It's Monday, November 9th: OpenAI put GPT-6 Cyber behind approved security-research access, and Anthropic connected Claude to bank accounts through Plaid with read-only permissions.
💡Help Shape The Collective
What should we cover next? Answer these three quick questions to help us choose the topics and technical depth you find most useful.
How would you describe your technical ability?
Which content would you most like to see?
Which content would you like to see less of?

Live today: Vaire Computing and Purple Orange AI
🎙️ Live today on The AI Collective Podcast
If today's stories caught your eye, join us live at 9 AM PT on The AI Collective Podcast. Catherine McMillan and AJ Green dig into the news, then talk with the people building the next layer of AI infrastructure and operations.
9:15 AM PT · Hannah Earley, Founder & CTO at Vaire Computing. Vaire builds reversible chips that recycle the energy normally lost as heat. Hannah holds a PhD from Cambridge, was named to MIT Technology Review's 35 Under 35 this year, and Vaire has raised more than $12 million.
9:45 AM PT · Max Markovtsev, Founder & CEO at Purple Orange AI. After three decades in global advertising, Max helps organizations turn AI into measurable operational advantage with automation, agentic workflows and a human approval layer.
Your AI Agent Works in a Demo. But Can It Survive Production? 🚀

Building an AI agent in an afternoon is easy. Getting it to operate autonomously, access real-time data, process payments, and deliver reliable results in production? That's where things get interesting.
On November 10 in San Francisco, some of the brightest minds in AI are coming together for Run by Apify, a one-day conference dedicated to taking AI agents from impressive demos to real-world deployment.
Here's who's taking the stage:
OpenAI × WorkOS × x402: Kevin Whinnery (OpenAI), Michael Grinich (WorkOS), and Kevin Leffew (x402 co-author) explore what happens when AI agents become first-class citizens of the internet.
Apify × Stripe: Apify founder Jan Čurn and Stripe's Lauren Dai unpack the shift from humans clicking through interfaces to autonomous AI workers getting things done.
IBM: Tejas Kumar dives into agentic taste - how to tell whether your AI agent is actually doing good work, featuring a live before-and-after demo.
And it's not just talks. The Apify team is flying in for the occasion, giving attendees a rare opportunity to connect directly with the builders shaping the future of autonomous AI.
The day wraps up with a networking party alongside speakers, founders, engineers, and the Apify team. 🥂
📅 When: Tuesday, November 10, 2026 | Doors open at 10 AM
📍 Where: The Pearl, 601 19th Street, San Francisco
🎟️ Exclusive AI Collective Discount: 50% OFF
Use code AICOLLECTIVE to get half off your ticket.

Covering what’s happening on the ground in AI, every Monday.
1️⃣ IN SCOPE: OpenAI Puts Stronger Cyber Agents Behind Restricted Access

Image from OpenAI.
OpenAI introduced GPT-6 Cyber for authorized vulnerability research, reporting stronger exploit discovery and improved adherence to scope while initially restricting access to approved Daybreak Red customers through its API and Codex.
OpenAI reports 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber. Its separate web penetration-testing evaluation puts vulnerability recall at 40.3%, compared with 32.0% for GPT-6 Astra, with models probing applications without direct source-code access.
In the company's scope-alignment evaluation, the model avoided exploiting unrelated services in 99.93% of tests without its deployment prompt and 100% with it. Those are controlled results; OpenAI still recommends sandboxing, action review and human oversight for high-risk workflows.
Firefox distinguished engineer Brian Grinstead says the model helped ship fixes to users. OpenAI reports 81 previously unknown high-severity Firefox vulnerabilities and working exploits for 948 Linux kernel vulnerabilities, with critical findings disclosed to maintainers alongside candidate patches.
Daybreak Red access starts with a select group of approved customers through the API and Codex. OpenAI classifies the model as Critical in cybersecurity under its Preparedness Framework and requires IP allowlisting for Red API access by default, with limited exceptions.
Security teams evaluating autonomous research now have more evidence to examine about both vulnerability discovery and adherence to authorization boundaries. Start with an isolated target and explicit rules of engagement, then compare the agent's actual actions against the deployment safeguards before expanding its access to sensitive systems.
2️⃣ BANK CONTEXT: Claude Gets Read-Only Access To Your Financial Accounts

Image from Plaid.
Anthropic began rolling out Personal Finance in Claude to paid U.S. consumers, using Plaid to connect selected financial accounts with read-only permissions for questions about balances, spending and transactions.
Personal Finance is rolling out in beta to U.S. Pro, Max and Ultra users. Claude can analyze spending across connected accounts, flag possible duplicate charges and explain financial terms; Anthropic says it cannot move, manage or invest money.
Plaid CTO Will Robinson says Plaid powers the account connections and refreshes the transaction data used in answers. Users authenticate through Plaid and choose the accounts they share, so Claude does not ask for or see their bank password.
The Claude Help Center says access includes balances, credit limits and up to 24 months of transactions. Investment-account access is limited to balances. Linked accounts remain unavailable in scheduled tasks, Cowork tasks and Claude Code, and holdings and trades stay outside the current data scope.
Anthropic's privacy explanation says conversations in the Personal Finance hub are excluded from model training; conversations outside that hub follow the user's privacy settings. Unlinking an account stops further access, while information already included in chats remains subject to retention policies.
Builders connecting assistants to sensitive records should pay attention to the permission boundaries and the paths through which data can be shared. Check the documented limitations, verify balances against bank statements and review artifact-sharing behavior before using linked financial data in a workflow that other people can open.
SGLang Summit: Learn From The Teams Building AI

November 12–13, 2026 · Fort Mason, San Francisco
Join 2,000+ engineers, researchers, founders, and AI builders at SGLang Summit, with speakers from NVIDIA, Google, Meta, Perplexity, Thinking Machines Lab, and more.
Bring ideas back to your next AI project: hear how teams improve agents, reduce inference costs, and deploy models in production. Workshops led by NVIDIA, AMD, Modal, Crusoe, and Nebius offer a chance to work through AI coding, deployment, and reinforcement learning workflows with expert instructors. Explore the GPU Expo and live demos, then compare notes with other builders in the lounges.
An Expo Pass includes both days’ technical breakouts, workshops, expo access, complimentary meals and refreshments, and onsite livestream of select main-stage sessions. Workshops require separate registration and approval; seats are limited.
Full Passes are sold out; Expo Passes are available, including Developer & Edu pricing for developers, students, and researchers.
AI Collective members: get 20% off with code COMM20-3BD454C.
📰 Other Headlines
LONGER VIDEO CLIPS: Google released Gemini Omni 1.2 Flash with 15-second clips and conversational editing, cutting 720p generation pricing 30% to $0.07 per second.
INKLING WEIGHTS PENDING: Thinking Machines launched Inkling 2 Small on Tinker, with fine-tuning available and full Apache 2.0 weights planned after external safety testing.
INTERACTIVE WORLD MODELS: NVIDIA added four Cosmos 3 checkpoints for interactive robot and driving simulation, including synchronized camera and LiDAR generation for autonomous-driving research.
MUSE VIDEO CALLS: Meta expanded Muse Voice to U.S. users, enabling conversations with customizable animated avatars on mobile, web and Mac ahead of planned hardware integrations.
COPILOT INSIDE XCODE: GitHub brought Copilot into Xcode's built-in coding assistant in public preview, starting with Xcode 27.3 beta and supporting every Copilot plan.
SOFTWARE FOUNDER SUPPORT: Microsoft and NVIDIA opened Frontier Launchpad applications for established software firms in the UK and Ireland, offering technical guidance and commercial support across two pathways.
AGING RESEARCH RENEWED: Johns Hopkins received a five-year, $20 million award to advance AI and sensor research for healthy aging, including cognitive health and mobility monitoring.
FUSION SIMULATION UPGRADE: ANU secured Google.org support to modernize SPEC, rebuilding the fusion simulation tool for GPUs and developing AI-assisted workflows with open-source software outputs.
RESEARCH SERVICE AGENTS: AiScholar and PolyU announced a funded multi-agent research collaboration to interpret scientific findings and explore an AI-native academic conference platform for research services.
🫵 Want your message in front of 500,000 AI builders?
Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.
For all inquiries, send us a note at [email protected].
The AI Collective is built by volunteers across 180+ chapters in 40 countries.
Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.
🧑💻 About the Editors

About Joy Dong
Joy is a news editor, writer, and entrepreneur at the intersection of AI and blockchain. Whether she is demystifying complex systems in her newsletter, TEA, or building streamlined solutions through her automation agency, Ownly, Joy’s mission is to make emerging tech accessible and actionable for everyone.

About Lindsay Gross
Lindsay is an AI engineer, researcher, and writer focused on how AI systems behave in practice and what it takes to make them safe. Her work sits at the intersection of AI safety, governance, and product design. At AIC and in her newsletter, Hidden Layer, she writes about the questions that matter most as these systems scale.
