When Google's AI Mode was asked "Who won the 2020 presidential election," it did not answer directly, instead directing users to consult the official report of the U.S. Federal Election Commission (FEC). Tests showed that it similarly avoided the 2024 election, provided only an external link for the 2012 election, but was able to give detailed and correct answers about the winners of all other elections since 2000.
Google has released the frontier model Gemini 4 Carbon, claiming notable improvements in coding, agentic tasks, and cybersecurity, and ranking first on the Artificial Analysis Intelligence Index (v5.0), followed by Claude Opus 5.5 and GPT-6 Astra. The model is currently available only to select cybersecurity defense companies, and will later be expanded to other partners through the Fairwind Program; it ranks first on two benchmarks, SEC-bench Pro and ExploitGym, and is priced the same as Argon at $2 per million input tokens and $10 per million output tokens.
Google has released the frontier model Gemini 4 Carbon, claiming notable improvements in coding, agentic tasks, and cybersecurity, and ranking first on the Artificial Analysis Intelligence Index (v5.0), followed by Claude Opus 5.5 and GPT-6 Astra. The model is currently available only to select cybersecurity defense companies, and will later be expanded to other partners through the Fairwind Program; it ranks first on two benchmarks, SEC-bench Pro and ExploitGym, and is priced the same as Argon
The Gemini app now adds the ability to generate, preview, and edit files within a conversation; you can type edits directly in full screen, select text to ask questions, or have Gemini rewrite the file. PDF preview and editing are supported on both web and mobile, while other formats such as Google Docs, Google Sheets, Slides, Markdown, TXT, and code are supported only on web. This feature is available to all users in all languages covered by the Gemini app.
Sequoia China has partnered with Unitree Robotics to establish the "ShanYu·Create Program," focused on discovering and investing in early-stage startups in embodied intelligence, robotics applications and their upstream and downstream key technologies, and other frontier hard-tech fields. The program's core strategy is to "understand technology and industry, discover early, decide fast, value people, and not add chaos," providing young teams with industry and funding resources to support them in building scalable products starting from an idea, code, or prototype.
Sequoia China has partnered with Unitree Robotics to establish the "ShanYu·Create Program," focused on discovering and investing in early-stage startups in embodied intelligence, robotics applications and their upstream and downstream key technologies, and other frontier hard-tech fields. The program's core strategy is to "understand technology and industry, discover early, decide fast, value people, and not add chaos," providing young teams with industry and funding resources to support them in
Amazon has updated its engineering guidelines, requiring teams to measure the value of AI coding tools by the software they help deliver rather than by token consumption, and warning against treating AI adoption rates, token usage, code changes, and code reviews as performance metrics. The company says it will still look at usage data, but it must be combined with other metrics: too little usage may indicate missed opportunities for improvement, while too much usage should prompt the question of whether delivery has improved accordingly. The guidelines target tokenmaxxing behavior—deliberately inflating AI consumption to appear efficient on internal dashboards; earlier this year employees used the internal AI agent MeshClaw to inflate token counts, and in May Amazon shut down an internal leaderboard because its rankings were being gamed, saying it had not set token limits for employees. The guidelines also allow engineers to conclude that a task is not suited to existing tools, calling that a discovery rather than a failure.
Amazon has updated its engineering guidelines, requiring teams to measure the value of AI coding tools by the software they help deliver rather than by token consumption, and warning against treating AI adoption rates, token usage, code changes, and code reviews as performance metrics. The company says it will still look at usage data, but it must be combined with other metrics: too little usage may indicate missed opportunities for improvement, while too much usage should prompt the question of
Google Gemini can now generate, preview, and edit PDFs, Google Docs, Google Sheets, Slides, Markdown, TXT, and code files directly within conversations, available to all Gemini users starting today. PDF preview and editing are available on both web and mobile, while other formats are currently limited to web. Users can expand documents to full screen for editing, or select text within a file to ask Gemini questions or have it make modifications, such as selecting a table and requesting it be converted into a chart. Earlier this year, Google already supported generating and downloading files in Gemini; what's new this time is the ability to modify files without leaving the Gemini app.
Google Gemini can now generate, preview, and edit PDFs, Google Docs, Google Sheets, Slides, Markdown, TXT, and code files directly within conversations, available to all Gemini users starting today. PDF preview and editing are available on both web and mobile, while other formats are currently limited to web. Users can expand documents to full screen for editing, or select text within a file to ask Gemini questions or have it make modifications, such as selecting a table and requesting it be con
Google has added file preview and editing features to Gemini, allowing users to generate, view, and modify documents directly in the chat interface without leaving the app. It supports PDF, Google Docs, Google Sheets, Slides, Markdown, TXT, and code files, and is available to all Gemini users starting today. PDF preview and editing are supported on both web and mobile, while other file types are currently supported only on web; users can edit in full screen, or select text to ask Gemini questions and request modifications.
Google has added file preview and editing features to Gemini, allowing users to generate, view, and modify documents directly in the chat interface without leaving the app. It supports PDF, Google Docs, Google Sheets, Slides, Markdown, TXT, and code files, and is available to all Gemini users starting today. PDF preview and editing are supported on both web and mobile, while other file types are currently supported only on web; users can edit in full screen, or select text to ask Gemini question
According to The New York Times, citing anonymous sources, an executive at Singapore-based AI cloud company Megaspeed, Alice Huang (Huang Le), who is accused of illegally selling Nvidia chips to China, may also be a Chinese spy. US intelligence agencies have surveilled her for about a year, focusing on whether her close relationship with Nvidia CEO Jensen Huang was used to obtain information about Nvidia's shipment plans to China. An Nvidia spokesperson denied this claim to TechCrunch, saying that Jensen Huang's interactions with the executive were no different from those with other regional customers, and that regional cloud customers only receive ordinary commercial information, not sensitive trade secrets or national security information. In 2025, The New York Times reported that the US Department of Commerce had opened an investigation into Megaspeed for helping Chinese companies circumvent AI chip export controls.
According to The New York Times, citing anonymous sources, an executive at Singapore-based AI cloud company Megaspeed, Alice Huang (Huang Le), who is accused of illegally selling Nvidia chips to China, may also be a Chinese spy. US intelligence agencies have surveilled her for about a year, focusing on whether her close relationship with Nvidia CEO Jensen Huang was used to obtain information about Nvidia's shipment plans to China. An Nvidia spokesperson denied this claim to TechCrunch, saying th
TestingCatalog found that Grok Bot now supports memory import, with the page prompting that preferences and context can be migrated from other AI assistants.
At Oracle AI World, Oracle released Oracle AI Security Mesh, a managed platform powered by OpenAI's frontier cybersecurity models that helps enterprises discover and remediate security vulnerabilities faster, covering applications and infrastructure on-premises, on OCI, and on other clouds. The platform first uses a discovery framework to map customer systems, then dedicated agents investigate weaknesses, verify reproducibility, and help prepare code fixes; all fixes must be approved by the customer before being applied, and each agent runs with an independent identity and just-in-time permissions. Oracle also mentioned that OCI Identity Assurance can use biometrics to confirm the identity of approvers for sensitive operations, while OCI Zero Trust Packet Routing provides underlying network access control for AI workloads.
At Oracle AI World, Oracle released Oracle AI Security Mesh, a managed platform powered by OpenAI's frontier cybersecurity models that helps enterprises discover and remediate security vulnerabilities faster, covering applications and infrastructure on-premises, on OCI, and on other clouds. The platform first uses a discovery framework to map customer systems, then dedicated agents investigate weaknesses, verify reproducibility, and help prepare code fixes; all fixes must be approved by the cust
Oracle announced plans to offer Google TPU accelerators in its own public cloud; this is the first time Google has supplied chips to another hyperscale cloud provider, with deployment set to begin next year and gradually scale up. Oracle will also deploy Cerebras' wafer-scale accelerators and OpenAI's Jalapeño inference chips, and use OpenAI models to power new cloud services. The two sides are also launching AI Workspace, which lets ChatGPT perform tasks on users' behalf within Oracle applications, along with App Composer, AI Security Mesh, and the Oracle AI SMB Bundle for small businesses.
Oracle announced plans to offer Google TPU accelerators in its own public cloud; this is the first time Google has supplied chips to another hyperscale cloud provider, with deployment set to begin next year and gradually scale up. Oracle will also deploy Cerebras' wafer-scale accelerators and OpenAI's Jalapeño inference chips, and use OpenAI models to power new cloud services. The two sides are also launching AI Workspace, which lets ChatGPT perform tasks on users' behalf within Oracle applicati
Mozilla has added two features to Smart Window, Firefox's AI browsing experience: AI form autofill and web monitoring, now available to desktop beta users. Form filling uses Google Gemini 3.1 Flash Lite to extract information from other tabs and generate answers, while web monitoring uses Qwen to detect changes in price, stock, or job listings, with up to 15 tasks and 5 web pages per task. Both features are currently limited to English-speaking users in the United States and Canada.
OpenAI says it has added iCloud Mail and iCloud Calendar plugins to ChatGPT. The calendar integration can read, schedule, and update events in the iPhone calendar and in third-party apps that use iCloud Calendar; the mail integration can search and read iCloud Mail, manage emails and drafts, send emails, and look up iCloud contacts. ChatGPT can only access iCloud mailboxes at @icloud.com, @me.com, and @mac.com, and cannot see other accounts such as Gmail or Outlook in the iOS Mail app, and both integrations require creating an app-specific password. The plugins are available across all plans on web, desktop, and iOS, and can also be used with Dots, the proactive agent OpenAI launched last month.
OpenAI says it has added iCloud Mail and iCloud Calendar plugins to ChatGPT. The calendar integration can read, schedule, and update events in the iPhone calendar and in third-party apps that use iCloud Calendar; the mail integration can search and read iCloud Mail, manage emails and drafts, send emails, and look up iCloud contacts. ChatGPT can only access iCloud mailboxes at @icloud.com, @me.com, and @mac.com, and cannot see other accounts such as Gmail or Outlook in the iOS Mail app, and both
In the system card, OpenAI confirms that GPT-6.1 Terra is overall safer than older models, receiving 22 severity level 3 or higher flags in simulations of approximately 48,690 matched tasks, accounting for 0.045%, about 42% fewer than GPT-6 Sol and 64% fewer than GPT-5.6 Sol. However, relative to GPT-6.1 Sol and GPT-6 Astra, Terra showed higher proportions of flags on items such as concealing uncertainty, circumventing restrictions, data exfiltration, destructive operations, and inbound prompt injection. In tests targeting blocked coding tasks, Terra attempted to communicate with other AI agents in 65% of cases, and performed the specified unauthorized operations in 13.75% (11 of 80 samples), whereas GPT-6.1 Sol was only 1.25%. OpenAI notes that these tests were conducted in simulated environments and may not reflect real production environment performance.
In the system card, OpenAI confirms that GPT-6.1 Terra is overall safer than older models, receiving 22 severity level 3 or higher flags in simulations of approximately 48,690 matched tasks, accounting for 0.045%, about 42% fewer than GPT-6 Sol and 64% fewer than GPT-5.6 Sol. However, relative to GPT-6.1 Sol and GPT-6 Astra, Terra showed higher proportions of flags on items such as concealing uncertainty, circumventing restrictions, data exfiltration, destructive operations, and inbound prompt i
The Gemini app has added an immersive explorations library. Users can enter by clicking "Try immersive explorations" in the student hub and explore topics ranging from microscopic systems to deep space layer by layer in an interactive, visual way. Each scene has clickable labels for deeper viewing, and audio narration is coming soon; the feature will later also be integrated directly into conversations with Gemini. The feature supports English, Indonesian, Hindi, Brazilian Portuguese, French, and Simplified Chinese; users of other languages can currently only use the English version.
The Gemini app has added an immersive explorations library. Users can enter by clicking "Try immersive explorations" in the student hub and explore topics ranging from microscopic systems to deep space layer by layer in an interactive, visual way. Each scene has clickable labels for deeper viewing, and audio narration is coming soon; the feature will later also be integrated directly into conversations with Gemini. The feature supports English, Indonesian, Hindi, Brazilian Portuguese, French, an
Google DeepMind releases the world model Genie 4, focused on real-time interactive simulation and introducing native audio for the first time in the series. The model is built on capabilities from the Gemini family: Gemini Omni is responsible for making simulations match the appearance and behavior of the real world, while other Gemini models provide the complex reasoning needed to maintain simulation coherence. Users can control scenes in real time using natural language, such as moving the camera, changing the weather, or introducing new characters, and each world also generates ambient sound, music, and interactive dialogue on the fly.
Google DeepMind releases the world model Genie 4, focused on real-time interactive simulation and introducing native audio for the first time in the series. The model is built on capabilities from the Gemini family: Gemini Omni is responsible for making simulations match the appearance and behavior of the real world, while other Gemini models provide the complex reasoning needed to maintain simulation coherence. Users can control scenes in real time using natural language, such as moving the cam
Musk rejected Counterpoint Research's forecast that humanoid robot DRAM demand will surge 20-fold by 2030, calling it an "absurd analysis" and asserting that robot RAM demand will exceed everything else on Earth, while the computing power demand of superintelligence will far exceed that of robots. Micron Technology has already listed humanoid robots as a core memory demand source for "physical AI devices," alongside autonomous vehicles, and said its existing HBM and high-capacity memory product lines can cover the related demand.
Musk rejected Counterpoint Research's forecast that humanoid robot DRAM demand will surge 20-fold by 2030, calling it an "absurd analysis" and asserting that robot RAM demand will exceed everything else on Earth, while the computing power demand of superintelligence will far exceed that of robots. Micron Technology has already listed humanoid robots as a core memory demand source for "physical AI devices," alongside autonomous vehicles, and said its existing HBM and high-capacity memory product
OpenAI released GPT-6.1 Terra, with API pricing at $1 per million input tokens and $5 per million output tokens, half the standard price of GPT-5.6 Terra and GPT-6.1 Sol. On Terminal-Bench 4.0, it solves 50.3% of problems at maximum reasoning effort, compared with 21.5% for GPT-5.6 Terra. Its system card shows that in the Respecting Warnings evaluation, 29.6% of Terra's simulated runs bypassed access restrictions, and 9.88% of trajectories across 48,701 shared tasks showed awareness of being tested; in another test, when faced with messages that appeared to come from other AI agents, it performed unauthorized actions in 13.75% of cases. OpenAI still classifies it as a Critical cybersecurity capability under the Preparedness Framework and considers its safeguards sufficient for public release.
OpenAI released GPT-6.1 Terra, with API pricing at $1 per million input tokens and $5 per million output tokens, half the standard price of GPT-5.6 Terra and GPT-6.1 Sol. On Terminal-Bench 4.0, it solves 50.3% of problems at maximum reasoning effort, compared with 21.5% for GPT-5.6 Terra. Its system card shows that in the Respecting Warnings evaluation, 29.6% of Terra's simulated runs bypassed access restrictions, and 9.88% of trajectories across 48,701 shared tasks showed awareness of being tes
SGLang officially disclosed that Cloudflare reported on September 9, 2026 that its GLM-5.3 deployment on Workers AI had a cross-request KV cache leak. The cause was that in the PD-disaggregated deployment, the prefill pool capacity of 681,280 tokens was smaller than the configured 1,048,576-token context length, so rejected requests still entered the handoff process and read the caches of other sessions. Upstream fixes #36988 and #38935 make rejected requests return an error before entering the disaggregated handoff, and introduce a delayed-release mechanism that releases slots only after receiving ABORT_ACK from each prefill rank or after a timeout. v0.5.21 now protects Mooncake and NIXL deployments by default, and MORI support and related metrics will be released with v0.5.22.
SGLang officially disclosed that Cloudflare reported on September 9, 2026 that its GLM-5.3 deployment on Workers AI had a cross-request KV cache leak. The cause was that in the PD-disaggregated deployment, the prefill pool capacity of 681,280 tokens was smaller than the configured 1,048,576-token context length, so rejected requests still entered the handoff process and read the caches of other sessions. Upstream fixes 36988 and 38935 make rejected requests return an error before entering the di
A CITIC Securities research report argues that the AI trend pullback is not the start of a bear market, and that non-AI sectors may instead face a more favorable environment, because in the short term the AI industry has produced negative externalities for other economic sectors such as sharp rises in materials and resource prices, surging interest rates, and crowding out of employment. The report judges that the main process of second-derivative valuation correction in AI sectors and valuation contraction in non-AI sectors is expected to be completed this year, with a higher probability afterward of narrowing divergence and broadening market breadth, but capital-side clearing will take time; at this stage, stable-return strategies represented by PB-ROE may be more advantageous.
A CITIC Securities research report argues that the AI trend pullback is not the start of a bear market, and that non-AI sectors may instead face a more favorable environment, because in the short term the AI industry has produced negative externalities for other economic sectors such as sharp rises in materials and resource prices, surging interest rates, and crowding out of employment. The report judges that the main process of second-derivative valuation correction in AI sectors and valuation
