OpenAI reported that Iranian regime agents used ChatGPT to fabricate fake news articles criticizing U.S. policy toward Iran. The operation ran from July to September 2025, producing nearly 100 articles across about 12 outlets using seven fake journalist personas. OpenAI blocked the accounts involved, though some outlets with large audiences, like Middle East Monitor with nearly two million Facebook followers, had already published the material.
Google's Gemini lost website traffic share for a fourth straight month, dropping from 26.6 percent in July to 25.0 percent in September. Over the same window, ChatGPT climbed from 54.3 percent to 56.4 percent. Anthropic's Claude held steady around 9 to 10 percent. Together, ChatGPT and Gemini still dominate with over 81 percent of tracked generative AI website traffic worldwide.
AMD faces renewed scrutiny after OpenAI sharply cut its annualized revenue estimate in early October 2026, signaling roughly 20 billion dollars less than previously expected. The revision raises questions about long-term demand for AI servers and accelerators. AMD highlighted its partnerships with OpenAI as a key growth pillar, supplying Instinct accelerators, EPYC CPUs, and Helios systems. The concern centers on whether hyperscalers and AI labs will slow large-scale infrastructure builds.
McKinsey argues that organizations are applying AI sovereignty too broadly. Instead, they should classify workloads by data sensitivity and apply sovereignty only where necessary. Sovereignty covers territorial, operational, technological, and legal control. Over-classifying everything increases cost and complexity without real benefit. Defense and national security workloads genuinely require full sovereignty, but most use cases do not.
Aventail CEO Joseph Ward highlighted that losing visibility across the AI data chain or depending on a single provider poses major risks. Organizations can reduce exposure by running open-weight models themselves. Growing regulation in Europe is pushing more companies to examine where their AI services actually operate.
MagicBeanStalk addresses a data problem in RLVR training. Research found that about 88 percent of problems in a large math corpus fall into a zero-advantage trap. The algorithm grows seed problems into new training families. Testing on eight math benchmarks improved gpt-oss-120b by 8.6 points and DeepSeek v4Flash by 5.4 points. In finance, it synthesized over 14,000 problems from 340 solutions.
Rollins reported softer consumer search activity tied to AI changes affecting digital lead recovery. Higher staffing and service costs pressured operating margins despite revenue growth. Management is testing new approaches across multiple AI platforms to restore residential lead flow. The company also integrates Romex Pest Control and other acquisitions targeting 2 to 3 percent annual revenue contribution.
Globe Life has deployed AI tools in administration, underwriting, and sales support to cut costs and boost agent productivity. The administrative expense ratio sits around 7.3 percent of premium. AI tools aim to shorten underwriting cycles and free agents for more selling time. Consensus forecasts project revenue growth of 6.2 percent yearly and earnings rising from about 1.2 billion to roughly 1.3 billion by 2029.
Meta's Muse AI agent reached 5 million downloads in just 22 days, drawing fresh investor interest in AI personal assistants as the next phase of the AI trade. Stephanie Link from Hightower Advisors, whose firm manages 8.5 billion dollars in assets, is buying Meta stock on the belief that AI investments will eventually generate returns. Meta plans to spend up to 145 billion dollars on capital expenditures this year. Hedge fund Minotaur Capital's Thomas Rice is also focused on this emerging area.
Scott Kupor, vice chair and director of the Office of Personnel Management, discussed the Super Intelligence Task Force on Squawk Box. The group is preparing a report covering risks and opportunities related to AI, including safety concerns and recursive self-improvement in AI systems. C3.ai, meanwhile, needs several more quarters before its business stabilizes, according to UBS, though customer engagement and value delivery are improving.
Key Takeaways
- OpenAI blocked ChatGPT accounts used by Iranian regime agents to publish nearly 100 fake articles across about 12 outlets from July to September 2025
- Gemini's traffic share fell for a fourth straight month to 25.0 percent while ChatGPT climbed to 56.4 percent, with Claude holding steady around 9 to 10 percent
- OpenAI cut its annualized revenue estimate by roughly 20 billion dollars in early October 2026, raising demand concerns for AMD's AI server and accelerator business
- McKinsey recommends classifying AI workloads by data sensitivity and applying sovereignty only where needed, warning that over-classifying increases cost and complexity
- Aventail CEO warns that losing visibility across the AI data chain or depending on one provider are major organizational risks
- MagicBeanStalk's data synthesis algorithm improved gpt-oss-120b by 8.6 points and DeepSeek v4Flash by 5.4 points on math benchmarks, and synthesized over 14,000 finance problems from 340 solutions
- Rollins is testing new approaches across multiple AI platforms to restore residential lead flow after AI changes disrupted digital search activity
- Globe Life deployed AI tools in administration and underwriting, targeting a 7.3 percent administrative expense ratio and projected earnings growth from 1.2 billion to 1.3 billion by 2029
- Meta's Muse AI agent reached 5 million downloads in 22 days, with Meta planning up to 145 billion dollars in capital expenditures this year
- Scott Kupor's Super Intelligence Task Force is preparing a report on AI risks and safety concerns, while UBS says C3.ai needs several more quarters to stabilize
Aventail CEO discusses AI control and security
Eve Goode interviewed Joseph Ward, CEO of Aventail, about AI sovereignty. Aventail helps organizations control which AI models they use, what data they share, and where models run. A major risk is losing visibility across the AI data chain or depending on one provider. Organizations can reduce risk by using open-weight models they run themselves. Growing regulation in Europe is pushing more companies to examine where their AI services operate.
McKinsey says not all AI needs to be sovereign
McKinsey argues that governments and companies are applying AI sovereignty too broadly. Instead, they should classify AI workloads based on data sensitivity and apply sovereignty only where needed. Sovereignty has four parts: territorial, operational, technological, and legal control. Over-classifying everything makes systems slower and more expensive. For most workloads, non-sovereign approaches are sufficient, while defense and national security need full sovereignty.
Rollins faces AI search disruption impact
Rollins reported softer consumer search activity due to AI changes affecting digital lead recovery. Higher staffing and service expenses pressured operating margins despite revenue growth. Management is testing new approaches across multiple AI platforms to restore residential lead flow. The company also integrates Romex Pest Control and other acquisitions aiming for 2% to 3% annual revenue contribution. Analysts project revenues of US$5.0b and earnings of US$709.9m by 2029.
Globe Life uses AI to cut costs
Globe Life has deployed AI tools in administration, underwriting, and sales support to reduce costs and boost agent productivity. The administrative expense ratio sits around 7.3% of premium. AI tools aim to shorten underwriting cycles and free agents for more selling time. Consensus forecasts project revenue growth of 6.2% yearly and earnings rising from about US$1.2b to roughly US$1.3b by 2029. Success depends on consistent rollout across legacy systems.
MagicBeanStalk fixes RLVR data problem
MagicBeanStalk is a data synthesis algorithm that creates effective training problems for RLVR models. Research found that about 88% of problems in a large math corpus fall into a zero-advantage trap. The algorithm grows seed problems into new, calibrated training families. Testing on 8 math benchmarks improved gpt-oss-120b by 8.6 points and DeepSeek v4Flash by 5.4 points. In finance, it synthesized over 14K problems from 340 solutions.
OpenAI says Iranian regime used ChatGPT to create fake articles
OpenAI reported that Iranian regime agents used ChatGPT to create fake news articles criticizing the U.S. war against the regime. The agents created seven fake Western journalist personas and got almost 100 articles published in about 12 small to medium news outlets from July 2025 to September. OpenAI blocked the ChatGPT accounts used in this operation. The fake articles included headlines like Trump's Iran Gamble Hardened the Regime it was Meant to Break. Some outlets with large online audiences, such as Middle East Monitor with nearly two million Facebook followers, published the fake articles.
Meta's Muse AI agent opens new chapter in AI investment race
Australia's top technology investors are now betting on AI personal assistants as the next big phase in the artificial intelligence trade. Hedge fund Minotaur Capital's Thomas Rice is among growth managers focusing on this new area. Meta's Muse AI agent has launched this new leg in the AI trade. Earlier in the year, investors made large returns from memory stocks like SK Hynix, Micron and Samsung Electronics that make chips for AI data centers.
Gemini traffic share falls for fourth straight month as ChatGPT regains ground
Google's Gemini lost website traffic share for four consecutive months according to SimilarWeb data. Gemini's share dropped from 26.6 percent in July to 25.0 percent in September. Over the same period, ChatGPT climbed from 54.3 percent to 56.4 percent. Anthropic's Claude held steady around 9 to 10 percent, staying in third place. ChatGPT and Gemini still account for more than 81 percent of tracked generative AI website traffic worldwide.
C3.ai needs more quarters to stabilize, UBS says
UBS says C3.ai needs several more quarters before its business stabilizes. The company's ability to connect with customers and deliver value is improving. However, full stabilization will take more time.
AMD faces new AI demand risk after OpenAI cuts revenue estimate
Advanced Micro Devices faces fresh attention after OpenAI sharply cut its annualized revenue estimate in early October 2026. The revised forecast raises questions about long term demand for AI servers, accelerators and related infrastructure. AMD highlighted its AI data center partnerships, including work with OpenAI, as an important growth pillar. OpenAI signaling about 20 billion dollars less in annualized revenue pulls attention into how durable AI workloads are for suppliers like AMD. The risk involves whether hyperscalers and AI labs slow rack scale buildouts that AMD serves with Instinct accelerators, EPYC CPUs and Helios systems.
Scott Kupor discusses Super Intelligence Task Force goals
Scott Kupor joined the Squawk Box program to talk about the Super Intelligence Task Force. He serves as the vice chair and director of the Office of Personnel Management. The group is working on a report that covers risks and opportunities related to artificial intelligence. They are also discussing how the government should manage safety concerns and recursive self-improvement in AI systems.
Analyst buys Meta stock as AI spending may pay off
Stephanie Link from Hightower Advisors says she is buying Meta stock because its AI investments could create future profits. Her firm manages 8.5 billion dollars in assets and she has 35 years of experience in finance. Meta recently launched a personal AI agent called Muse, which reached 5 million downloads in just 22 days. The company plans to spend up to 145 billion dollars on capital expenditures this year to build new AI businesses. Link believes Meta has the resources to turn this heavy spending into returns for investors over time.
Sources
- Taking control of AI with Aventail
- Governments want to safeguard their tech. But not all AI should be sovereign
- What Rollins (ROL)'s AI Search Disruption Means For Shareholders
- Should AI Cost Cuts Require Action From Globe Life (GL) Investors?
- MagicBeanStalk: Escaping the RLVR Zero Advantage Trap with Recursive Data Synthesis
- Open AI says Iranian regime used ChatGPT to create fake articles
- Meta’s Muse AI agent unleashes new leg in AI trade that has Minotaur, Loftus Peak all in
- Gemini’s Traffic Share Falls For Fourth Straight Month As ChatGPT Regains Ground: SimilarWeb Data
- C3.ai Needs More Quarters to Stabilize, UBS Says
- Does Advanced Micro Devices (AMD) Face New AI Demand Risk After A Customer Reset?
- Super intelligence task force goals: Vice chair Scott Kupor on future of AI regulation
- Investors can now see how Meta's big AI spending will pay off with Muse. Why I am buying now
Comments
Please log in to post a comment.