WEBVTT

1
00:00:00.080 --> 00:00:04.440
<v Speaker 1>AI Daily Briefing. I'm at Sharp, thanks for joining me today.

2
00:00:04.559 --> 00:00:08.560
<v Speaker 1>Pentagon's real time AI warfare A Censure just won an

3
00:00:08.599 --> 00:00:11.759
<v Speaker 1>eight hundred and twenty one million dollar Pentagon contract to

4
00:00:11.800 --> 00:00:14.880
<v Speaker 1>build the War Data Platform. And that's the clearest signal

5
00:00:14.960 --> 00:00:18.440
<v Speaker 1>yet that the US military isn't experimenting with AI anymore,

6
00:00:18.800 --> 00:00:23.000
<v Speaker 1>it's operationalizing it. The War Data Platform connects more than

7
00:00:23.039 --> 00:00:26.280
<v Speaker 1>fifteen hundred data sources and is designed to feed AI

8
00:00:26.359 --> 00:00:30.480
<v Speaker 1>systems making real time battlefield decisions. This thing started as

9
00:00:30.519 --> 00:00:34.359
<v Speaker 1>a financial audit tool back in twenty eighteen. It's now

10
00:00:34.399 --> 00:00:38.320
<v Speaker 1>the connective tissue for AI powered warfare. That's a significant

11
00:00:38.320 --> 00:00:41.520
<v Speaker 1>shift and a short amount of time. Here's what matters

12
00:00:41.560 --> 00:00:45.079
<v Speaker 1>most about this contract. There are no publicly stated safeguards

13
00:00:45.079 --> 00:00:49.679
<v Speaker 1>against AI systems acting outside intended parameters. The platform is

14
00:00:49.719 --> 00:00:53.200
<v Speaker 1>moving from experimental to operational without a clear public answer

15
00:00:53.200 --> 00:00:56.439
<v Speaker 1>on what keeps autonomous systems in check. That's the open

16
00:00:56.479 --> 00:01:01.840
<v Speaker 1>question the contract doesn't answer on the security. Microsoft released

17
00:01:01.840 --> 00:01:05.959
<v Speaker 1>a specialized AI model called MAI dash cyber Dash one

18
00:01:06.079 --> 00:01:10.400
<v Speaker 1>dash Flash, trained on decades of vulnerability data. It's scored

19
00:01:10.480 --> 00:01:14.840
<v Speaker 1>ninety six percent on the cyber Gym security benchmark. That's

20
00:01:14.920 --> 00:01:18.400
<v Speaker 1>twelve points ahead of anthropics mythos model on the same test.

21
00:01:19.120 --> 00:01:22.159
<v Speaker 1>The important distinction here is what kind of AI is

22
00:01:22.159 --> 00:01:25.879
<v Speaker 1>winning the security race. MAI cyber one Flash isn't a

23
00:01:25.920 --> 00:01:31.519
<v Speaker 1>general purpose model doing security tasks. Its domain specific purpose

24
00:01:31.519 --> 00:01:36.640
<v Speaker 1>built and outperforming broader competitors on specialized benchmarks. The signal

25
00:01:36.719 --> 00:01:40.400
<v Speaker 1>is that in high stakes enterprise verticals, general intelligence may

26
00:01:40.439 --> 00:01:44.480
<v Speaker 1>not be enough. Vendors may need proprietary models built for

27
00:01:44.519 --> 00:01:49.040
<v Speaker 1>the domain while the pentagon scales up and vendors specialize.

28
00:01:49.400 --> 00:01:53.319
<v Speaker 1>California is building a regulatory pile up. The state is

29
00:01:53.359 --> 00:01:58.159
<v Speaker 1>moving forward with multiple AI bills simultaneously covering content disclosure,

30
00:01:58.400 --> 00:02:02.799
<v Speaker 1>suicide detection tools, and labour protections for workers displaced by automation.

31
00:02:03.599 --> 00:02:06.560
<v Speaker 1>Here's the thing. The problem isn't the intent behind any

32
00:02:06.640 --> 00:02:11.080
<v Speaker 1>single bill, it's the fragmentation. States are passing roughly eighteen

33
00:02:11.159 --> 00:02:15.680
<v Speaker 1>hundred AI bills annually. Congress hasn't established a federal baseline.

34
00:02:16.199 --> 00:02:20.080
<v Speaker 1>That means companies operating nationally on navigating a patchwork that

35
00:02:20.240 --> 00:02:23.919
<v Speaker 1>changes by state, by sector, and by year. Some of

36
00:02:23.960 --> 00:02:28.199
<v Speaker 1>the California proposals carry strict liability standards that could trigger

37
00:02:28.240 --> 00:02:33.680
<v Speaker 1>significant litigation exposure for developers industry groups are pushing back hard,

38
00:02:33.759 --> 00:02:37.719
<v Speaker 1>and that fight isn't close to settled. TSMC raised its

39
00:02:37.759 --> 00:02:41.560
<v Speaker 1>twenty twenty six capital expenditure guidance to between sixty and

40
00:02:41.639 --> 00:02:45.400
<v Speaker 1>sixty four billion dollars, and the market didn't reward that confidence.

41
00:02:46.360 --> 00:02:49.120
<v Speaker 1>The stock has fallen seventeen percent from its record high

42
00:02:49.120 --> 00:02:51.800
<v Speaker 1>of four hundred and eighty dollars and is now testing

43
00:02:51.840 --> 00:02:55.400
<v Speaker 1>the three hundred ninety dollars support level. The investor concern

44
00:02:55.560 --> 00:02:59.879
<v Speaker 1>is structural AI infrastructure demand has driven an aggressive capex

45
00:03:00.039 --> 00:03:04.240
<v Speaker 1>cycle across the semiconductor industry. The question is whether hyperscalar

46
00:03:04.319 --> 00:03:07.360
<v Speaker 1>spending holds long enough to justify returns on that investment.

47
00:03:08.280 --> 00:03:11.240
<v Speaker 1>If the big cloud players pull back even slightly, the

48
00:03:11.319 --> 00:03:15.080
<v Speaker 1>ripple through the supply chain could be sharp. Meanwhile, a

49
00:03:15.080 --> 00:03:19.199
<v Speaker 1>state backed Chinese manufacturer is entering production of immersion DUV

50
00:03:19.360 --> 00:03:22.840
<v Speaker 1>lithography systems, the kind of chip making equipment China has

51
00:03:22.840 --> 00:03:27.120
<v Speaker 1>been cut off from through US export controls. Deliveries are

52
00:03:27.120 --> 00:03:31.639
<v Speaker 1>expected before year end, though production volumes and quality remain unconfirmed.

53
00:03:32.840 --> 00:03:37.400
<v Speaker 1>Consider this. If the timeline holds, this represents a meaningful

54
00:03:37.400 --> 00:03:41.479
<v Speaker 1>erosion of the hardware advantage. Western export controls were designed

55
00:03:41.479 --> 00:03:46.159
<v Speaker 1>to maintain the ambiguity around volumes matters, but the direction

56
00:03:46.240 --> 00:03:51.000
<v Speaker 1>of travel is clear. One more development worth tracking, SAP

57
00:03:51.240 --> 00:03:55.280
<v Speaker 1>released an AI agent hub with automated discovery and compliance

58
00:03:55.360 --> 00:04:00.840
<v Speaker 1>verification across enterprise platforms. Governance tools are becoming a product category,

59
00:04:01.280 --> 00:04:04.800
<v Speaker 1>not just a policy conversation. The three watch points heading

60
00:04:04.800 --> 00:04:08.639
<v Speaker 1>into tomorrow are the Pentagon platform safeguard question, where the

61
00:04:08.680 --> 00:04:13.680
<v Speaker 1>California's liability provisions survive industry pushback, and whether China's lithography

62
00:04:13.680 --> 00:04:18.319
<v Speaker 1>deliveries materialize on schedule. Those three together tell you a

63
00:04:18.360 --> 00:04:23.079
<v Speaker 1>lot about where AI power is actually concentrating. Thanks for listening.

64
00:04:23.519 --> 00:04:26.959
<v Speaker 1>This podcast was built using AI technology, a Yes We

65
00:04:27.079 --> 00:04:27.519
<v Speaker 1>production