WEBVTT

1
00:00:10.000 --> 00:00:24.677
Welcome to the Global Advisors A.I Voices podcast. I'm Maya.
Today is Thursday, the nineteenth of November 2026.
This is your morning catch-up on A.I. Please like and follow the show for your daily A.I update.

2
00:00:26.726 --> 00:00:48.806
The A.I Daily Brief: Artificial Intelligence News reports that enterprise A.I usage jumped by 36 percent over the past month, while spending only increased by 2 percent. This divergence highlights a shift in the enterprise A.I landscape that favors the buyer, as usage explodes even while overall spending remains relatively flat.

3
00:00:49.156 --> 00:01:15.055
The driver behind this gap is falling costs. The average price per million tokens dropped by 21 percent over the last month. This means companies are getting significantly more computational work done for nearly the same budget. The A.I Daily Brief argues that this isn't just about companies switching to cheaper Chinese models; rather, vicious competition among major providers is forcing price decreases across the board.

4
00:01:15.405 --> 00:01:35.885
This dynamic gives enterprises more leverage than they might realize. The A.I Daily Brief suggests that companies are not fully exerting this power in negotiations. With spend highly concentrated among a small number of large buyers, these key customers are in a strong position to demand better terms as the market matures.

5
00:01:38.284 --> 00:02:01.431
Microsoft is intensifying the competition for developer mindshare. At its Ignite conference, the company announced seven new models under its MAI brand. The headline claim is for MAI Code 2, which Microsoft says reaches 96 percent of the performance of Anthropic's Claude Opus 5.5 across coding benchmarks, but at just 5 percent of the cost.

6
00:02:01.781 --> 00:02:25.034
It is important to note that these performance and pricing comparisons come from Microsoft's internal evaluation harness. They have not been independently reproduced or verified by external leaderboards. However, in a blind frontend test, developers reportedly picked MAI Code 2 over OpenAI's GPT-6 Sol, suggesting a strong user preference for the new tool.

7
00:02:25.384 --> 00:02:49.832
The full lineup includes specialized models like MAI Cyber 2 Flash and MAI Image 2.7. Microsoft is also introducing MAI Licensed Weights, a move that allows organizations to run customized versions of these models within their own Azure environments. This strategy aims to undercut competitors on both price and deployment flexibility.

8
00:02:52.231 --> 00:03:11.175
Satya Nadella, the C E O of Microsoft, is warning that the rise of A.I agents will fundamentally reshape the economics of the internet. Speaking at The New York Times DealBook Summit, he argued that the current model of the 'human web,' which relies heavily on advertising, is facing disruption.

9
00:03:11.525 --> 00:03:30.597
Nadella predicts that companies deeply invested in advertising-driven business models will have to scramble to adapt. He cited Shopify as a potential example of a platform that could gain an advantage by allowing A.I agents to interact directly with its services, bypassing traditional traffic metrics.

10
00:03:30.947 --> 00:03:51.854
This forecast suggests a shift where value accrues to platforms that facilitate agent-to-agent transactions rather than those that monetize human attention. For businesses relying on ad revenue, this represents a significant strategic risk that may force a rethinking of how they compete for customers in an agent-driven economy.

11
00:03:54.253 --> 00:04:18.509
Palmer Luckey, co-founder of Anduril, is issuing a stark warning about the state of cybersecurity. In an interview reported by TechCrunch, Luckey stated that while everything can be hacked for now, the core issue is a mismatch in speed. Attackers are using new A.I technology to move quickly, while defenders are stuck with old, outdated systems that cannot adopt new tools as fast as they are developed.

12
00:04:18.859 --> 00:04:37.761
Luckey predicts that in the long run, advantages will accrue to the side that is most effectively using A.I for defense. He argues that the current pace of technological change is leaving many systems vulnerable to a new wave of A.I-enabled attacks that traditional defenses simply cannot handle.

13
00:04:38.111 --> 00:04:52.361
Palmer Luckey concluded that people must pick up the pace, stating, 'It's time to start running.' He emphasized that organizations need to accelerate their adoption of modern security tools to keep up with the evolving threat landscape.

14
00:04:54.760 --> 00:05:20.872
Google Threat Intelligence Group has identified what it calls the first fully autonomous A.I hacking tool, named 'exploitagent'. The tool was discovered on an exposed staging server and integrates with several large language models to automate the entire attack process. Google reports that threat actors used this framework to fully automate vulnerability discovery and exploitation without human intervention for the first time.

15
00:05:21.222 --> 00:05:44.028
The mechanism is straightforward but dangerous. The A.I tool monitors software patches, compares code before and after fixes to identify vulnerabilities, and then automatically drafts and runs exploits. Google stated in its A.I Threat Tracker report that 'the era of autonomous attacks has arrived,' signaling a major shift in how cyber threats are generated.

16
00:05:44.378 --> 00:06:11.450
This automation is contributing to a rise in zero-day exploits, which jumped to 22 in August 2026. However, Google notes that current autonomous tools still have weaknesses, such as leaving recognizable traces and generating failed exploits. They also observed open-source A.I attack tools like CAIRN and PentAGI being used in real intrusions, indicating that this technology is becoming more accessible.

17
00:06:13.849 --> 00:06:34.969
Ryan Greenblatt, appearing on The MAD Podcast with Matt Turck, argues that a proposed deal requiring 'total research transparency' in A.I development would undermine one of the largest moats for frontier A.I companies. He suggests that such a mandate would force companies like OpenAI and Anthropic to compete on different axes.

18
00:06:35.319 --> 00:06:55.692
Under this transparency regime, Greenblatt posits that these companies would have to compete on user experience, customization, reliability, and safety rather than raw capabilities. He suggests this shift would greatly reduce their valuations, as the competitive advantage of proprietary research secrets would be erased.

19
00:06:56.042 --> 00:07:21.877
This perspective contrasts with current safety frameworks. For instance, OpenAI has established a framework requiring safety reviews before frontier models move from research to internal deployment and external release, stating they will pause training or withhold models if evidence falls short. Greenblatt's argument highlights the tension between regulatory transparency goals and the commercial incentives of A.I labs.

20
00:07:24.276 --> 00:07:44.436
Google is expanding the capabilities of its Gemini app, making its 'Tasks' feature free to hundreds of millions of users in six countries. Powered by the Gemini 4 Hydrogen model, this update allows the assistant to perform actions beyond simple chat, such as calling businesses, booking appointments, and completing purchases.

21
00:07:44.786 --> 00:08:06.055
Users can switch from Chat to Task mode, enabling the assistant to contact businesses and schedule appointments. For purchases, the feature uses Google's Agent Payments Protocol with digital tokens. Users can set spending limits, and transactions require approval, adding a layer of control over automated financial actions.

22
00:08:06.405 --> 00:08:26.693
This feature was initially launched in May as 'Gemini Spark' for select testers and Google A.I Ultra subscribers before being renamed and expanded. To protect sensitive information during task execution, Google implements security measures including isolated virtual environments and user-adjustable permissions.

23
00:08:29.093 --> 00:08:54.437
KPMG has officially launched its new A.I-enabled audit platform, KPMG Clara intelligence, in the United States. The system integrates specialized A.I agents directly into auditor workflows to connect client data, technical knowledge, and risk assessment. This move follows a period of testing where KPMG staff used these agents on privately held clients across multiple countries.

24
00:08:54.787 --> 00:09:16.419
During the pilot phase, the A.I agents were capable of touching more than 500 audits, performing tasks such as selecting financial transactions for closer inspection. The platform is designed to allow auditors to spend more time on direction, supervision, and review, while the agents automate routine tasks and first-pass documentation.

25
00:09:16.769 --> 00:09:39.169
Christian Peo, Vice Chair of Audit and Assurance at KPMG United States, stated that while technology helps remove barriers between data and action, auditor judgment cannot be automated. This highlights the firm's stance that A.I serves as a tool to enhance human oversight rather than replace the critical decision-making required in financial auditing.

26
00:09:41.568 --> 00:10:03.584
SID A.I has released SID-2, a new agentic search model designed to replace fixed retrieval-augmented generation pipelines. Unlike traditional systems that follow a static path, SID-2 dynamically decides its search steps, operating as a specialist that iteratively searches, reads, and adapts to find information.

27
00:10:03.934 --> 00:10:30.771
The model introduces a 'Swarm' mode, where multiple agents coordinate to divide and conquer search tasks. In self-reported benchmarks, SID-2 Swarm scored 0.720 on the NDCG metric, outperforming GPT-6 Astra's score of 0.704. However, it is important to note that these performance claims have not been independently verified, and no model weights or size details have been disclosed.

28
00:10:31.121 --> 00:10:54.225
SID-2 also features 'Metric Steering,' which allows users to adjust the trade-off between precision and recall for different applications. While the model offers a large context window, latency figures provided by SID A.I assume specific high-end GPU configurations, suggesting that real-world performance may vary based on infrastructure.

29
00:10:54.575 --> 00:11:11.023
That's A.I Voices for today. I'm Maya. Thanks for listening.
Follow the show on Spotify, YouTube, Apple Podcasts, Amazon Music, or your favourite podcast platform, and explore the original voices and sources in the episode notes. See you tomorrow.

