1
00:00:04,889 --> 00:00:07,800
China curbs overseas travel for AI talent.

2
00:00:08,050 --> 00:00:10,430
Researchers at leading Chinese labs now need

3
00:00:10,430 --> 00:00:12,770
government approval for any overseas travel,

4
00:00:12,920 --> 00:00:16,719
closing the informal channels that once let ideas cross borders quickly.

5
00:00:17,010 --> 00:00:21,250
Domestic labs will accelerate fully sovereign training runs to make up the gap.

6
00:00:21,649 --> 00:00:24,649
US Frontier Labs respond by screening candidates more

7
00:00:24,649 --> 00:00:27,520
aggressively for recent ties to those same firms.

8
00:00:27,809 --> 00:00:32,709
Two parallel talent pools stop sharing benchmarks or techniques inside 18 months.

9
00:00:33,389 --> 00:00:36,450
Mastercard details trusted Agentic commerce vision.

10
00:00:36,860 --> 00:00:40,299
Every agent-driven purchase now requires an auditable chain

11
00:00:40,299 --> 00:00:42,889
back to user intent before the payment clears.

12
00:00:43,220 --> 00:00:45,819
Banks must rewrite fraud detection around machine

13
00:00:45,819 --> 00:00:48,729
generated sequences instead of individual clicks.

14
00:00:49,020 --> 00:00:53,049
Smaller FinTech players face a choice to adopt the new trust specs or

15
00:00:53,049 --> 00:00:55,580
watch merchants route volume elsewhere once

16
00:00:55,580 --> 00:00:58,209
large retailers standardize on the requirements.

17
00:00:58,540 --> 00:01:01,900
Early tests will expose whether the added security checks introduce

18
00:01:01,900 --> 00:01:05,440
latency that undercuts the speed agents promise at checkout.

19
00:01:05,980 --> 00:01:09,370
When Deep Seek makes V4 Pro price cut permanent,

20
00:01:09,739 --> 00:01:14,360
Deepeek's permanent reduction locks output tokens at 87 cents per million

21
00:01:14,500 --> 00:01:16,860
and ends the era of temporary discounts.

22
00:01:17,059 --> 00:01:20,739
Anthropic must now either match the rate across its public endpoints

23
00:01:20,739 --> 00:01:24,730
or shift entirely to enterprise deals that bundle compliance extras.

24
00:01:25,050 --> 00:01:27,900
Startups relying on cheap inference migrate first.

25
00:01:28,220 --> 00:01:31,339
Higher priced providers are left competing on safety features

26
00:01:31,339 --> 00:01:34,440
that few buyers prioritize when budgets tighten next year.

27
00:01:35,419 --> 00:01:38,940
Wait till US Census reports rising business AI adoption.

28
00:01:39,260 --> 00:01:44,319
Numbers stuck between 17 and 20% over six months show the low hanging fruit is gone.

29
00:01:44,739 --> 00:01:49,120
Growth now concentrates in firms large enough to absorb custom integration work.

30
00:01:49,339 --> 00:01:52,339
Vertical software vendors will have to cut per user AI

31
00:01:52,339 --> 00:01:56,169
fees sharply or watch smaller businesses stall out entirely.

32
00:01:56,419 --> 00:01:57,260
Without those cuts,

33
00:01:57,379 --> 00:02:00,580
the headline adoption rate stays flat through the end of next year.

34
00:02:01,370 --> 00:02:04,050
States enact new AI healthcare regulations.

35
00:02:04,330 --> 00:02:07,010
Hospitals now face a choice between slowing their AI

36
00:02:07,010 --> 00:02:10,210
pilots or building entire oversight teams from scratch.

37
00:02:10,529 --> 00:02:13,100
The new rules require every automated denial to

38
00:02:13,100 --> 00:02:15,529
carry a human sign off within 48 hours.

39
00:02:15,860 --> 00:02:18,460
That timeline alone will push smaller insurers toward

40
00:02:18,460 --> 00:02:21,039
legacy systems they abandoned two years ago.

41
00:02:21,259 --> 00:02:24,300
Expect prior authorization vendors to embed real-time

42
00:02:24,300 --> 00:02:27,250
logging that state auditors can query directly.

43
00:02:27,619 --> 00:02:30,059
Mid-sized plans in California and New York

44
00:02:30,059 --> 00:02:33,839
will likely freeze new deployments until Q4 2026

45
00:02:34,020 --> 00:02:36,229
while they retrofit compliance layers.

46
00:02:36,460 --> 00:02:38,779
Agentic AI breaks traditional security.

47
00:02:38,895 --> 00:02:39,544
Assumptions.

48
00:02:39,884 --> 00:02:43,134
Security teams will soon discover that their endpoint agents no

49
00:02:43,134 --> 00:02:46,634
longer see the real actor once autonomous workflows take over.

50
00:02:46,895 --> 00:02:50,164
Existing controls assume a human clicks through approvals,

51
00:02:50,214 --> 00:02:52,895
yet agents chain API calls across Slack,

52
00:02:53,095 --> 00:02:53,645
Salesforce,

53
00:02:53,735 --> 00:02:55,604
and finance systems in seconds.

54
00:02:56,014 --> 00:02:58,975
CIOs must rewrite data access policies around

55
00:02:58,975 --> 00:03:01,445
agent identities rather than user roles.

56
00:03:01,735 --> 00:03:03,975
The first firms to lose control will be those

57
00:03:03,975 --> 00:03:07,214
still relying on MFA prompts that agents can simulate.

58
00:03:07,610 --> 00:03:12,110
This forces CrowdStrike to ship agent-specific baselines by early next year.

59
00:03:12,410 --> 00:03:15,240
Clawed memory syncs across AI platforms.

60
00:03:15,410 --> 00:03:19,169
The move quietly hands anthropic control over continuity data

61
00:03:19,169 --> 00:03:21,729
that every competing agent will now need to match.

62
00:03:22,089 --> 00:03:26,229
Users can keep one persistent memory across chat GPT and Gemini

63
00:03:26,449 --> 00:03:28,449
without rebuilding context each time.

64
00:03:28,809 --> 00:03:31,339
That convenience raises the switching cost for

65
00:03:31,339 --> 00:03:33,660
anyone already invested in Claude's profile.

66
00:03:33,919 --> 00:03:39,199
OpenAI therefore faces pressure to expose equivalent memory APIs within 6 months

67
00:03:39,380 --> 00:03:43,770
or see enterprise teams standardize on anthropic for multi-model workflows.

68
00:03:44,100 --> 00:03:46,580
Smaller startups lose ground fastest without the

69
00:03:46,580 --> 00:03:48,979
scale to negotiate similar sync deals.

70
00:03:49,330 --> 00:03:51,839
AI voice cloning scams hit families.

71
00:03:52,210 --> 00:03:54,490
Families hit hardest will be those who still trust

72
00:03:54,490 --> 00:03:56,869
a familiar voice over a simple callback number.

73
00:03:57,479 --> 00:04:02,070
Scammers already route calls through VOIP services that strip carrier metadata,

74
00:04:02,289 --> 00:04:04,369
making trace back nearly impossible.

75
00:04:04,589 --> 00:04:06,460
Expect banks to mandate video verification.

76
00:04:07,100 --> 00:04:10,610
for transfers above $500 within the next two quarters.

77
00:04:10,940 --> 00:04:14,690
That change alone will add friction to legitimate family support payments.

78
00:04:15,059 --> 00:04:19,858
Telecom providers must deploy voice watermarking on their networks by mid-2026

79
00:04:19,858 --> 00:04:23,739
or absorb rising fraud losses passed through from insurers and banks.

80
00:04:24,670 --> 00:04:27,739
Nursing education adopts AI and VR training.

81
00:04:28,109 --> 00:04:30,690
State licensing boards now face the real test.

82
00:04:30,869 --> 00:04:34,670
Will they accept VR procedure logs as equivalent to bedside hours?

83
00:04:34,910 --> 00:04:35,670
If they do,

84
00:04:35,869 --> 00:04:40,420
the schools still using mannequins will see enrollment drop by 50% within 2 years.

85
00:04:40,709 --> 00:04:43,470
Hospitals are already signaling preference for graduates

86
00:04:43,470 --> 00:04:46,500
who logged 200+ simulated critical events.

87
00:04:46,829 --> 00:04:49,149
That preference turns the adoption question into

88
00:04:49,149 --> 00:04:51,570
a survival issue for smaller programs.

89
00:04:52,160 --> 00:04:55,279
Baidu pushes Chinese AI independence with Ernie.

90
00:04:55,519 --> 00:04:57,799
The bigger move sits in the chip supply chain,

91
00:04:57,959 --> 00:04:59,309
not the model releases.

92
00:04:59,600 --> 00:05:03,160
Baidu's domestic silicon push means foreign GPU vendors

93
00:05:03,160 --> 00:05:06,160
lose their easiest path into Chinese training clusters.

94
00:05:06,399 --> 00:05:09,959
Expect Microsoft and Nvidia to accelerate joint ventures with

95
00:05:09,959 --> 00:05:13,100
local foundries before the next round of restrictions lands.

96
00:05:13,320 --> 00:05:13,880
Otherwise,

97
00:05:13,959 --> 00:05:16,279
their inference margins in the region collapse.

98
00:05:17,239 --> 00:05:20,059
Agenic AI threat modeling gains urgency.

99
00:05:20,320 --> 00:05:24,119
Enterprise security teams are already rewriting playbooks around agents

100
00:05:24,119 --> 00:05:27,309
that can execute multi-step actions without human review.

101
00:05:27,600 --> 00:05:30,880
The first companies to ship dedicated agent threat models will

102
00:05:30,880 --> 00:05:33,500
set the procurement checklist for the rest of the sector.

103
00:05:33,760 --> 00:05:36,119
Everyone else ends up retrofitting after the

104
00:05:36,119 --> 00:05:38,880
first autonomous incident hits a production SLA.

105
00:05:39,760 --> 00:05:42,709
Banks shift to AI specialists from traditional roles.

106
00:05:42,959 --> 00:05:44,980
The quiet constraint is data access,

107
00:05:45,239 --> 00:05:46,109
not headcount.

108
00:05:46,440 --> 00:05:49,660
Banks cutting relationship manager roles still need clean

109
00:05:49,660 --> 00:05:52,790
labeled transaction histories to train the new models.

110
00:05:53,070 --> 00:05:56,000
That requirement hands disproportionate power to the handful

111
00:05:56,000 --> 00:05:59,070
of FinTechs already sitting on consented data sets.

112
00:05:59,440 --> 00:06:03,640
Larger institutions will have to buy or partner rather than build from scratch.