

116
00:05:48.990 --> 00:05:53.100
The compromise of AI models can lead to incorrect decisions

117
00:05:53.100 --> 00:05:55.950
or unauthorized access to sensitive data.

118
00:05:55.950 --> 00:05:58.440
These threats are potential attack paths

119
00:05:58.440 --> 00:06:00.030
for malicious actors.

120
00:06:00.030 --> 00:06:03.300
Threats to the model include prompt injection,

121
00:06:03.300 --> 00:06:07.170
unsecured output handling, training data poisoning,

122
00:06:07.170 --> 00:06:11.370
model denial of service, supply chain vulnerabilities,

123
00:06:11.370 --> 00:06:14.190
model theft, and model inversion.

124
00:06:14.190 --> 00:06:16.590
Prompt injection occurs when an attacker

125
00:06:16.590 --> 00:06:20.430
manipulates input prompts to alter the model's behavior.

126
00:06:20.430 --> 00:06:22.950
Prompt injection may then produce harmful

127
00:06:22.950 --> 00:06:24.990
or unintended outputs.

128
00:06:24.990 --> 00:06:27.720
Unsecured output handling can be exploited

129
00:06:27.720 --> 00:06:31.470
to leak sensitive information or execute malicious commands.

130
00:06:31.470 --> 00:06:33.720
Next, training data poisoning

131
00:06:33.720 --> 00:06:36.960
is the intentional manipulation of training data

132
00:06:36.960 --> 00:06:39.990
to skew the model's predictions or performance.

133
00:06:39.990 --> 00:06:42.660
Training data poisoning can lead to bias

134
00:06:42.660 --> 00:06:46.230
or incorrect decision making and operational failures,

135
00:06:46.230 --> 00:06:49.350
security violations or ethical issues

136
00:06:49.350 --> 00:06:53.610
that undermine the trust and reliability of the AI model.

137
00:06:53.610 --> 00:06:55.620
Model denial of service attacks

138
00:06:55.620 --> 00:06:58.380
aim to overwhelm the AI system,

139
00:06:58.380 --> 00:07:00.930
preventing legitimate users from using it.

140
00:07:00.930 --> 00:07:03.930
Next, supply chain vulnerabilities emerge

141
00:07:03.930 --> 00:07:07.110
when compromised components or third party libraries

142
00:07:07.110 --> 00:07:08.880
used in model development

143
00:07:08.880 --> 00:07:12.570
are exploited to insert malicious code, backdoors,

144
00:07:12.570 --> 00:07:15.780
or alter the functionality of the AI system.

145
00:07:15.780 --> 00:07:19.980
AI supply chain attacks can lead to unauthorized access,

146
00:07:19.980 --> 00:07:22.920
data breaches, or the deployment of flawed models

147
00:07:22.920 --> 00:07:27.240
that behave unpredictably or maliciously in production.

148
00:07:27.240 --> 00:07:30.270
Model theft is the unauthorized extraction

149
00:07:30.270 --> 00:07:32.400
of a model's intellectual property.

150
00:07:32.400 --> 00:07:35.310
Model theft could allow attackers to replicate

151
00:07:35.310 --> 00:07:37.080
or misuse the model.

152
00:07:37.080 --> 00:07:40.230
Finally, model inversion allows attackers

153
00:07:40.230 --> 00:07:44.010
to reconstruct sensitive data used during AI training.

154
00:07:44.010 --> 00:07:46.380
By analyzing the model's outputs,

155
00:07:46.380 --> 00:07:49.680
an attacker could use model inversion to extract

156
00:07:49.680 --> 00:07:53.070
and expose sensitive data, such as personal information

157
00:07:53.070 --> 00:07:54.780
or proprietary details

158
00:07:54.780 --> 00:07:57.780
that were used during training the AI model.

159
00:07:57.780 --> 00:08:00.810
This could lead to privacy breaches and data theft.

160
00:08:00.810 --> 00:08:04.380
In application, an attacker might use prompt injection

161
00:08:04.380 --> 00:08:07.560
to cause a chat bot to generate harmful advice,

162
00:08:07.560 --> 00:08:09.690
exploit unsecured output handling

163
00:08:09.690 --> 00:08:12.030
to extract confidential information,

164
00:08:12.030 --> 00:08:14.400
and engage in training data poisoning

165
00:08:14.400 --> 00:08:18.330
to bias an AI model used for content moderation.

166
00:08:18.330 --> 00:08:21.870
Each of these potential vulnerabilities underscore the need

167
00:08:21.870 --> 00:08:24.480
for managing threats to the AI model.

168
00:08:24.480 --> 00:08:28.290
Following that, we will look at AI enabled attacks.

169
00:08:28.290 --> 00:08:30.270
AI enabled attacks are attacks

170
00:08:30.270 --> 00:08:32.280
that leverage artificial intelligence

171
00:08:32.280 --> 00:08:35.100
to enhance the sophistication, scale,

172
00:08:35.100 --> 00:08:37.560
and effectiveness of malicious activities,

173
00:08:37.560 --> 00:08:41.190
making them more difficult to detect and defend against.

174
00:08:41.190 --> 00:08:45.090
AI enabled attacks include unsecure plugin design,

175
00:08:45.090 --> 00:08:49.320
AI pipeline injectors, and automated exploit generation.

176
00:08:49.320 --> 00:08:52.200
An AI plugin is an add-on or extension

177
00:08:52.200 --> 00:08:55.170
that integrates artificial intelligence capabilities

178
00:08:55.170 --> 00:08:57.810
into existing software or platforms.

179
00:08:57.810 --> 00:09:02.070
AI plugins enable enhanced functionality such as automation,

180
00:09:02.070 --> 00:09:05.940
data analysis, or personalized user experiences.

181
00:09:05.940 --> 00:09:09.120
Unsecured plugin design can leave AI systems

182
00:09:09.120 --> 00:09:12.180
vulnerable to exploitation, allowing attackers

183
00:09:12.180 --> 00:09:14.160
to gain unauthorized access

184
00:09:14.160 --> 00:09:17.220
through poorly secured extensions or integrations.

185
00:09:17.220 --> 00:09:21.060
AI pipeline injectors place malicious data or code

186
00:09:21.060 --> 00:09:23.910
in the AI training or deployment pipeline.

187
00:09:23.910 --> 00:09:27.510
AI pipeline injection can alter the model's behavior

188
00:09:27.510 --> 00:09:31.110
in harmful ways, such as introducing bias,

189
00:09:31.110 --> 00:09:34.110
generating incorrect or misleading outputs,

190
00:09:34.110 --> 00:09:37.080
creating backdoors for unauthorized access,

191
00:09:37.080 --> 00:09:39.510
or compromising the model's integrity.

192
00:09:39.510 --> 00:09:42.900
Automated exploit generation leverages AI

193
00:09:42.900 --> 00:09:44.070
to quickly discover

194
00:09:44.070 --> 00:09:47.310
and create exploits for unpatched vulnerabilities.

195
00:09:47.310 --> 00:09:50.880
Automated exploit generation significantly increases

196
00:09:50.880 --> 00:09:53.430
the speed and precision of attacks.

197
00:09:53.430 --> 00:09:55.860
For example, an attacker could exploit

198
00:09:55.860 --> 00:09:59.490
an unsecured AI plugin in a web application,

199
00:09:59.490 --> 00:10:02.550
inject malicious code into the AI model

200
00:10:02.550 --> 00:10:04.110
during its training phase,

201
00:10:04.110 --> 00:10:08.130
and then use AI to automatically generate exploits

202
00:10:08.130 --> 00:10:10.410
targeting the compromised system.

203
00:10:10.410 --> 00:10:14.190
Then we will explore AI manipulation attacks.

204
00:10:14.190 --> 00:10:17.820
AI manipulation attacks involve the intentional alteration

205
00:10:17.820 --> 00:10:21.330
or deception of AI systems to produce harmful

206
00:10:21.330 --> 00:10:22.950
or misleading outcomes.

207
00:10:22.950 --> 00:10:24.840
AI enabled manipulation attacks

208
00:10:24.840 --> 00:10:27.780
include social engineering and deepfakes

209
00:10:27.780 --> 00:10:30.720
through digital media and interactivity.

210
00:10:30.720 --> 00:10:33.030
Social engineering refers to the techniques

211
00:10:33.030 --> 00:10:34.830
that exploit human trust

212
00:10:34.830 --> 00:10:37.440
to deceive individuals into interacting

213
00:10:37.440 --> 00:10:39.660
with manipulated AI outputs.

214
00:10:39.660 --> 00:10:43.230
Interactivity may occur through interactive chatbots

215
00:10:43.230 --> 00:10:45.420
or social media platforms.

216
00:10:45.420 --> 00:10:49.470
Manipulating AI outputs could include deep fake videos

217
00:10:49.470 --> 00:10:53.520
or audio that appear genuine but are entirely fabricated.

218
00:10:53.520 --> 00:10:57.270
Deepfake technology uses AI to create realistic

219
00:10:57.270 --> 00:10:59.280
but fake digital media.

220
00:10:59.280 --> 00:11:02.100
Deep fakes can be distributed through social media,

221
00:11:02.100 --> 00:11:04.980
video conferencing, and streaming services

222
00:11:04.980 --> 00:11:08.970
to convincingly mimic real people, leading to misinformation

223
00:11:08.970 --> 00:11:10.890
or reputational damage.

224
00:11:10.890 --> 00:11:15.090
For example, an attacker might use deepfake technology

225
00:11:15.090 --> 00:11:17.640
to create a video of a company executive

226
00:11:17.640 --> 00:11:19.530
making false statements,

227
00:11:19.530 --> 00:11:22.500
then employ social engineering tactics

228
00:11:22.500 --> 00:11:24.720
to distribute the video widely,

229
00:11:24.720 --> 00:11:27.840
causing significant harm to the company's reputation

230
00:11:27.840 --> 00:11:29.490
and misleading the public.

231
00:11:29.490 --> 00:11:32.490
Next, we will look at AI usage risks.

232
00:11:32.490 --> 00:11:35.160
AI usage risks are the potential dangers

233
00:11:35.160 --> 00:11:37.110
and unintended consequences

234
00:11:37.110 --> 00:11:39.000
that arise from the implementation

235
00:11:39.000 --> 00:11:42.180
and reliance on artificial intelligence systems.

236
00:11:42.180 --> 00:11:45.600
AI usage risks may be realized when AI systems

237
00:11:45.600 --> 00:11:48.480
are not properly managed or understood.

238
00:11:48.480 --> 00:11:51.960
Risks of AI usage include overreliance,

239
00:11:51.960 --> 00:11:54.720
sensitive information disclosure to the model,

240
00:11:54.720 --> 00:11:57.660
sensitive information disclosure from the model,

241
00:11:57.660 --> 00:12:01.260
and excessive agency of the AI.

242
00:12:01.260 --> 00:12:04.260
Over-reliance on AI can lead to situations

243
00:12:04.260 --> 00:12:06.750
where human judgment is undervalued,

244
00:12:06.750 --> 00:12:08.490
potentially resulting in decisions

245
00:12:08.490 --> 00:12:11.400
that overlook important contextual factors

246
00:12:11.400 --> 00:12:13.740
that AI cannot calculate for.

247
00:12:13.740 --> 00:12:16.590
Sensitive information disclosure to the model

248
00:12:16.590 --> 00:12:19.110
occurs when private or confidential data

249
00:12:19.110 --> 00:12:23.910
is input into AI systems without adequate safeguards.

250
00:12:23.910 --> 00:12:26.550
Sensitive information disclosure to the model

251
00:12:26.550 --> 00:12:29.940
raises the risk that disclosure data could be mishandled

252
00:12:29.940 --> 00:12:32.700
or exploited by the AI model.

253
00:12:32.700 --> 00:12:35.280
Sensitive information disclosure from the model

254
00:12:35.280 --> 00:12:38.820
involves the unintentional release of private data

255
00:12:38.820 --> 00:12:41.610
that the AI has learned during its training.

256
00:12:41.610 --> 00:12:44.100
Sensitive information disclosure from the model

257
00:12:44.100 --> 00:12:46.290
can lead to privacy breaches.

258
00:12:46.290 --> 00:12:51.290
Finally, excessive agency of the AI refers to situations

259
00:12:51.300 --> 00:12:55.860
where AI systems are given too much control or autonomy.

260
00:12:55.860 --> 00:12:58.800
AIs with excessive agency can make decisions

261
00:12:58.800 --> 00:13:01.110
or take actions that are harmful

262
00:13:01.110 --> 00:13:03.870
or not aligned with societal values.

263
00:13:03.870 --> 00:13:07.200
In poor practice and demonstrating overreliance,

264
00:13:07.200 --> 00:13:10.800
a healthcare provider might rely on an AI system

265
00:13:10.800 --> 00:13:12.480
for diagnosing patients,

266
00:13:12.480 --> 00:13:14.430
inputting sensitive patient data

267
00:13:14.430 --> 00:13:17.010
without sufficient privacy protections.

268
00:13:17.010 --> 00:13:19.920
And then the risk the AI inadvertently reveals

269
00:13:19.920 --> 00:13:22.350
private information during its operations,

270
00:13:22.350 --> 00:13:25.020
while also allowing the AI too much autonomy

271
00:13:25.020 --> 00:13:27.450
in making critical health decisions.

272
00:13:27.450 --> 00:13:30.600
Finally, we will explore AI bots.

273
00:13:30.600 --> 00:13:33.330
AI bots are automated software programs

274
00:13:33.330 --> 00:13:35.520
that interact with users or systems

275
00:13:35.520 --> 00:13:37.800
to perform specific tasks.

276
00:13:37.800 --> 00:13:40.710
AI bots often simulate human behavior

277
00:13:40.710 --> 00:13:44.100
in areas like customer service or data management.

278
00:13:44.100 --> 00:13:48.120
AI bots may also serve as assistance and digital workers.

279
00:13:48.120 --> 00:13:52.800
AI bot concepts include access and permissions, guardrails,

280
00:13:52.800 --> 00:13:57.150
data loss prevention, and the disclosure of AI usage.

281
00:13:57.150 --> 00:14:00.300
Access and permissions refer to the controls put in place

282
00:14:00.300 --> 00:14:03.210
to ensure that AI bots only interact with

283
00:14:03.210 --> 00:14:07.350
data and systems they are explicitly authorized to access.

284
00:14:07.350 --> 00:14:09.600
By doing this, access and permissions

285
00:14:09.600 --> 00:14:12.300
prevent unauthorized data manipulation

286
00:14:12.300 --> 00:14:14.100
and security breaches.

287
00:14:14.100 --> 00:14:17.430
Guardrails are predefined boundaries or rules

288
00:14:17.430 --> 00:14:20.250
that restrict the actions of AI bots.

289
00:14:20.250 --> 00:14:22.140
Guardrails ensure AI bots

290
00:14:22.140 --> 00:14:25.320
operate with safe and intended parameters.

291
00:14:25.320 --> 00:14:29.310
Data loss prevention strategies are used with AI bots

292
00:14:29.310 --> 00:14:32.370
to prevent sensitive information from being accessed

293
00:14:32.370 --> 00:14:34.050
or leaked inadvertently.

294
00:14:34.050 --> 00:14:38.400
Finally, disclosure of AI usage involves informing users

295
00:14:38.400 --> 00:14:40.950
that they are interacting with an AI bot

296
00:14:40.950 --> 00:14:42.600
rather than a human.

297
00:14:42.600 --> 00:14:45.990
Disclosure of AI usage is important for transparency

298
00:14:45.990 --> 00:14:47.820
and maintaining trust.

299
00:14:47.820 --> 00:14:50.400
In application, a customer service bot

300
00:14:50.400 --> 00:14:53.220
with proper access controls and guardrails

301
00:14:53.220 --> 00:14:56.760
might be used to assist users with account inquiries.

302
00:14:56.760 --> 00:14:59.670
Data loss prevention measures would ensure the bot

303
00:14:59.670 --> 00:15:03.240
cannot access or leak sensitive financial information,

304
00:15:03.240 --> 00:15:06.300
and clear disclosure would let customers know

305
00:15:06.300 --> 00:15:10.140
they're interacting with an AI system and not a human.

306
00:15:10.140 --> 00:15:12.780
To finish things off, we'll take a short quiz

307
00:15:12.780 --> 00:15:15.630
to see what you learned during this section of the course,

308
00:15:15.630 --> 00:15:19.320
and we will review each of those quiz questions fully

309
00:15:19.320 --> 00:15:22.380
to ensure you can explain why the right answers were right

310
00:15:22.380 --> 00:15:24.210
and the wrong answers were wrong.

311
00:15:24.210 --> 00:15:27.630
So let's get ready to dive into artificial intelligence

312
00:15:27.630 --> 00:15:29.523
in this section of the course.