1
00:00:01,679 --> 00:00:06,000
Speaker 1: Espionage has also been now influenced by AI. But how

2
00:00:06,040 --> 00:00:07,799
are they using it. Let's take a look and see

3
00:00:07,839 --> 00:00:10,199
what's going on. Something. The things you can see AI

4
00:00:10,320 --> 00:00:14,000
is helping with is scaling and speeding analysis. Like ocent sigent,

5
00:00:14,480 --> 00:00:18,519
AI rapidly processes huge volumes of open source data that

6
00:00:18,600 --> 00:00:24,079
include social media, foreign news, darknet also signals, intelligence intercepts,

7
00:00:24,079 --> 00:00:28,039
and traffic and imagery immine from satellites and drones that

8
00:00:28,120 --> 00:00:31,440
lets analysts find new leads, track networks and detect changes

9
00:00:31,519 --> 00:00:35,399
far faster than ever. Also, another thing that's been incredible

10
00:00:35,479 --> 00:00:39,200
has been large language and translation models that speed translation

11
00:00:39,280 --> 00:00:42,520
of captured documents, voice intercepts, and help triage which material

12
00:00:42,600 --> 00:00:46,439
is worth deeper human review. So this is probably going

13
00:00:46,479 --> 00:00:48,479
to affect a lot of the jobs that are used

14
00:00:48,479 --> 00:00:53,840
for translation, link analysis, entity resolution, and network detection, where

15
00:00:53,920 --> 00:00:57,359
machine learning uncovers relationships and messy data sets who met, whom,

16
00:00:57,399 --> 00:01:01,039
money flows, and device associations. This is of course central

17
00:01:01,039 --> 00:01:05,640
to counter espionage and tracking clandestine networks. AI also improves

18
00:01:05,680 --> 00:01:10,159
object to person detection and imagery, automates change detection, and

19
00:01:10,359 --> 00:01:13,519
enables persistent monitoring without needing human eyes in every frame.

20
00:01:14,840 --> 00:01:17,879
And finally, AI also assists cyber operators to craft more

21
00:01:17,879 --> 00:01:22,359
convincing phishing, automate reconnaissance, and speed malware development and opphification.

22
00:01:23,159 --> 00:01:25,799
But of course there's a downside. Some of the things

23
00:01:25,799 --> 00:01:28,840
that could hurt intelligence include deep fakes, where audio video

24
00:01:28,879 --> 00:01:32,239
imagery fakes can be weaponized to impersonate officials, fabricate events,

25
00:01:32,359 --> 00:01:36,599
or discredit targets. Detection is an arms race at this point.

26
00:01:37,359 --> 00:01:40,519
Another one is AI generated content. Automated campaigns make it

27
00:01:40,560 --> 00:01:44,920
hard to prove who acted, complicating diplomatic or legal responses.

28
00:01:46,120 --> 00:01:48,799
AI lowers a skill expense barrier for state and non

29
00:01:48,799 --> 00:01:53,239
state actors to run sophisticated cybro and disinformation campaigns. The

30
00:01:53,280 --> 00:01:56,760
same AI tools help both defenders and attackers. Recent reporting

31
00:01:56,799 --> 00:02:02,040
shows multiple adversary states rapidly adopting generative AI for these

32
00:02:02,040 --> 00:02:09,000
particular operations. Over Reliance and bias could also affect those

33
00:02:09,000 --> 00:02:12,120
who are trained on Bias or partial data can mislead analysts,

34
00:02:12,759 --> 00:02:16,120
and finally, adversaries can try to contaminate training data craft

35
00:02:16,159 --> 00:02:21,919
inputs that trick the models. CIA and other agencies operationalizing

36
00:02:21,960 --> 00:02:25,960
AI are saying that the AI has moved beyond pilots

37
00:02:25,960 --> 00:02:29,000
and it's scaling AI into many mission areas now analytics,

38
00:02:29,080 --> 00:02:33,479
cyber finance personnel, often via cloud and vendor partnerships. US

39
00:02:33,520 --> 00:02:37,960
cyber defense organizations and SAY personnel sited using mL to

40
00:02:38,039 --> 00:02:42,599
detect hacking behavior at AI assistant tools to trace adversary activity.

41
00:02:43,319 --> 00:02:46,960
US agencies and homeland security organizations have also published warnings

42
00:02:47,000 --> 00:02:52,879
and guidance about generative AI threats like deep figs so

43
00:02:53,000 --> 00:02:57,159
operational implications for espionage well. AI generated identities and media

44
00:02:57,199 --> 00:03:00,960
increase the burden on human sources and authentication workflows. AI

45
00:03:01,120 --> 00:03:04,479
enables quicker queuing of operations, but faster tempo increases risk

46
00:03:04,560 --> 00:03:09,840
of acting incorrectly. Bottom line, there's pros and cons, and

47
00:03:09,879 --> 00:03:14,280
it could be transformational for modern espionage since it multiplies

48
00:03:14,319 --> 00:03:16,960
what analysts can see him do at a scale and

49
00:03:17,000 --> 00:03:19,400
speed never seen before, but as well, at the same

50
00:03:19,439 --> 00:03:24,400
time multiplies threats. The balance between benefit and harm depends

51
00:03:24,439 --> 00:03:26,800
on a lot of factors, and agencies that pair with

52
00:03:26,879 --> 00:03:30,039
AI with strong validation and governance get the gains while

53
00:03:30,080 --> 00:03:33,120
limitating while limiting strategic risk.