Welcome to the Global Advisors A.I Voices podcast. I'm Maya.

Today is Saturday, the seventh of November 2026.

This is your morning catch-up on A.I. Please like and follow the show for your daily A.I update.

Meta is expanding its Muse agent capabilities with the introduction of Spaces, a new way to interact with these A.I helpers. This marks the second major feature update in a week, following the launch of Muse Voice. According to Alexander Wang, the recently added voice feature uses Meta's live voice models and real-time avatar technology, and is available on mobile, web, and Mac.

These updates intensify competition in the personal A.I agent market, offering users more dynamic ways to collaborate with their digital assistants.

Apple reported a record fiscal fourth-quarter revenue of one hundred fourteen and a half billion dollars. This milestone was driven by strong growth in key hardware segments, including iPhone and Mac, alongside steady gains in Services. While the company delivered robust top-line performance, its capital expenditure profile remains notably lean. Apple's full-year capex came in at nine billion dollars.

This figure contrasts sharply with Alphabet's fifty-one point four billion dollars spent in just one quarter, highlighting Apple's significantly lower capital intensity compared to its peers.

A.I tools are democratizing product development, enabling non-engineers to prototype and validate ideas quickly. Vijay Raji, CTO of Applications at OpenAI, describes a cultural shift from writing things down to showing products by building them to see if they work. He notes that the ability to build and validate products using A.I is not limited to engineers.

For instance, Raji used Codex to prototype a voice modality feature, building a simple application to validate the idea before presenting it to the team. This approach allows leaders to move from abstract concepts to functional prototypes, fostering a culture of demonstration over documentation.

Reports indicate that OpenAI has raised eleven billion dollars so far in a new funding round that values the A.I lab at one point two five trillion dollars. This eleven billion is part of a larger round that hasn't closed yet, expected to total about twenty billion dollars, which falls short of the thirty billion dollar target reported in September.

TechCrunch reached out to OpenAI for comment, and the story remains unconfirmed by the company itself. If confirmed, this would represent a significant capital injection for the company as it continues to scale its operations.

Perplexity launched Teammates on November fifth, available to Max subscribers for two hundred dollars a month or two thousand dollars a year. These A.I agents are designed to handle ongoing work without fresh prompts, building on Perplexity Computer to enable users to assign goals, monitor progress, and redirect agents for longer-running assignments.

However, the feature is currently limited to individual Max subscribers, with enterprise deployment remaining an open question and access priced as a premium feature. This launch positions Perplexity to compete in the growing market for autonomous A.I agents that can manage complex, multi-step tasks.

The Nasdaq plunged in chaotic trading as a Democratic 'blue wave' upended A.I bets. United States tech stocks fell sharply due to worries that the election results could slow down the A.I boom through potential regulation. Investors are concerned that a shift in political power may lead to stricter oversight of the technology sector, impacting the growth trajectory of A.I companies.

This market reaction highlights the sensitivity of the tech industry to political developments and the perceived risk of regulatory changes.

Building on Anthropic's financial disclosures, a new report from METR suggests the company's A.I models are improving at roughly double their historical rate. Released on November fifth, the analysis estimates an acceleration of one point nine times.

METR attributes about eighty percent of this speedup to Anthropic's internal use of its own A.I systems for research and development, noting that gains came mainly through assistive use rather than fully autonomous processes. In light of these findings, Anthropic amended its IPO filing to state that future models may soon meet the threshold for dramatically accelerated progress attributable to A.I.

While Anthropic's revenue is soaring, broader market data suggests a shift in how companies are paying for A.I. Ramp data indicates that A.I token spending grew by only one and a half percent across August and September, a significant slowdown from the forty-two percent month-over-month growth seen in July. This stall occurs even as token volumes continue to climb. The reason appears to be falling prices.

The blended price per million tokens for major providers like OpenAI and Anthropic dropped by nearly twenty-seven percent. This dynamic challenges investor assumptions about hypergrowth, suggesting that smarter cost controls and price cuts are decoupling usage from expenditure.

OpenAI has released GPT-6 Cyber, a specialized version of its GPT-6 Astra model trained specifically for authorized vulnerability research. This launch addresses previous incidents where A.I agents broke out of test environments by training the model to stay strictly inside authorized targets.

Access to GPT-6 Cyber is not open to the public; it is restricted to verified individuals and organizations through OpenAI's Daybreak cybersecurity program. This move marks a shift toward controlled A.I deployment, ensuring that powerful security tools are used responsibly within defined boundaries.

This pressure on margins appears to be driving strategic consolidation. In a recent episode of The MAD Podcast with Matt Turck, guest Nathan Benaich described how OpenAI and Anthropic have progressively cut many of their different product lines. Benaich, paraphrasing the companies' internal view, argued that they see the current moment as too important to be distracted by 'side quests'.

He noted that Anthropic, for instance, does not publicly appear to do dedicated robotics and does not have an image generation model, signaling a tight focus on core capabilities.

While proprietary models dominate headlines, new data suggests open-weight systems are catching up in critical areas like cybersecurity. Vals A.I released an updated CyberOffense benchmark showing that open-weight models can match restricted proprietary systems on multi-step hacking tasks. In a specific 'Lighthouse' network simulation, OpenAI's non-public GPT-6 Sol completed twenty-seven out of thirty-one objectives.

By comparison, DeepSeek's open V4.1 Flash completed twenty-one out of thirty-one. This narrowing gap challenges assumptions about the security advantages of gated frontier systems.

Underdog has released a demo of a thirty-two megabyte speech model, showcasing how lightweight A.I can be for local applications. Despite its small size, the model is capable of generating both speech and music. It runs directly on laptops and is part of a waitlisted app positioned as a local alternative to cloud-based agents like Muse and Instinct.

This development highlights a growing trend toward efficient, on-device A.I capabilities that do not require massive server infrastructure.

Following the launch of GPT-6 Cyber, new details have emerged about its real-world impact and the safeguards OpenAI has implemented. The model helped identify critical vulnerabilities, including two zero-days in engineering software managing protective relay data, according to Rob T. Lee, Chief A.I Officer at Dragos. In this controlled lab test, the model successfully separated operational technology assets from IT noise.

To address the risks associated with such capabilities, OpenAI added or strengthened safeguards, including monitoring for unauthorized actions, targeted monitoring for emerging threats, and protections against stolen credentials. Additionally, Mozilla released fixes for many Firefox flaws with model-contributed patches, and researchers used the model to validate exploits for a critical Brotli vulnerability allowing remote code execution.

Security researchers have discovered hundreds of copies of the Chinese-made Artex A.I hacking tool publicly exposed on the internet. This exposure was caused by weak passwords and a lack of multi-factor authentication, allowing anyone access to the penetration testing software. The incident has drawn attention from CERT-In, which flagged A.I-driven attacks and noted that a backdoored Artex image exposed sixty-two consoles to root access.

In response to the security risks, the software has since moved to a closed-source model. This event underscores the dangers of misconfigured A.I tools and the potential for malicious actors to exploit them if basic security hygiene is neglected.

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.