WEBVTT

1
00:00:00.080 --> 00:00:04.799
<v Speaker 1>AI daily breathing. I'm d sharp, thanks for joining me today.

2
00:00:05.000 --> 00:00:09.519
<v Speaker 1>The standard's body trap who controls AI guardrails. The most

3
00:00:09.560 --> 00:00:12.759
<v Speaker 1>significant proposal in AI governance right now isn't coming from

4
00:00:12.800 --> 00:00:16.120
<v Speaker 1>a regulator. It's coming from the CEO of Google deep Mind.

5
00:00:16.679 --> 00:00:19.960
<v Speaker 1>Demas Hassebis published a framework this week calling for a

6
00:00:20.079 --> 00:00:23.920
<v Speaker 1>US South led federal private standards body to test frontier

7
00:00:23.960 --> 00:00:27.320
<v Speaker 1>AI models before they're released. The model he's pointing to

8
00:00:27.440 --> 00:00:32.560
<v Speaker 1>is FINRA, the financial industry self regulatory organization. Start voluntary,

9
00:00:32.840 --> 00:00:37.039
<v Speaker 1>prove it works, then make it mandatory. The pitch sounds reasonable,

10
00:00:37.479 --> 00:00:42.880
<v Speaker 1>the implications are considerably more complicated. Here's what matters about

11
00:00:42.880 --> 00:00:46.119
<v Speaker 1>who's making this proposal. Hassabis runs one of the two

12
00:00:46.240 --> 00:00:49.640
<v Speaker 1>or three most capable AI labs on Earth. A pre

13
00:00:49.759 --> 00:00:53.759
<v Speaker 1>release testing regime built around a threshold called frontier model

14
00:00:54.079 --> 00:00:58.079
<v Speaker 1>would almost certainly cover Google deep minds competitors faster than

15
00:00:58.119 --> 00:01:02.000
<v Speaker 1>it covers Google deep Mind. That's not a conspiracy, that's

16
00:01:02.119 --> 00:01:05.319
<v Speaker 1>just how standards bodies work. The entity with the most

17
00:01:05.400 --> 00:01:09.799
<v Speaker 1>influence over the definition controls the playing field. The enforcement

18
00:01:09.920 --> 00:01:13.400
<v Speaker 1>question is also unresolved. This proposed body would have US

19
00:01:13.480 --> 00:01:17.480
<v Speaker 1>South jurisdiction. It's unclear how it binds labs operating outside

20
00:01:17.560 --> 00:01:21.599
<v Speaker 1>US South territory, how the frontier threshold gets defined and updated,

21
00:01:21.959 --> 00:01:24.519
<v Speaker 1>and where the voluntary compliance is viable in a market

22
00:01:24.519 --> 00:01:28.599
<v Speaker 1>where speed to deployment is a competitive advantage. Directly connected

23
00:01:28.640 --> 00:01:31.239
<v Speaker 1>to that governance question is a separate development from the

24
00:01:31.280 --> 00:01:35.159
<v Speaker 1>State Department, which officially raised serious concerns about US South

25
00:01:35.200 --> 00:01:40.480
<v Speaker 1>companies adopting Chinese AI models, specifically Deepseek and z dot Ai.

26
00:01:41.040 --> 00:01:43.400
<v Speaker 1>This is the first high level public warning on that

27
00:01:43.439 --> 00:01:48.159
<v Speaker 1>specific issue. Prior coverage focused on capability comparisons. The framing

28
00:01:48.200 --> 00:01:51.159
<v Speaker 1>has now shifted to policy threat. The key distinction is

29
00:01:51.159 --> 00:01:55.319
<v Speaker 1>this Deep Seek's cost advantage is real and measurable, policy

30
00:01:55.359 --> 00:01:58.400
<v Speaker 1>pressure is not. If the price differential is large enough,

31
00:01:58.560 --> 00:02:02.280
<v Speaker 1>security concerns tend to loose use and procurement decisions. That's

32
00:02:02.319 --> 00:02:05.280
<v Speaker 1>the actual risk the State Department is flagging, and it's

33
00:02:05.319 --> 00:02:08.560
<v Speaker 1>one that export controls alone are unlikely to solve cleanly.

34
00:02:09.080 --> 00:02:11.759
<v Speaker 1>The signal here is that the US self governance debate

35
00:02:11.879 --> 00:02:15.879
<v Speaker 1>isn't only about regulating domestic AI development. It's about who

36
00:02:15.919 --> 00:02:19.159
<v Speaker 1>controls the standards that determine which AI is safe to

37
00:02:19.280 --> 00:02:23.000
<v Speaker 1>use at all. Here's the thing Hassebi's also published what

38
00:02:23.039 --> 00:02:27.280
<v Speaker 1>he's calling a manifesto, framing agi arrival as only years away,

39
00:02:27.479 --> 00:02:30.800
<v Speaker 1>not decades. He describes the impact as ten times the

40
00:02:30.840 --> 00:02:35.319
<v Speaker 1>Industrial Revolution at ten times the speed he flagged cybersecurity

41
00:02:35.319 --> 00:02:40.840
<v Speaker 1>and biological risks. Specifically, the claim deserves calibration. Altman Ammaday

42
00:02:40.919 --> 00:02:44.080
<v Speaker 1>and others have made comparable statements. None of them attach

43
00:02:44.159 --> 00:02:48.919
<v Speaker 1>precise probabilities or specific timelines. What's different here is the

44
00:02:49.000 --> 00:02:52.520
<v Speaker 1>institutional weight behind the framing. When a Deep Mind CEO

45
00:02:52.639 --> 00:02:57.360
<v Speaker 1>says AGI's imminent and transformative, it justifies safety research funding,

46
00:02:57.840 --> 00:03:01.960
<v Speaker 1>sovereign compute investment, and yes X pre release testing regimes.

47
00:03:02.360 --> 00:03:05.639
<v Speaker 1>The urgency narrative and the governance proposal are not separate.

48
00:03:06.080 --> 00:03:11.599
<v Speaker 1>They're reinforcing the capital flows today. Make that reinforcement concrete. Helsing,

49
00:03:11.840 --> 00:03:15.719
<v Speaker 1>a European defense AI startup, raised one point eight billion

50
00:03:15.759 --> 00:03:19.599
<v Speaker 1>dollars and is building a drone manufacturing facility in West Virginia,

51
00:03:19.759 --> 00:03:23.240
<v Speaker 1>producing more than two thousand HX two strike drones monthly.

52
00:03:23.840 --> 00:03:27.280
<v Speaker 1>The company is now valued at eighteen billion dollars. Its

53
00:03:27.319 --> 00:03:30.639
<v Speaker 1>drones are already deployed in Ukraine. That's not a bet

54
00:03:30.680 --> 00:03:33.919
<v Speaker 1>on a future conflict. That's production scale commitment to an

55
00:03:33.919 --> 00:03:37.840
<v Speaker 1>active one consider this. Quantum Systems followed with a one

56
00:03:37.840 --> 00:03:41.080
<v Speaker 1>point two billion dollar raise valued at eight billion dollars

57
00:03:41.360 --> 00:03:45.520
<v Speaker 1>focused on multi domain autonomous systems. Combined with other defense

58
00:03:45.560 --> 00:03:49.039
<v Speaker 1>and aerospace deals, roughly seventy percent of today's three billion

59
00:03:49.080 --> 00:03:53.400
<v Speaker 1>dollars in total funding landed in geopolitically aligned categories. The

60
00:03:53.479 --> 00:03:56.879
<v Speaker 1>valuation meth on these deals is aggressive eighteen billion for

61
00:03:56.960 --> 00:04:00.159
<v Speaker 1>helsing an eight billion for quantum systems in play by

62
00:04:00.240 --> 00:04:05.199
<v Speaker 1>revenue multiples that require sustained conflict, sustained government contracts, or both.

63
00:04:05.759 --> 00:04:10.479
<v Speaker 1>If either condition softens, those numbers get tested hard outside defense.

64
00:04:10.639 --> 00:04:14.000
<v Speaker 1>Two other rounds are worth tracking. Picks Verse, a Singapore

65
00:04:14.039 --> 00:04:17.560
<v Speaker 1>based video generation startup backed by Ali Barber, closed the

66
00:04:17.560 --> 00:04:20.759
<v Speaker 1>four hundred and thirty nine million dollars Series C extension

67
00:04:20.800 --> 00:04:24.600
<v Speaker 1>at a valuation above two billion dollars, and CHI Discovery

68
00:04:24.759 --> 00:04:27.759
<v Speaker 1>raised four hundred million dollars for AI driven drug design,

69
00:04:28.120 --> 00:04:32.160
<v Speaker 1>valued at approximately three point eight billion dollars. Both rounds

70
00:04:32.240 --> 00:04:35.360
<v Speaker 1>confirm that mega round appetite in applied AI hasn't slowed,

71
00:04:35.759 --> 00:04:39.160
<v Speaker 1>but the gravitational center of capital today is clearly defense

72
00:04:39.199 --> 00:04:42.879
<v Speaker 1>and sovereignty infrastructure. The practical test for everything raised today

73
00:04:43.160 --> 00:04:46.439
<v Speaker 1>is where the standards become protectionist. A US South led

74
00:04:46.480 --> 00:04:49.879
<v Speaker 1>pre release testing body applied selectively could function as a

75
00:04:49.879 --> 00:04:53.240
<v Speaker 1>competitive mode rather than a safety mechanism. The difference between

76
00:04:53.279 --> 00:04:56.720
<v Speaker 1>those two outcomes depends entirely on how the frontier threshold

77
00:04:56.759 --> 00:05:00.519
<v Speaker 1>is defined and who defines it. Watch the Hassabis proposal

78
00:05:00.600 --> 00:05:04.079
<v Speaker 1>for specifics on that threshold. Watch the State Department's next

79
00:05:04.079 --> 00:05:06.959
<v Speaker 1>move on Chinese model adoption for whether this stays at

80
00:05:06.959 --> 00:05:10.120
<v Speaker 1>the level of concern or becomes enforceable policy. And watch

81
00:05:10.160 --> 00:05:12.680
<v Speaker 1>whether the defense tech funding pace holds when the next

82
00:05:12.759 --> 00:05:16.399
<v Speaker 1>quarterly numbers come in. The governance race and the capital

83
00:05:16.519 --> 00:05:19.560
<v Speaker 1>race are running on the same track right now, that's

84
00:05:19.560 --> 00:05:22.399
<v Speaker 1>the thing to keep in focus. Thanks for listening. This

85
00:05:22.519 --> 00:05:26.079
<v Speaker 1>podcast was built using AI technology, a yes We production