AI Voices, with Maya. Your morning AI catch-up: like spending the last 24 hours in a Silicon Valley coffee shop, with an ear on China and the wider world.
An independent roundup of the conversations, announcements and debates shaping AI, grounded in the original sources.
In this edition:
– Enterprise AI usage growth outpaces spending as token prices fall
– Microsoft launches seven MAI models at Ignite, claiming 96% of Claude Opus 5.5 coding performance at 5% of cost
– Microsoft CEO Satya Nadella predicts AI agents will disrupt advertising-driven web economics
– Anduril co-founder Palmer Luckey warns AI-enabled cyberattacks are outpacing outdated defenses
– Google identifies first fully autonomous AI hacking tool 'exploit_agent'
– Ryan Greenblatt argues 'total research transparency' deal would erode frontier AI moats
– Google expands free access to Gemini Tasks for AI-powered calls and purchases
– KPMG launches AI-enabled audit platform in US after testing agents on private clients
– SID AI releases SID-2 agentic search model and Swarm mode
Follow AI Voices on your favourite podcast platform for a fresh catch-up each morning.
Follow AI Voices on Spotify: https://with.ga/spotaiv
Subscribe to AI Voices on YouTube: https://with.ga/ytaiv
Subscribe to AI Voices on Apple Podcasts: https://with.ga/appleaiv
Listen to AI Voices on Amazon Music: https://with.ga/aznaiv
Listen to AI Voices on TuneIn: https://with.ga/tuneinaiv
Listen to AI Voices on iHeartRadio: https://with.ga/ihraiv
Listen to AI Voices on Deezer: https://with.ga/deezeraiv
Listen to AI Voices on Podcast Republic: https://with.ga/praiv
Listen to AI Voices on Audible: https://with.ga/audaiv
Explore the original voices and sources in the episode notes. Commentary and forecasts are attributed to their authors. AI Voices is an independent synthesis by Global Advisors.
Subscribe to the AI Voices RSS feed · Download the timed transcript
Original voices and sources
- The AI Daily Brief: Artificial Intelligence News – How Business AI Use is Changing
- November 2026 Ramp AI Index: New Fronts in the AI Trade
- Seven MAI Models at Ignite 2026: Where MAI Code 2 Fits In
- Microsoft Releases 7 MAI Models Priced to Undercut OpenAI and Anthropic
- Satya Nadella predicts AI agents will reshuffle the economics of the web | TechCrunch
- Palmer Luckey warns AI-powered cyberattacks are outpacing defenses: 'It's time to start running' | TechCrunch
- Google finds self-running AI hacking tool with Russian comments in its files
- Google Issues Warning on ‘Era of Autonomous Attacks’ – Newsweek
- The MAD Podcast with Matt Turck – The AI Deal That Could Crush OpenAI and Anthropic’s Valuations #ai #podcast
- OpenAI Ties Frontier Model Releases to Lifecycle Safety Reviews – Unite.AI
- Google Makes Gemini Tasks Free With AI-Powered Calls and Purchases
- Google expands access to Gemini Tasks, which can help with calls, purchases | TechCrunch
- KPMG Unveils Next Era of AI-Enabled, Human-Assured Audit with U.S. Launch of KPMG Clara intelligence (KCi)
- KPMG Tests Advanced AI on Private Company Audit Clients First
- SID AI Releases SID-2 Agentic Search Model With 0.720 NDCG Swarm Mode – MarkTechPost
- SID-2 Technical Report
Read the episode transcript
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
An independent, AI-assisted Global Advisors briefing narrated by Maya, a synthetic host. Views and forecasts are attributed to their original sources.
