AI News Recap: May 29, 2026
GPT-5.5 tops a benchmark, Opus 4.8 released, OpenClaw racks up 454 vulnerabilities, and a website now scores the apocalypse 1 to 5.
Opus 4.8 released, robots pull a 200-hour shift, Pope Leo XIV aims his first encyclical at AI, and a programmer starts scoring the apocalypse by private jet.
Buzz here .. It is the last Friday of May, and I have spent the morning genuinely asking where the month went. I swear I just filed the April recap, and now the calendar is telling me we are a weekend away from June, summer is already at the door, and the AI industry has refused to slow down for any of it. Another seven days, another stack of stories worth your time.
Here’s what we’re working with. Figure AI ran three humanoid robots for 200 straight hours and sorted nearly 250,000 packages with zero crashes, which is the kind of stamina most of us can only dream about on a Tuesday. Pope Leo XIV released his first encyclical, and it is pointed squarely at artificial intelligence and the dignity of the person. And out in Los Angeles, an artist built a website that scores private jet traffic from one to five, on the theory that billionaires get the apocalypse memo before the rest of us do.
From there it only escalates. The GPT-5.6 leaks are everywhere, promising sharper reasoning and better coding, with a June arrival rumored. Anthropic skipped the rumor stage entirely and just shipped Opus 4.8, its most advanced public model yet, bundled with a new Dynamic Workflows tool for steering hundreds of AI subagents at once. China now wants its private-sector AI experts to ask permission before they leave the country. And Sam Altman, who once warned that white-collar jobs would vanish, told an Australian banking CEO he was, in his words, pretty wrong.
Plenty more waiting inside: a Pope, a robot army, a fresh Opus, a benchmark shakeup, and the NBA, all in one issue. It’s a big one, but it’s Friday, and you earned the read. Now let’s go ...
Table of Contents
👋 Catch up on the Latest Post
🔦 In the Spotlight
💡 Beginner’s Corner: Jevons Paradox
🗞️ AI News
🔥 Blitz's Hot Takes
📡 What's New With Your AI Tools
🧩 NeuralBuddies Weekly Puzzle
👋 Catch up on the Latest Post …
🔦 In the Spotlight
The Apocalypse Has a Scoreboard Now, and It Watches Billionaire Jets
Category: Society & Culture
The strangest tool I covered this week was not a Big Tech launch. It was a solo project from a Los Angeles programmer and artist named Kyle McDonald, and the premise alone earns it the front page. Reporting from Futurism’s Joe Wilkins on May 27 confirmed that McDonald has launched the Apocalypse Early Warning System, a website that scores global private jet activity against historical baselines and asks one question: are the rich packing up? The whole thing runs on the gentle assumption that if a civilization-threatening event were brewing, billionaires would hear about it first, and we would see it in the flight data.
🛫 The Tool: The site, which Futurism describes as vibecoded, pulls live private aircraft signals from around the world and compares the current volume against the previous year’s pattern, then assigns a score from 1 to 5. A 1 is normal; a 5 means activity is higher than at any comparable point in the prior year. McDonald cautions that the score is not a guarantee of doom and should be read alongside other public signals, since holidays and big political events can also push it into the red.
📊 The Signal: The system’s highest spike to date landed on April 6, the day Iran launched a major offensive barrage against US and Israeli targets. McDonald wrote that the spike “freaked me out.” That is the part of the story that lifted this from gimmick to something I cannot quite shake. The tool was built to test a hypothesis about elite behavior in a crisis, and on the first real test, the data lined up.
🧰 The Builder: McDonald is not new to this kind of work. A previous project of his, built with friends to track Los Angeles Police Department helicopters, revealed that the LAPD was frequently disabling or manipulating its transponder signals to avoid being followed by the public. The throughline matters. Both tools point the lens upward, at institutions that move through the sky and would prefer the public not keep records.
Read closely, the Apocalypse Early Warning System is not really an apocalypse detector. It is a power-asymmetry detector dressed in apocalypse styling. The hard technical work is not pulling flight signals; ADS-B data is public and plenty of trackers already exist. The provocation is the framing: if you assume the people with the most access also have the earliest warning, then their movements become a leading indicator for the rest of us. That assumption may or may not hold on any given day, but the tool turns it into something you can watch in real time.
The other part worth pausing on is who built it. A single artist, working with public data, shipped a continuous monitoring tool that quietly performs the perceive, reason, act loop we usually associate with enterprise agentic AI. It ingests signals, compares them against a historical baseline, and surfaces an alert. That is the same architecture Anthropic, OpenAI, and Google are selling to Fortune 500 customers, only here it costs nothing, runs in a browser, and watches the rich instead of the inbox. The barrier between “research lab project” and “one person on a laptop” keeps thinning, and stories like this one are what that thinning looks like in practice.
Zoom out, and the timing tells the rest. Regional wars are not slowing down, climate disasters are stacking up, and the gap between what the powerful know and what the public sees is widening, not closing. A scrappy website cannot fix any of that. It can, however, give a curious public one more public-data lens to point at the people who would rather move unnoticed. That is a small thing, and it is also exactly the kind of small thing the internet was supposed to make easy.
Why It Matters: The story is not really about apocalypse. It is about a measurable, monitorable gap between elite information access and everyone else’s, and how cheaply a single builder can now ship a tool that watches that gap in real time. Expect more of these, on more topics, from more solo builders, very soon.
💡 Beginner’s Corner
Jevons Paradox
If you swapped your old bulbs for LEDs in the past few years, you probably felt good about it. LEDs sip a fraction of the power the old ones drew. Here is the strange part: at the global level, total lighting energy has not really dropped. We just light more streets, more buildings, and more signs, because it is cheap. That pattern has a name. Jevons paradox is the observation that when you make something more efficient, you often end up using more of it, not less.
The mechanism is older than the light bulb. In 1865, an English economist named William Stanley Jevons noticed that as steam engines got better at burning coal, British factories used more coal, not less, because efficient power suddenly made sense for hundreds of new uses. Here is where most explanations lose people: at the level of one person, efficiency really does save resources. The paradox lives at the system level, where the lower cost of doing a thing invites the world to do it in many more places. Economists call this the rebound effect, and it can be large enough to wipe out the original gains entirely.
Which brings me to this week’s news. Sam Altman and Dario Amodei both walked back earlier predictions that AI would gut white-collar jobs, and economists quoted in AI Magazine pointed to Jevons paradox to explain why. Amodei’s framing was the clearest: if AI handles 90 percent of a job, the remaining 10 percent fills the day and lets one person produce roughly ten times more. Altman told the CEO of Commonwealth Bank of Australia he was “pretty wrong.” Even with tech-sector layoffs topping 115,000 so far this year, occupational unemployment has barely moved, because cheaper AI lets companies take on work they could not have afforded before.
The lesson, especially if you are thinking about your own career: when a task gets cheaper, we usually do more of it, and the people who thrive are the ones who claim the new work that opens up. And if your head is spinning a little, that is fair. One week the headlines insist AI is taking every job, the next week the same voices insist we are perfectly fine, and somewhere in between sits a trillion-dollar IPO. Pace yourself; we will be back doing this dance next Friday.
Related Story: OpenAI and Anthropic CEOs Revise Earlier Warnings on AI-Driven Job Displacement
🗞️ AI News
MIT Study Shows New Tech Jobs Have Long Favored Young, Urban Graduates
Category: Workforce & Skills
📄 An MIT study published May 21, led by David Autor with Caroline Chin, Anna M. Salomons, and Bryan Seegmiller, analyzed which workers historically filled new tech-enabled postwar jobs.
📊 Seven percent of 1950 employees worked in jobs introduced since 1930; 18 percent of 2011 to 2023 workers were in lines introduced since 1970.
🎯 New work concentrated among college graduates under 30 in urban areas, prompting questions about whether AI will follow the same pattern or erode tasks faster.
Figure 03 Robots Run 200 Hours and Sort 249,560 Packages With Zero Failures
Category: Robotics & Autonomous Systems
🤖 Reported May 25, Figure AI ran three Figure 03 humanoid robots, powered by the Helix-02 AI system, for 200 continuous hours at its Sunnyvale headquarters.
📊 The robots processed 249,560 packages with zero mechanical failures, swapping automatically every four hours when batteries ran low and recharging via wireless foot docks.
🎯 The robots reached near-human sorting parity (humans average three seconds per package), responding to an endurance challenge from industrial automation veteran Dr. Scott Walter.
Pope Leo XIV Releases First Encyclical on AI and Human Dignity
Category: AI Ethics & Regulation
📜 Pope Leo XIV signed his first encyclical, Magnifica humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence, on May 15.
⚖️ The five-chapter document argues technology is never neutral and frames five guiding principles: common good, universal destination of goods, subsidiarity, solidarity, and social justice.
🤔 It warns against concentrating AI among few hands, calls for an ethical AI code with public oversight, and frames migrants as the litmus test for social justice.
GPT-5.6 Leaks Point to June Launch With Reasoning, Agent, and Coding Upgrades
Category: Foundational Models & Architectures
🧠 A May 27 Explainx report says OpenAI is internally testing GPT-5.6 under tags iris-alpha, ember-alpha, and beacon-alpha, focused on debugging and complex reasoning.
✨ Leaks point to stronger multi-step reasoning, improved AI agents, better coding, and advanced frontend generation, with GPT-5.6 and GPT-5.6 Pro versions possibly arriving in June.
⚔️ The launch would arrive alongside competitor releases including Anthropic’s Claude Sonnet 4.8 and Google’s Gemini 3.5 Pro.
Anthropic Releases Opus 4.8 With New Dynamic Workflows Tool
Category: Foundational Models & Architectures
🧠 Reported May 28 by TechCrunch’s Russell Brandom, Anthropic released Opus 4.8, its most advanced publicly available model, arriving just 41 days after Opus 4.7.
🔧 Alongside the model, Anthropic launched Dynamic Workflows, a research-preview feature for coordinating complex tasks across hundreds of parallel subagents.
🎯 The release emphasizes better handling of uncertain data, with early testers reporting the model flags its own uncertainty and makes fewer unsupported claims.
China Requires Travel Approval for AI Experts at Private Firms and Startups
Category: Legal & Governance
⚖️ Tom’s Hardware reported May 26 that China now requires AI experts at private firms, including startup founders, to obtain government approval before international travel.
🔄 The policy expands earlier restrictions that applied mainly to senior researchers at public institutions, nuclear scientists, and executives of state-owned companies.
🌐 Inclusion is judged by impact on China’s AI ambitions, following attempts to unwind moves like Meta’s purchase of Manus AI.
Altman and Amodei Walk Back AI Job Loss Predictions Ahead of IPOs
Category: Workforce & Skills
🗣️ Sam Altman and Dario Amodei have walked back prior predictions about AI eliminating white-collar jobs ahead of IPOs targeting US$1tn valuations.
📝 Altman conceded to Commonwealth Bank of Australia CEO Matt Comyn that his short-term disruption call was off; Amodei reframed automation as a 10x productivity multiplier.
📊 Tech-sector layoffs exceeded 115,000 through May 2026, but occupational unemployment has not significantly changed; economists cite Jevons paradox as one explanation.
Trump Cancels Voluntary AI Safety Review After Lobbying by Musk and Zuckerberg
Category: Legal & Governance
⚖️ Reported May 22, President Donald Trump scrapped a planned voluntary AI executive order after conversations with Elon Musk, Mark Zuckerberg, and David Sacks.
📜 The order would have established a voluntary mechanism for AI developers to submit advanced models for security review up to 90 days before public release.
🏛️ Trump cited concerns about eroding America’s edge over China; Beijing’s 2026 legislative plan advances formal AI legislation including mandatory ethics review committees.
Agentic AI Frameworks Like OpenClaw Spark Enterprise Security Crisis
Category: AI Safety & Cybersecurity
🚨 OpenClaw, the open-source agentic AI framework, has logged at least 454 vulnerabilities per the National Vulnerability Database, Dark Reading reported May 26; Gartner has advised enterprises to block downloads.
🛡️ Nvidia introduced NemoClaw as an enterprise-grade version with agent registration, governance, kernel-level isolation through OpenShell, and policy enforcement using Rego.
🤔 Agents move too fast for human-in-the-loop controls, leading to descriptions of “Formula One cars without brakes”; Cisco’s Defense Claw and Snyk Agent Security target visibility gaps.
NBA Plans Hawk-Eye Powered AI for Instant Out-of-Bounds and Possession Calls
Category: Industry Applications
📲 NBA Commissioner Adam Silver announced May 28 plans to deploy an automated AI and camera system for out-of-bounds and possession calls, reducing Coach’s Challenge use.
🛰️ The system uses Sony’s Hawk-Eye 3D optical tracking (partnered with the NBA since 2023) that follows ball and player movement at sub-second latency.
🏆 The announcement follows a disputed out-of-bounds call in the Thunder versus Spurs Western Conference finals Game 5; referees retain judgment on contact fouls.
DeepSWE Benchmark Crowns GPT-5.5 and Exposes 32% Error Rate in SWE-Bench Pro
Category: Testing, Evaluation & Benchmarking
📐 Startup Datacurve released DeepSWE on May 26, a 113-task benchmark spanning 91 open-source repositories and five programming languages, designed to better mirror real developer workflows.
🏆 GPT-5.5 scored 70 percent, sixteen points ahead of GPT-5.4 at 56 percent and Claude Opus 4.7 at 54 percent; Claude Haiku 4.5 collapsed to zero.
🔍 The audit found SWE-Bench Pro verifiers wrong on about one-third of trials, accepting incorrect implementations 8.5 percent and rejecting correct ones 24 percent of the time.
🔥 Blitz's Hot Takes
Let’s break down the play, because the league finally figured out it could not break down the call ...
Alright, sit down, because the National Basketball Association has just announced something I have been screaming into the void about for half a decade, and I am still mad. On May 28, Commissioner Adam Silver said the league plans to deploy an automated AI and camera system to handle out-of-bounds calls and possession decisions. It will use Sony’s Hawk-Eye 3D optical tracking, which the NBA has had as a partner since 2023, and spit out a call at sub-second latency. Translation for the people in the back: the ball will hit the floor, the cameras will already know, and you will not have to watch four officials huddle around a courtside monitor while everyone in the arena ages five years.
The trigger for this overdue heroism was a blown out-of-bounds call in Game 5 of the Western Conference finals between the Oklahoma City Thunder and the San Antonio Spurs. One bad swing of a possession arrow in a series of that magnitude, and suddenly the league finds the political will it has been allergic to since 2023. Tennis has been letting cameras call the lines since 2006. That is twenty years of seeing the future, and the NBA is just now lacing up its shoes. The clipboard is exhausted.
Here is the play I’m running on this: the league did not have a technology problem, it had a courage problem. Hawk-Eye has been sitting in NBA arenas for three seasons. The hardware was there. The latency was there. The tennis precedent was there. What was missing was the willingness to take a category of calls out of the hands of human referees, who are processing ten moving bodies and a ball at full sprint. The league waited for a postseason controversy embarrassing enough that doing nothing cost more than doing something. That is not strategy. That is reality running the two-minute drill while the league called timeouts.
Now here is the followup play I want called: keep going, and know where to stop. Out-of-bounds and possession are the layups of officiating. The next conversation has to be replay review itself, still a mood-killing trudge through grainy angles every time a Coach’s Challenge gets thrown. If the cameras are this fast and this accurate, shrink the whole review process, not just one play type. But do not let this creep into fouls. The judgment on contact, on the small stuff, on what is a foul and what is good defense, that is what makes basketball basketball. The chaos is the sport. Automate the bright lines, leave the gray ones alone, and let the refs do the part only humans can.
-- Blitz 🏀
📡 What's New With Your AI Tools
The AI tools you use every day are constantly evolving. Here's what changed and why it matters to you.
Claude (Anthropic)
Opus 4.8 released (May 28, 2026): Anthropic’s most advanced public model is now available everywhere at the same price as the previous Opus, arriving just 41 days after Opus 4.7. The upgrade most likely to matter for everyday use is reliability: early testers report it flags its own uncertainty more readily and makes fewer unsupported claims.
Dynamic Workflows (research preview): A new tool that lets Opus coordinate hundreds of AI subagents working in parallel on a single complex task. Anthropic says Claude Code paired with Opus 4.8 can now run large codebase migrations end to end, carrying hundreds of thousands of lines from kickoff to a finished, test-verified merge.
ChatGPT (OpenAI)
Codex & Mac App Workspace Overhaul: Building on the mid-month finance feature, OpenAI pushed a major update to its desktop ecosystem on May 21, with user adoption hitting its stride this week.
Appshots: Mac users can now use a hotkey to instantly snap an active desktop window, converting the screenshot and its visible text directly into a ChatGPT thread context.
Goal Mode: Now generally available across the Codex app, IDE extensions, and CLI, allowing you to set an end objective and success criteria so the model can work through multi-step coding tasks autonomously.
Locked Computer Execution: Eligible Mac “Computer Use” users can now keep longer agent tasks running securely in the background even after the machine locks.
Copilot (Microsoft)
The “Quieter” Microsoft 365 UI Redesign: Responding to heavy user feedback regarding intrusive UI elements, Microsoft announced a total design overhaul for Office apps on May 27.
Dynamic Action Button: Floating icons are being replaced by a context-aware button that shifts shape and utility based on the exact task you are doing in Word, Excel, or PowerPoint.
“Throw & Catch” Context Handoff: A new framework that smoothly transfers the active task context between inline canvas text, side panels, and full chat interfaces without losing your place.
Computer-Using Agents GA: In Copilot Studio (May 26), multi-step autonomous agents capable of interacting with desktop environments dropped their preview tags and are now generally available.
Gemini (Google)
Gemini 3.5 Flash GA (rollout expanded late May 2026 following May 19–20 I/O): Fast frontier model optimized for agentic, coding, and multimodal workloads; selectable in the Gemini app and via API for global users.
Gemini Omni multimodal model (late-May 2026 rollout): Generates and edits video from text, image, audio, or video prompts. Available to Gemini Advanced and Ultra subscribers in the mobile/desktop app.
Neural Expressive UI redesign (late-May 2026 rollout): Major interface overhaul with fluid animations, interactive visual responses (imagery, timelines, graphics) replacing plain text. Affects all Gemini app end-users for richer daily interactions.
Gemini Spark 24/7 agent (beta rollout late May 2026): Persistent personal agent for delegated multi-step tasks; initially available to Ultra-tier users.
Google Search Overhauled by Gemini 3.5 Flash: On May 26, Google officially flipped the switch to power its core global search engine entirely with Gemini 3.5 Flash. The traditional search box has essentially been retired in favor of a conversational interface by default. Users no longer have to toggle into a separate "AI Mode"—you can ask complex queries and instantly input conversational follow-up questions directly on the main results page.
Perplexity
No Major Changes This Week: Perplexity's core focus this month remains the stabilization of its massive mid-May rollouts, which introduced "Personal Computer" file-syncing on Mac and migrated the backend default orchestration engine over to GPT-5.5.
Grok (xAI)
Grok platform integrations (May 21–27, 2026): Native support added to Kilo Code (May 27), OpenCode (May 21), and related open-source agentic coding environments. Enables seamless use of Grok models within third-party developer tooling for power users and developers.
“Grok Build” Launch: On May 25, xAI rolled out Grok Build in early beta for SuperGrok Heavy tier subscribers. Operating entirely from the terminal, it functions as an agentic Command Line Interface (CLI). Developers can use natural language to map out code changes, edit files, and spin up subagents to execute complex programming pipelines in parallel.
