In partnership with
It's Monday, April 20th: Anthropic shipped Claude Opus 4.7 with automatic cyber-misuse blocking, while OpenAI launched GPT-5.4-Cyber and expanded its Trusted Access program to thousands of verified defenders.
Head over to our Events Portal to get the latest on upcoming AI Collective events near you. Search by city, date, or event format, and join thousands of builders at events across 180+ chapters on every continent (except Antarctica, for now).
Find an event in your city using the link below.👇
The top AI stories from last week, filtered for what will help you stay in the know.
1️⃣ AN EYE TOWARD SAFETY: Anthropic Ships Claude Opus 4.7 With Built-In Cyber Safeguards
Anthropic shipped Claude Opus 4.7 on April 16, rolling the model out across Claude products, the API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. Pricing stays at $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.6. The release is the first Claude model to ship with automatic blocking of high-risk cybersecurity requests and a verification program for legitimate security researchers.
Opus 4.7 posts 70% on CursorBench (up from 58% on 4.6), resolves 3x more production tasks on Rakuten-SWE-Bench, and improves CodeRabbit recall by more than 10%. On Databricks' OfficeQA Pro evaluation the model makes 21% fewer errors than 4.6, and on a 93-task internal coding benchmark Anthropic measured a 13% lift.
Vision now accepts images up to 2,576 pixels on the long edge, more than 3x the resolution of prior Claude models, and Opus 4.7 reached 98.5% on XBOW's visual-acuity test (up from 54.5%). Launch customers on the page include Stripe, Cursor, Replit, Devin, Harvey, Notion, Databricks, Ramp, Bolt, Vercel, and Rakuten.
Policy is the larger shift in this release. Opus 4.7 auto-detects and blocks prompts consistent with offensive cyber use, while letting verified red-teamers, pen-testers, and vulnerability researchers apply for access at claude.com/form/cyber-use-case. Anthropic is treating cyber capability as a gated resource, a workable approach only if the verification pipeline keeps pace with adoption.
For builders shipping developer tools or security products on Claude, the ceiling on what Opus 4.7 will help write just moved, and legitimate workflows now have a documented path to restore it.
2️⃣ DEFENDER ACCESS: OpenAI Launches GPT-5.4-Cyber and Scales Trusted Access Program
OpenAI announced GPT-5.4-Cyber on April 14, a variant of GPT-5.4 fine-tuned for cybersecurity workflows that prior models declined. The model handles binary reverse engineering, patch-diff interpretation, and detection-logic generation for verified researchers and defense teams. Access flows through the Trusted Access for Cyber program, which scales from a February pilot to thousands of individual defenders and hundreds of enterprise teams.
Individual researchers can apply at chatgpt.com/cyber with identity verification, and enterprises request access through an OpenAI account representative. OpenAI held back public benchmark scores for GPT-5.4-Cyber itself and instead pointed to its broader cyber track record, including Codex agents that have fixed over 3,000 critical and high-severity vulnerabilities and Codex for Open Source running on more than 1,000 projects.
Hard prohibitions stay in place for malware creation, data exfiltration, and unauthorized testing. Zero-data-retention environments are limited for GPT-5.4-Cyber, U.S. government access is currently paused pending additional governance review, and OpenAI's $10M cybersecurity grant program remains open for approved defenders who need API credits.
GPT-5.4-Cyber lands a week after Anthropic's Claude Mythos began seeding a tight circle of 40 organizations inside Project Glasswing, and the rollout styles diverge by two orders of magnitude in breadth. OpenAI is betting that identity-based access at scale can replace blanket capability restrictions, while Anthropic is keeping its most permissive cyber model inside a narrow partner ring.
For security startups, the practical takeaway is that major labs are now treating offensive-adjacent capabilities as access-gated rather than prohibited. Expect the KYC layer to become the real competitive surface for AI-native security tooling over the next two quarters.
🔗 Other News
AGENT MAC: Perplexity launched Personal Computer for Mac Max subscribers, a $200/month always-on agent that plugs into Mail, Finder, Slack, Messages, and Calendar.
ROBOT BRAIN: Physical Intelligence released π0.7, a 4B-parameter model that folded unseen laundry on a UR5e with zero laundry-specific training data.
OPEN WEIGHTS: Alibaba open-sourced Qwen3.6-35B-A3B under Apache 2.0, a 35B-parameter MoE with 3B active params targeting agentic coding and spatial reasoning.
VOICE AGENTS: Google launched Gemini 3.1 Flash TTS, a voice model that takes stage directions on pacing, emotion, and tone for narrator-style audio production.
DRUG DISCOVERY: OpenAI released GPT-Rosalind, a research-preview reasoning model for life sciences, with Amgen, Moderna, and Thermo Fisher as enterprise customers.
AI INDEX: Stanford HAI's 2026 AI Index found the US-China model performance gap has collapsed to 2.7% and US-bound AI researcher migration is down 89% since 2017.
SNAP LAYOFFS: Snap cut roughly 1,000 employees after activist investor pressure, with CEO Evan Spiegel saying AI now generates 65% of new code at the company.
CODEX EXPANDS: OpenAI pushed Codex beyond coding into computer use, web workflows, image generation, memory, and automations across existing apps.
SIDE-BY-SIDE SEARCH: Google rolled out AI Mode in Chrome desktop with a conversation view that sits alongside the webpage and pulls content from open tabs into queries.
AGENTIC CODING: xAI confirmed Grok Build and Grok CLI ship next week with parallel and arena execution modes, targeting Claude Code, Codex, and Cursor.
CLAUDE DESIGN: Anthropic debuted Claude Design, a conversational tool that turns prompts into prototypes, slide decks, and marketing collateral.
COMMUNITY SPOTLIGHT
📌 SUPERION: Local AI That Matches GPT-4o in Coding
A 4.5 billion parameter AI model just matched GPT-4o in code generation and math reasoning — and it runs entirely on a laptop. No cloud required. No data leaving your device.
Punky Tiger Labs published benchmark results recently showing their NYMPH AI, powered by the proprietary SUPERION engine, performing at GPT-4o levels on HumanEval and GSM8K benchmarks — while outperforming open-source models up to 6x its size.
“More than 95% of AI will be local. Not just because it keeps getting better — but because it is physically impossible for cloud infrastructure to serve humanity at scale.”
— Álvaro López Zúñiga, Founder & CEO
This breakthrough opens the path for the inevitable shift to local AI as well as supporting regulated companies that cannot share data with the cloud models.
Learn more about what’s driving the local AI shift, the SUPERION engine, and the planned launch of Nymph Pulse in the Mac App Store.
Your pulse on the biggest events and announcements and happening in AI this week, from Noah Frank ⚡️
📅 Events We’re Watching
Mark your calendars and be sure to sign up for these landmark events we’re watching. Be sure to look out for special AIC discounts where available.
April 27 – 29: AIM-2026 (San Francisco, California)
The Third International Conference on Artificial Intelligence and Machine Learning, with keynote speakers from Stanford, University of Maryland, and York University. More academic than trade show. Registration runs $299 to $1,099.
May 4 – 7: IBM Think 2026 (Boston, Massachusetts)
IBM's flagship technology conference, covering enterprise AI, cloud computing, and quantum. Heavy on real-world implementation and use cases across industries like healthcare, finance, and supply chain.
May 27 – 28: AI DevSummit 2026 (South San Francisco, California)
A two-day conference on shipping real-world AI, with tracks on management, machine learning, and enterprise integration. Speakers include Logan Ramalingam (Google Cloud), Kordel France (Toyota), and AIC’s very own Mary Grygleski! Registration starts at $1,080.
June 15 – 18: Databricks Data + AI Summit 2026 (San Francisco, California)
The leading event at the intersection of data engineering, machine learning, and AI, hosted by Databricks. In-person passes run $1,395 to $1,895, but virtual access is free. If you can only attend one event this summer, this is a strong pick.
🔦 Spotlight On: OpenClaw Hype Reveals… Old Truths
Image from Turing College
I've been traveling a lot for work the last few weeks.
In my meetings with both personal friends and business executives, OpenClaw has been a continuing topic of conversation. Donning a lobster pin and all, Peter Steinberger, the creator of OpenClaw and OpenAI's hottest new hire, gave a TED talk that’s been going viral in many AI circles. As he notes, "chatbots give up, agents improvise... you have to experience an agent, it's hard to explain." There's a lot to admire in Steinberger's story. Once a burned-out founder, Steinberger tells the story of rediscovering his spark by sending a WhatsApp bot a voice message and watching it solve the problem in 9 seconds flat.
Fast forward a few months and people are buying Mac Minis specifically to run OpenClaw, to the point that Apple pulled high-memory configurations from sale in early April because of demand (maybe even you!). ClawCons are popping up on multiple continents, lobsters in tow.
But there is an unsexy part to this story that doesn't get talked about nearly enough – the gap between the demo and the “daily driver” using agentic. And we’ve heard this story countless times since the breakout rise of ChatGPT: 88% of organizations now use AI in some capacity, but agent deployment across actual business functions remains in single digits, according to Stanford's 2026 AI Index. Further, Gartner expects nearly 40% or more of agentic AI projects to be canceled by end of 2027.
And so we arrive to that same truth about building useful things. Steinberger's beer sommelier and São Paulo teenager make for a great stage story, but automation is only valuable when you know what problems need solving, and knowing what problems need solving is the hard part that agents don't do for you.
The lobster is loose, and it's not going back in the tank. But just like the Claude Code app-building frenzy that captivated many in our community earlier this year, the "SaaS is dead" narrative keeps getting ahead of itself. Not because enterprise procurement is stubborn, but because people who build quality things will continue to build quality things, inside the lobster tank or not.
The builders who win are the ones who know not how, but why they’re building, and who they’re building for.
Smart starts here.
You don't have to read everything — just the right thing. 1440's daily newsletter distills the day's biggest stories from 100+ sources into one quick, 5-minute read. It's the fastest way to stay sharp, sound informed, and actually understand what's happening in the world. Join 4.5 million readers who start their day the smart way.
🫵 Do You Belong on Our Newsletter?
Share your message with the world’s largest AI community. To inquire about partnership availability, reach out to our team below.
The AI Collective is a community of volunteers, made for volunteers. All proceeds directly fund future initiatives that benefit this community.
Before You Go…
💬 Join Slack: The AI Collective
🧑💼 LinkedIn: The AI Collective
𝕏 Twitter / X: @AICollectiveCo
📸 Instagram: @AICollectiveCo
Get Involved in Your Community
Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.
About Noah Frank
Noah is a researcher, innovation strategist, and ex-founder thinking and writing about the future of AI. His work and body of research explores the economics of emerging technology and organizational strategy.
About Joy Dong
Joy is a news editor, writer, and entrepreneur at the forefront of the emerging tech landscape. A former educator turned media strategist, she currently writes TEA, where she demystifies complex systems to make AI and blockchain accessible for all.