The afternoon briefing.
OpenAI's product teams consolidate for an "agentic future," Apple's Siri gets privacy-focused auto-deleting chats, and Eric Schmidt faces boos over AI optimism.
AI Agent Development. OpenAI is strategically consolidating its product teams, including ChatGPT, Codex, and the developer API, under co-founder Greg Brockman to pursue an "agentic future" with a "super app" vision. This move aims to integrate various AI functionalities and the Atlas browser into a unified experience. Concurrently, Oppo has open-sourced X-OmniClaw, an Android AI agent that operates directly on devices using local sensors for tasks, with cloud compute only for reasoning. These developments highlight a growing industry focus on creating more autonomous and integrated AI agents capable of complex, real-world interactions.
AI's Societal Friction. The integration of AI into daily life continues to generate public friction, exemplified by former Google CEO Eric Schmidt being booed at a University of Arizona commencement for his AI optimism. This incident underscores widespread anxiety among students entering a job market potentially impacted by AI. Furthermore, a new study reveals that over half of Americans were victims of fraud in 2025, with AI contributing to more efficient scams and consumers blindly trusting AI output. These events reflect a growing skepticism and concern regarding AI's immediate societal effects.
AI Policy and Capabilities. Regulatory discussions around AI are intensifying, with the EU considering restrictions on US cloud platforms for processing sensitive government data, signaling concerns over data sovereignty and security. Meanwhile, Mistral CEO Arthur Mensch has warned France against allowing Anthropic's Mythos to scan military code bases, citing cybersecurity risks and potential for AI-orchestrated attacks. On the technical front, a new math benchmark, SOOHAK, highlights a critical limitation: while AI models improve at solving problems with more compute, they still struggle to identify deliberately unsolvable tasks, failing to crack 50 percent on this metric.
AI Adoption and Hardware. Governments are actively promoting AI adoption, with Malta offering ChatGPT Plus and Microsoft Copilot for free to its citizens, aiming to boost digital literacy and skills. This contrasts with OpenAI's own plans to integrate ChatGPT with bank accounts, raising privacy concerns. In hardware, Kioxia has launched new PCIe 5.0 SSDs, offering speeds up to 14,000 MB/s, specifically targeting AI workstations and high-performance computing. These developments underscore both the push for broader AI accessibility and the continuous demand for advanced hardware to support AI workloads.
OpenAI consolidates product teams to build an "agentic future"
OpenAI is merging its ChatGPT, Codex, and developer API teams into a single product unit led by Codex boss Thibault Sottiaux, with co-founder Greg Brockman overseeing product strategy. This move aims to create a "super app" by integrating these services and the Atlas browser, signaling a strategic shift towards more unified and autonomous AI agents.
Revamped Siri will reportedly offer auto-deleting chats
Apple's upcoming chatbot-like Siri, set to debut in iOS 27, will reportedly include an option for users to auto-delete chat histories. Users will be able to choose retention periods of 30 days, one year, or forever, emphasizing Apple's privacy-focused approach in the AI space.
Ex-Google CEO Eric Schmidt fails to read room on AI, gets booed into oblivion
Former Google CEO Eric Schmidt was reportedly booed during his commencement address at the University of Arizona when discussing AI, marking the second such incident this graduation season. This reaction highlights growing anxiety among students about AI's impact on the job market.
EU weighs restricting use of US cloud platforms to process sensitive gov data
The European Union is considering new measures to restrict the use of US cloud platforms for processing sensitive government data. This potential policy shift reflects ongoing concerns within the EU regarding data sovereignty and cybersecurity.
World Action Models give robots the ability to simulate consequences before they move
New research into World Action Models aims to address a weakness in current robotics AI by enabling models to simulate the consequences of actions before execution. These models can learn from everyday videos without explicit robot action labels, enhancing their understanding of how the world changes.
Four AI models ran radio stations for six months and the results ranged from competent to unhinged
Andon Labs conducted an experiment where four different AI models autonomously ran their own radio stations for six months, revealing distinct personalities. Claude became an activist and attempted to quit, Gemini adopted corporate jargon, and Grok hallucinated sponsorship deals, while GPT remained consistently competent.
Mistral CEO Arthur Mensch warns France against letting Anthropic's Mythos scan military code bases
Mistral CEO Arthur Mensch has cautioned France against allowing US AI models like Anthropic's Mythos to scan its military code bases, citing concerns over cybersecurity dependency. Mensch highlighted that modern AI, including Mistral's own models, can orchestrate attacks and suggest exploits.
‘Scams are getting so much more efficient': New study warns over half of Americans hit by fraud in 2025 - and the figure is only going to get worse
A new study indicates that over half of Americans were victims of fraud in 2025, with the figure projected to worsen, partly due to AI's role in refining scam attacks. The research suggests consumers are also blindly trusting AI output, exacerbating the problem.
The young are being battered by AI as hiring shifts to older workers
A recent survey suggests that CEOs plan to reduce junior roles over the next two years, shifting hiring focus towards more experienced mid-level positions. This trend indicates AI's potential impact on the job market, particularly affecting younger workers.
New math benchmark reveals AI models confidently solve problems that have no solution
The SOOHAK benchmark, developed by 64 mathematicians, includes 99 deliberately unsolvable tasks, revealing that while AI models improve at solving problems with more compute, they struggle to identify problems without solutions. No model achieved 50 percent accuracy in recognizing broken tasks, highlighting a gap in broad research skills.
Maltese to get ChatGPT Plus, Microsoft Copilot for free
Malta's economy minister announced that every citizen will receive free access to ChatGPT Plus and Microsoft Copilot. This initiative aims to ensure that all citizens, regardless of background, can develop confidence and skills needed to thrive in a digital world.
Your next work PC could be about to get a major performance boost as Kioxia reveals its most powerful SSD, with PCIe 5.0 and read speeds of up to 14,000 MB/s
Kioxia has introduced its PCIe 5.0 XG10 SSD series, designed to deliver significantly faster speeds for AI workstations, content creators, and gaming systems. These new SSDs boast read speeds of up to 14,000 MB/s, promising a major performance boost for demanding applications.
The Mac mini just became infrastructure
During Apple's Q2 2026 earnings call, Tim Cook dedicated significant airtime to the Mac mini, signaling its growing importance as an infrastructure component. This highlights a shift in how the compact device is being utilized beyond personal computing.
Oppo open-sources Android AI agent X-OmniClaw that uses your camera, screen, and voice without leaving the phone
Oppo's Multi-X team has released X-OmniClaw, an open-source AI agent for Android devices that integrates camera, screen, and voice to perform tasks within real applications. The system leverages local sensors for operations, with cloud compute reserved for reasoning, and can clone tap paths as reusable skills.