WEBVTT 1 00:00:00.000 --> 00:00:01.290 In this lesson, 2 00:00:01.290 --> 00:00:04.020 we will learn about artificial intelligence, 3 00:00:04.020 --> 00:00:07.560 or AI manipulation attacks. 4 00:00:07.560 --> 00:00:12.090 AI manipulation attacks involve the intentional alteration 5 00:00:12.090 --> 00:00:16.530 or deception of AI systems to produce harmful, 6 00:00:16.530 --> 00:00:19.050 or misleading outcomes. 7 00:00:19.050 --> 00:00:23.790 AI enabled manipulation attacks include social engineering 8 00:00:23.790 --> 00:00:28.500 and deepfakes through digital media and interactivity. 9 00:00:28.500 --> 00:00:31.200 Social engineering refers to techniques 10 00:00:31.200 --> 00:00:35.370 that exploit human trust to deceive individuals 11 00:00:35.370 --> 00:00:39.480 into interacting with manipulated AI outputs. 12 00:00:39.480 --> 00:00:43.410 Interactivity may occur through interactive chatbots 13 00:00:43.410 --> 00:00:48.390 or social media platforms, and include deepfake videos, 14 00:00:48.390 --> 00:00:53.390 or audio that appear genuine, but are entirely fabricated. 15 00:00:53.490 --> 00:00:57.900 Let's learn more about AI enabled manipulation attacks 16 00:00:57.900 --> 00:01:01.290 to include social engineering and deepfakes 17 00:01:01.290 --> 00:01:05.430 delivered through digital media and interactivity. 18 00:01:05.430 --> 00:01:07.740 Then we will conduct a demo 19 00:01:07.740 --> 00:01:10.988 of an AI-generated voice deception. 20 00:01:10.988 --> 00:01:14.490 First, we have social engineering. 21 00:01:14.490 --> 00:01:18.240 AI manipulation in social engineering attacks 22 00:01:18.240 --> 00:01:21.930 uses AI to exploit human trust, 23 00:01:21.930 --> 00:01:26.280 creating deceptive situations that seem legitimate. 24 00:01:26.280 --> 00:01:30.300 For instance, an attacker might employ a chatbot 25 00:01:30.300 --> 00:01:33.000 that impersonates a known contact, 26 00:01:33.000 --> 00:01:35.998 using language and conversational patterns 27 00:01:35.998 --> 00:01:38.310 tailored to the victim. 28 00:01:38.310 --> 00:01:41.070 By establishing a sense of trust, 29 00:01:41.070 --> 00:01:44.430 the manipulating AI system can lead victims 30 00:01:44.430 --> 00:01:48.120 into revealing personal data, passwords, 31 00:01:48.120 --> 00:01:51.000 or even financial information. 32 00:01:51.000 --> 00:01:56.000 The scary part is that with advanced AI language models, 33 00:01:56.160 --> 00:01:59.160 these interactions feel natural and real, 34 00:01:59.160 --> 00:02:02.700 making social engineering attacks more persuasive 35 00:02:02.700 --> 00:02:05.190 and challenging to identify. 36 00:02:05.190 --> 00:02:08.670 AI manipulation social engineering attacks 37 00:02:08.670 --> 00:02:12.420 are increasingly common on social media platforms, 38 00:02:12.420 --> 00:02:16.350 where fake accounts can be created to impersonate friends, 39 00:02:16.350 --> 00:02:19.860 family members, or authority figures. 40 00:02:19.860 --> 00:02:24.030 In these cases, AI can analyze user data 41 00:02:24.030 --> 00:02:26.640 and craft personalized messages, 42 00:02:26.640 --> 00:02:30.030 enhancing the illusion of authenticity. 43 00:02:30.030 --> 00:02:34.710 Even automated phishing messages are evolving with AI, 44 00:02:34.710 --> 00:02:37.980 learning to adapt their language and content 45 00:02:37.980 --> 00:02:40.350 based on user behavior, 46 00:02:40.350 --> 00:02:44.640 making it easier for attackers to gain a victim's trust. 47 00:02:44.640 --> 00:02:48.330 So the use of AI in social engineering 48 00:02:48.330 --> 00:02:50.580 means individuals and companies 49 00:02:50.580 --> 00:02:55.580 need to exercise greater caution with online interactions, 50 00:02:55.860 --> 00:02:58.320 double checking authenticity, 51 00:02:58.320 --> 00:03:02.160 and being wary of unsolicited requests. 52 00:03:02.160 --> 00:03:06.420 Second, we have deepfakes delivered through digital media 53 00:03:06.420 --> 00:03:08.280 and interactivity. 54 00:03:08.280 --> 00:03:12.120 Deepfakes delivered through digital media and interactivity 55 00:03:12.120 --> 00:03:16.230 represent another form of AI manipulation attack 56 00:03:16.230 --> 00:03:18.840 that can cause serious harm. 57 00:03:18.840 --> 00:03:23.130 Deepfake technology generates hyperrealistic videos 58 00:03:23.130 --> 00:03:28.130 or audio files that mimic real people, often public figures, 59 00:03:28.890 --> 00:03:30.990 making it challenging to distinguish 60 00:03:30.990 --> 00:03:34.350 between what is real, and what is fake. 61 00:03:34.350 --> 00:03:39.150 For example, a deepfake video of a well-known executive 62 00:03:39.150 --> 00:03:42.882 could be fabricated to show them making damaging statements 63 00:03:42.882 --> 00:03:47.340 about a company, or endorsing a fraudulent scheme. 64 00:03:47.340 --> 00:03:50.850 Delivered through social media or streaming platforms, 65 00:03:50.850 --> 00:03:54.150 such a video could reach a wide audience, 66 00:03:54.150 --> 00:03:58.020 damaging reputations and spreading misinformation 67 00:03:58.020 --> 00:04:01.260 before it is detected as false. 68 00:04:01.260 --> 00:04:03.990 The interactive aspect of deepfakes 69 00:04:03.990 --> 00:04:06.750 is what makes them more deceptive. 70 00:04:06.750 --> 00:04:10.560 Deepfakes can now be used in live video calls, 71 00:04:10.560 --> 00:04:13.200 or digital conferencing platforms. 72 00:04:13.200 --> 00:04:16.920 Attackers may even use AI-generated voices 73 00:04:16.920 --> 00:04:20.910 and face-swapping technology in real time, 74 00:04:20.910 --> 00:04:23.820 making it appear as though the person is present 75 00:04:23.820 --> 00:04:26.670 and speaking directly to others. 76 00:04:26.670 --> 00:04:30.690 This real-time manipulation increases the risk 77 00:04:30.690 --> 00:04:33.180 of fraud and misinformation, 78 00:04:33.180 --> 00:04:36.060 especially in high stakes environments, 79 00:04:36.060 --> 00:04:40.440 like business meetings, or online news broadcasts. 80 00:04:40.440 --> 00:04:44.160 So as deepfakes become harder to detect, 81 00:04:44.160 --> 00:04:47.520 organizations need advanced verification methods 82 00:04:47.520 --> 00:04:50.940 and awareness programs to combat the spread 83 00:04:50.940 --> 00:04:54.270 of AI-driven misinformation. 84 00:04:54.270 --> 00:04:59.270 Now let's conduct a demonstration of an AI-generated voice. 85 00:04:59.550 --> 00:05:03.540 For this demonstration, I am using Descript, 86 00:05:03.540 --> 00:05:08.520 and I am going to create an AI speaker that is of my voice 87 00:05:08.520 --> 00:05:12.390 and then write some text for that speaker to say. 88 00:05:12.390 --> 00:05:16.980 Hopefully, in the end, it sounds a little bit like me. 89 00:05:16.980 --> 00:05:18.810 I'm here in AI speakers, 90 00:05:18.810 --> 00:05:21.930 and you can see that there's already one with my name on it. 91 00:05:21.930 --> 00:05:25.630 So I'm going to create a new AI speaker. 92 00:05:25.630 --> 00:05:30.630 I'm going to call this new AI Speaker Alt, or A-L-T Me. 93 00:05:31.651 --> 00:05:34.050 Let's go ahead and confirm. 94 00:05:34.050 --> 00:05:36.720 Now, the next thing I need to do is train 95 00:05:36.720 --> 00:05:39.210 and authorize this speaker. 96 00:05:39.210 --> 00:05:41.580 While I read this text here, 97 00:05:41.580 --> 00:05:44.250 I'm going to mute my microphone, 98 00:05:44.250 --> 00:05:48.840 so that I can clearly read this text into Descript. 99 00:05:48.840 --> 00:05:51.723 I'll go ahead and start my recording now. 100 00:06:10.140 --> 00:06:12.510 All right, I read the script. 101 00:06:12.510 --> 00:06:15.390 Now I need to authorize my speaker. 102 00:06:15.390 --> 00:06:17.583 So I'll click the Submit button. 103 00:06:18.990 --> 00:06:20.670 And it'll take a few minutes 104 00:06:20.670 --> 00:06:23.190 for the voice model to be created. 105 00:06:23.190 --> 00:06:25.829 You can see it's already started. 106 00:06:25.829 --> 00:06:29.760 In the meantime, I'm going to set up a new project 107 00:06:29.760 --> 00:06:32.250 in a private workspace. 108 00:06:32.250 --> 00:06:34.230 Here I am in the private workspace, 109 00:06:34.230 --> 00:06:37.710 and I'm going to select New Project. 110 00:06:37.710 --> 00:06:40.920 And this will be an audio project. 111 00:06:40.920 --> 00:06:44.910 Now that my audio project is up and running, 112 00:06:44.910 --> 00:06:47.880 I'm going to write some AI speech. 113 00:06:47.880 --> 00:06:51.780 It's not going to be Cedric talking, it'll be me. 114 00:06:51.780 --> 00:06:54.780 But let's go ahead and get this written first. 115 00:06:54.780 --> 00:06:57.540 What do I want the AI me to say? 116 00:06:57.540 --> 00:07:00.450 Well, let's just have me introduce myself. 117 00:07:00.450 --> 00:07:04.173 Hello, I am Jeremiah Minner. 118 00:07:07.080 --> 00:07:11.973 I am an instructor for Dion Training. 119 00:07:13.560 --> 00:07:16.080 All right, let's go ahead and add me 120 00:07:16.080 --> 00:07:18.813 or the alternate me as the speaker. 121 00:07:19.770 --> 00:07:23.673 And there it goes. It's generating the AI speech. 122 00:07:24.990 --> 00:07:26.763 This will take just a minute. 123 00:07:29.820 --> 00:07:31.710 Once the speech is generated, 124 00:07:31.710 --> 00:07:34.770 I'll just play it using the play key down here. 125 00:07:34.770 --> 00:07:37.050 But when it's generated, I will know, 126 00:07:37.050 --> 00:07:39.780 'cause I'll get a little green check mark that tells me that 127 00:07:39.780 --> 00:07:42.633 that AI speech has been completed. 128 00:07:45.780 --> 00:07:50.130 All right, AI speech completed and download is completed. 129 00:07:50.130 --> 00:07:53.670 All completed, perfect. Let's go ahead and play it. 130 00:07:53.670 --> 00:07:56.280 Hello, I am Jeremiah Minner. 131 00:07:56.280 --> 00:07:59.043 I am an instructor for Dion Training. 132 00:08:00.990 --> 00:08:02.010 There we go. 133 00:08:02.010 --> 00:08:05.490 This demonstrates how quickly an AI voice 134 00:08:05.490 --> 00:08:07.800 can be created and used. 135 00:08:07.800 --> 00:08:10.080 Though some of that generated speech 136 00:08:10.080 --> 00:08:12.210 sounded a bit mechanical, 137 00:08:12.210 --> 00:08:15.480 it's important to remember that this right here 138 00:08:15.480 --> 00:08:19.920 is as bad as AI generated voices will ever sound. 139 00:08:19.920 --> 00:08:21.360 From this point forward, 140 00:08:21.360 --> 00:08:23.850 they will only be getting better and better 141 00:08:23.850 --> 00:08:26.010 as technology advances, 142 00:08:26.010 --> 00:08:28.890 and attackers will only get better and better 143 00:08:28.890 --> 00:08:32.520 at implementing this type of deception. 144 00:08:32.520 --> 00:08:35.520 So remember, AI, 145 00:08:35.520 --> 00:08:39.330 or artificial intelligence manipulation attacks 146 00:08:39.330 --> 00:08:42.480 intentionally deceive AI systems 147 00:08:42.480 --> 00:08:45.900 to create harmful or misleading outcomes. 148 00:08:45.900 --> 00:08:49.080 These attacks often involve social engineering 149 00:08:49.080 --> 00:08:50.730 and deepfakes, 150 00:08:50.730 --> 00:08:55.560 both of which exploit human trust and digital platforms. 151 00:08:55.560 --> 00:08:58.470 Social engineering tactics use AI 152 00:08:58.470 --> 00:09:02.430 to create convincing, personalized interactions 153 00:09:02.430 --> 00:09:06.240 through chatbots or fake social media profiles, 154 00:09:06.240 --> 00:09:09.930 leading individuals to share sensitive information 155 00:09:09.930 --> 00:09:12.600 or perform specific actions. 156 00:09:12.600 --> 00:09:16.890 Deepfakes, on the other hand, use artificial intelligence 157 00:09:16.890 --> 00:09:21.060 to create realistic but false videos or audio, 158 00:09:21.060 --> 00:09:23.820 making it difficult for viewers to distinguish 159 00:09:23.820 --> 00:09:27.660 between genuine and fabricated content. 160 00:09:27.660 --> 00:09:31.110 As these attacks become more sophisticated, 161 00:09:31.110 --> 00:09:34.230 organizations need advanced security 162 00:09:34.230 --> 00:09:36.150 and verification methods 163 00:09:36.150 --> 00:09:41.150 to protect against AI-driven deception and misinformation.