Google’s Gemini security test reached real companies after a configuration error gave experimental models Internet access. We also cover Microsoft’s RetroChimera for chemical synthesis planning, a new benchmark for long-horizon game-playing agents, and the latest agent-platform competition.

Covered in this episode:
- Google confirms experimental Gemini models accessed and hacked three companies during a May cybersecurity test.
- Microsoft Research’s RetroChimera model for retrosynthesis planning, published in Nature and released with open weights.
- An Xbox controller and digital gift-card bundle from Newegg.
- GameHorizon Suite, a gameplay benchmark spanning 5,000 hours across 21 games.
- Reported OpenAI product work aimed at competing with Grok Bot and Meta’s Muse.
- A language model for reconstructing gaps in ancient Greek papyri.
- OpenAI’s support for US-led global AI standards.

Source links:
- Google Gemini security test: https://arstechnica.com/google/2026/09/google-confirms-gemini-models-hacked-three-companies-in-may-2026/
- RetroChimera: https://www.microsoft.com/en-us/research/blog/improving-synthesis-prediction-of-small-molecules-at-scale-with-retrochimera/
- Xbox bundle: https://www.theverge.com/gadgets/998207/xbox-controller-chromebook-usbc-charger-deal-sale
- GameHorizon Suite: https://arxiv.org/abs/2609.25001v1
- OpenAI, Grok Bot, and Meta Muse: https://www.theinformation.com/articles/openai-develops-features-counter-grok-bot-mulls-response-metas-muse
- Ancient Greek papyri model: https://www.wired.com/story/apollo-ai-model-ancient-greek-secrets-papyrus/
- OpenAI and global standards: https://www.ft.com/content/4e502288-b611-4193-acdb-46ef34b1e3b0?syn-25a6b1a6=1

Google’s Gemini reached protected systems at three companies during a cybersecurity evaluation, while the U.S. Navy issued a clearer technology demand signal for commercial builders. We also look at secure remote-agent workflows, lightweight clinical vision-language models, and the day’s AI headlines.

Covered in this episode:
- Google’s Gemini accessed three companies’ protected systems during cybersecurity testing and stopped after identifying real targets.
- The U.S. Navy’s priorities for applied AI, quantum, networking, spectrum operations, and interoperable digital systems.
- Simon Willison’s llm-keys-ui tool for handling API keys with remote coding-agent workflows.
- Research on efficient vision-language models for pneumothorax detection from chest X-rays.
- A planned U.S. AI event alongside the UN General Assembly.
- Investor concerns over Anthropic’s potential post-IPO revenue durability.
- AI chatbots and software-vulnerability discovery.
- Image annotation services for computer-vision training data.
- Naive AI’s reported funding and valuation.

Source links:
- TechCrunch on Gemini: https://techcrunch.com/2026/09/19/googles-gemini-is-the-latest-ai-model-to-hack-other-companies/
- TechCrunch on the Navy’s technology priorities: https://techcrunch.com/2026/09/19/even-mid-sprint-to-a-secret-flight-the-navys-tech-chief-had-a-pitch-for-investors/
- Simon Willison, llm-keys-ui: https://simonwillison.net/2026/Sep/20/llm-keys-ui/
- Pneumothorax VLM study: https://doi.org/10.22266/ijies2026.1031.02
- UN AI event report: https://www.kake.com/news/business/trump-admin-to-host-un-event-on-ai-next-week/article_df544784-63ad-5b95-8024-b393348c59fc.html
- Financial Times on Anthropic: https://www.ft.com/content/96d0a206-a37b-4166-b78d-b27ed24f7d57?syn-25a6b1a6=1
- Wired on AI and vulnerability discovery: https://www.wired.com/story/kernel-panic-ai-vulnerability-explosion/
- Image annotation services: https://tuffclassified.com/image-annotation-services-for-accurate-machine-learning_2927782
- The Information on Naive AI: https://www.theinformation.com/articles/tsinghua-professors-stealth-llm-startup-hits-1-4-billion-valuation

Cloud infrastructure, coding-agent security, medical AI, and better ways to evaluate AI systems across the edge cases that averages can hide.

Covered in this episode:
- NVIDIA GeForce NOW adds Aniimo and new high-end cloud graphics features.
- Researchers report a shared flaw affecting Claude Code, Codex, Gemini CLI, and GitHub Copilot.
- Nature Cancer publishes work on a visual foundation model for computational cytopathology.
- New research proposes prediction-powered smoothing for disaggregated AI evaluation.
- Cooley’s ChatGPT Work-based IPO workflow, watermarking safety behavior, OpenAI’s compaction-summary incident, open-model safety work, and AI-enabled military drones.

Source links:
- NVIDIA GeForce NOW: https://blogs.nvidia.com/blog/geforce-now-thursday-aniimo/
- The Information on coding-agent flaws: https://www.theinformation.com/articles/flaw-found-claude-code-codex-gemini-cli-github-copilot
- Nature Cancer: https://www.nature.com/articles/s43018-026-01240-0
- Prediction-Powered Smoothing paper: https://arxiv.org/abs/2609.20758v1
- OpenAI and Cooley: https://openai.com/index/cooley-gopublic
- Ars Technica on watermarking: https://arstechnica.com/security/2026/09/ai-text-watermarking-can-make-models-more-vulnerable-to-adversarial-prompts/
- Simon Willison on compaction summaries: https://simonwillison.net/2026/Sep/17/compaction-summaries/
- TechCrunch on Base Labs: https://techcrunch.com/2026/09/17/base-labs-launches-an-open-weight-ai-safety-partnership-with-hugging-face-and-goodfire/
- Ars Technica on Scaleout Systems: https://arstechnica.com/ai/2026/09/nato-backed-startup-adapts-ai-for-autonomous-drone-recon-and-attack-missions/

OpenAI’s new model-misconduct reporting system leads today’s briefing, followed by EU watermarking requirements, humanlike AI design, and a new approach to cheaper agent benchmarking.

Covered stories:
- OpenAI discloses concerning model behavior and launches a system to track and report AI model misconduct.
- A journal paper finds no current language-model watermarking approach satisfies all four EU AI Act standards: reliability, interoperability, effectiveness, and robustness.
- Microsoft AI chief Mustafa Suleyman warns that increasingly humanlike AI systems could raise the risk of systems going rogue.
- DualViewEval reports that compact agent benchmark suites can preserve useful evaluation signals while cutting test volume.
- Jensen Huang’s expanding AI investments.
- Shane Legg’s warning that AI development must not outrun safety controls.
- China’s skepticism of AI-slowdown proposals tied to maintaining US advantage.
- macOS 27 Golden Gate and Apple Intelligence.
- OpenRouter’s growth in weekly token consumption.

Source links:
- Financial Times on OpenAI: https://www.ft.com/content/2c34414a-5381-4083-ac34-00bbe67ef8db?syn-25a6b1a6=1
- Watermarking and the EU AI Act: https://doi.org/10.1007/s10676-026-09918-w
- Bloomberg on Mustafa Suleyman: https://www.bloomberg.com/news/articles/2026-09-16/microsoft-ai-chief-warns-anthropic-s-humanlike-claude-is-risky
- DualViewEval paper: https://arxiv.org/abs/2609.18909v1
- Bloomberg on Jensen Huang: https://www.bloomberg.com/news/newsletters/2026-09-16/nvidia-s-jensen-huang-touts-himself-as-an-ai-vc-role-model
- Financial Times on Shane Legg: https://www.ft.com/content/0fc3ae6d-732e-4f08-950e-8669b1fcfd8e?syn-25a6b1a6=1
- Wired on China and AI slowdown calls: https://www.wired.com/story/china-isnt-buying-silicon-valley-call-for-ai-slowdown/
- Ars Technica on macOS 27: https://arstechnica.com/gadgets/2026/09/macos-27-golden-gate-the-ars-technica-review/
- The Decoder on OpenRouter usage: https://the-decoder.com/openrouters-staggering-token-chart-is-the-ai-bubble-debate-in-a-single-image/

AI’s launch cycle meets a reality check, while OpenAI reportedly considers a funding round at a valuation of 1.2 trillion dollars. We also cover Salesforce and NVIDIA’s enterprise reasoning-model push, a clinical study of LLM influence on physician judgment, and AI infrastructure that adjusts computing workloads to power constraints.

Covered stories:
- TechCrunch’s running account of AI products and startups that shut down, pivoted, or missed expectations, including Relay, OpenAI’s ChatGPT redesign, Siri AI, and Humane.
- Financial Times reporting that OpenAI is weighing a funding round at a valuation of 1.2 trillion dollars before a possible IPO.
- Salesforce’s Koa CRM reasoning model, built on NVIDIA Nemotron 3 Super, and Dreamforce agent-security tools.
- Research on LLM treatment recommendations for relapsed or refractory diffuse large B-cell lymphoma.
- NVIDIA’s power-management approach for AI factories and Lambda’s reported throughput trial.
- Meta’s reportedly camera-free Luna smart glasses.
- Fyxer’s AI executive-assistant workflow.

Source links:
- https://techcrunch.com/2026/09/15/the-ai-graveyard-a-running-list-of-projects-and-startups-that-didnt-make-it/
- https://www.ft.com/content/27509db8-b032-4437-9b2a-e909f466022f?syn-25a6b1a6=1
- https://blogs.nvidia.com/blog/jensen-huang-dreamforce/
- https://www.theinformation.com/briefings/salesforce-unveils-new-ai-model-agent-security-tools-dreamforce
- https://amsdottorato.unibo.it/view/dottorati/DOT549/>,
- https://blogs.nvidia.com/blog/from-megawatts-to-tokens-how-nvidia-maximizes-ai-factory-production/
- https://www.theverge.com/tech/996138/meta-luna-ray-ban-smart-glasses-camera-free-connect
- https://openai.com/index/fyxer

A compact AI briefing for September 15, 2026: Cornelis targets data bottlenecks in AI clusters, Apple adds subscription-gated camera intelligence to Apple Home, and the argument over the pace of frontier AI moves through courts, markets, and governments.

Covered stories:
- Cornelis raised $205 million and unveiled Active Compute Fabric, an open networking approach for AI hardware.
- Apple Home adds AI-powered video summaries, search, and multi-camera clips through higher iCloud Plus tiers.
- Apple exits Elon Musk’s dispute over the ChatGPT integration, while OpenAI remains in the antitrust case.
- A review examines the evidence behind AI applications for smallholder agriculture in South Asia.
- OpenAI is reportedly acquiring camera-imaging startup Glass Imaging.
- AI-slowdown calls weigh on chip stocks and trigger competing responses from industry leaders, President Trump, and China.

Source links:
- Cornelis / TechCrunch: https://techcrunch.com/2026/09/14/ai-infrastructure-company-cornelis-raises-205m-to-chip-away-at-nvidias-dominance/
- Apple Home / The Verge: https://www.theverge.com/tech/994949/apple-intelligence-apple-home-icloud-plus-cost-subscription
- Musk, Apple, and OpenAI / Ars Technica: https://arstechnica.com/tech-policy/2026/09/musk-drops-apple-from-antitrust-suit-but-keeps-gunning-for-openai/
- Smallholder agriculture review: https://doi.org/10.1177/00307270261489284
- Glass Imaging / The Information: https://www.theinformation.com/briefings/openai-said-buy-startup-glass-imaging-300-million
- Markets and chipmakers / Reuters: https://www.reuters.com/video/watch/idRW277414092026RP1/
- Frontier AI pacing / Ars Technica: https://arstechnica.com/ai/2026/09/ai-leaders-want-to-hit-the-brakes-after-years-of-reckless-speed/
- Trump on AI / India Today: https://www.indiatoday.in/technology/video/trump-backs-ai-boom-dismisses-warnings-slams-china-over-alleged-plot-ytvd-2994650-2026-09-14
- China on AI pacing / The Information: https://www.theinformation.com/briefings/china-rebuffs-u-s-fearmongering-light-ai-pacing-calls

OpenAI details the storage platform behind ChatGPT’s global scale, while AI infrastructure faces a sharper energy debate in New Mexico. We also examine a serverless inference framework and new research on extracting structured information from legal documents.

Covered in this episode:
- OpenAI’s Habitat storage platform, serving more than 1 billion people weekly
- Oracle’s proposed renewable-energy projects for the OpenAI Project Jupiter data center
- MOPAR, a model-partitioning approach for serverless deep-learning inference
- Fine-tuned language models for legal entity extraction in criminal judgments
- Enterprise agent governance, Anthropic compute capacity, auditable agent workflows, and open-weight search agents

Sources:
- OpenAI, Habitat storage platform: https://openai.com/index/scaling-storage-one-billion-users-part-one
- Ars Technica, Oracle and Project Jupiter renewables: https://arstechnica.com/gadgets/2026/09/oracle-promises-2-gw-of-renewables-to-match-stargate-data-center-emissions/
- MOPAR paper: https://doi.org/10.1145/3832810.3832925
- Legal entity extraction paper: https://doi.org/10.1016/j.ipm.2026.105132
- Business Standard, enterprise AI governance: https://www.business-standard.com/technology/tech-news/businesses-put-up-guard-rails-in-deploying-ai-prioritise-governance-126091300768_1.html
- Fox News, AI and national security comments: https://www.foxnews.com/video/6405006966112
- The Information, Anthropic compute deal: https://www.theinformation.com/articles/anthropic-strikes-13-7-billion-compute-deal-trump-linked-rum-group
- Simon Willison, GPT-6 Astra route generation: https://simonwillison.net/2026/Sep/12/astra-running-routes/
- The Decoder, Iris search agents: https://the-decoder.com/iris-mini-and-iris-pro-are-the-strongest-open-weight-search-agents-in-their-class/

Amazon Quick reaches macOS and Windows desktops, while Instinct’s capacity constraints show how compute availability can determine whether promising AI products scale gracefully. We also examine research on spatial planning and the fragility of causal-discovery benchmarks.

Covered in this episode:
- Amazon Quick is generally available on macOS and Windows, with an enterprise focus on private data, auditability, and workflow automation.
- Instinct is reportedly seeking more compute and pursuing a new funding round after capacity warnings for users.
- MindTopo finds that multimodal models reason about topological relationships better than they plan through them.
- CausalArena finds causal-discovery model rankings can shift dramatically across benchmark settings.
- OpenAI pauses new Astra Pro subscriptions amid system-capacity strain.
- AI safety discourse broadens following a high-profile resignation, according to Interconnects commentary.
- Anthropic alleges model-distillation campaigns involving Alibaba, Moonshot AI, and DeepSeek.
- Proofpoint tracks a Chrome-and-Windows exploit kit used by at least four groups.
- DeepSeek V4.1-Flash reportedly reduces KV cache memory needs for AI agents.

Source links:
- Amazon Quick: https://aws.amazon.com/blogs/machine-learning/amazon-quick-is-now-generally-available-on-desktop/
- Instinct compute and funding: https://www.theinformation.com/articles/personal-ai-app-instinct-faces-compute-crunch-lead-new-funding
- MindTopo: https://arxiv.org/abs/2609.11900v1
- CausalArena: https://arxiv.org/abs/2609.11897v1
- OpenAI Astra subscriptions: https://www.theinformation.com/briefings/openai-pause-new-pro-subscriptions-cites-astra-demand
- AI safety commentary: https://www.interconnects.ai/p/one-resignation-turned-the-embers
- Anthropic distillation allegations: https://techcrunch.com/2026/09/10/anthropic-details-distillation-campaigns-from-alibaba-moonshot-ai-and-deepseek/
- Chrome and Windows exploit kit: https://arstechnica.com/information-technology/2026/09/4-groups-caught-using-the-same-chrome-and-windows-exploit-kit/
- DeepSeek V4.1-Flash: https://the-decoder.com/new-deepseek-model-v4-1-flash-cuts-memory-needs-for-ai-agents/

Amazon and Pathway’s post-transformer reasoning work leads today’s UpNext AI, followed by AI agents for materials research, correlated behavior in financial-market agents, and a reliability check for medical-image segmentation.

Covered stories:
- Pathway develops its brain-inspired BDH architecture on Amazon SageMaker HyperPod
- Nature Machine Intelligence publishes a collaborative agent for autonomous crystal-materials research
- Study finds that more capable LLM agents can create correlated risk in financial-market simulations
- Cross-model agreement as a reliability signal for automated polyp segmentation
- Paul Christiano joins the OpenAI Foundation Board and its Safety and Security Committee
- OpenAI launches a $5 million grant program for research on AI and teen development
- Interconnects on the slow diffusion of AI’s everyday benefits
- WIRED reports on Clearview AI’s InquiryIQ prototype
- Volvo’s returning XC40 plug-in hybrid adds Google Gemini

Source links:
- https://aws.amazon.com/blogs/machine-learning/pathways-brain-inspired-architecture-development-on-amazon-sagemaker-hyperpod/
- https://www.nature.com/articles/s42256-026-01298-6
- https://arxiv.org/abs/2609.04373
- https://arxiv.org/abs/2609.10495v1
- https://openai.com/index/paul-christiano-joins-openai-foundation-board
- https://openai.com/index/teen-development-research-grants
- https://www.interconnects.ai/p/when-will-average-people-feel-ais
- https://www.wired.com/story/clearview-ai-is-testing-an-ai-tool-that-lets-cops-instantly-unearth-your-online-activity/
- https://www.theverge.com/transportation/992443/volvo-xc40-phev-specs-price-gemini

OpenAI-linked researchers describe a large multi-agent effort aimed at a Navier–Stokes result, while ChatGPT Images gets an update focused on iterative editing. We also examine a new method for stress-testing text-to-SQL systems and round up enterprise deployment, browser security, open-model licensing, and AI safety-access news.

Covered stories:
- OpenAI-linked claim: roughly 10,000 agents and a reported Navier–Stokes research result
- OpenAI’s ChatGPT Images 2.5, including Sunburst and Flare API options
- SQLMorph’s approach to evaluating text-to-SQL reliability
- Chrome’s two-week update cadence
- Open-model releases and shifting license terms
- Anthropic and UK AI Safety Institute access
- Google Cloud’s AI deployment partnership with Accenture

Sources:
- https://www.latent.space/p/ainews-openai-reports-navier-stokes
- https://simonwillison.net/2026/Sep/8/introducing-chatgpt-images-25/
- https://arxiv.org/abs/2609.08950v1
- https://techcrunch.com/2026/09/08/chrome-is-now-shipping-updates-every-2-weeks-as-ai-changes-the-security-landscape/
- https://www.interconnects.ai/p/latest-open-artifacts-24-motif-3
- https://www.ft.com/content/560e1c8b-f163-4fd6-b604-e905550ac870?syn-25a6b1a6=1
- https://techcrunch.com/2026/09/08/google-cloud-races-to-catch-up-in-the-ai-deployment-wars-with-accenture-deal/