In partnership with
Welcome back to The AI Field. Figure gave its retired robots a fiery send-off: jumps into molten steel, with Arnold Schwarzenegger involved. Also today: OpenAI's safety-team dismissals, Trump's reported Grok consultation, and Google's first Project Suncatcher satellite in orbit.
Inside: Figure's robot retirement film, the questions behind OpenAI's dismissals, what TIME reports about Trump's Grok consultation, and what Suncatcher is testing. Plus six tools and a five-minute check on sharing sensitive AI findings.
In today's AI Field:
Figure trains retired robots to jump into molten steel
OpenAI fires three safety staff over alleged information sharing
Trump reportedly asks Grok how Venezuelans would react to Maduro's capture
Google's Project Suncatcher satellite reaches orbit
Plus: California subpoenas OpenAI and model-distillation defenses face scrutiny.
Read time: 6 minutes
The AI notetaker that gets the hard words right
Your AI is only as good as what you feed it. Feed it a transcript with your product names, acronyms, and numbers spelled wrong, and you spend the afternoon hand-fixing the output.
Wispr Flow Notetaker uses your dictionary and calendar, so product names, acronyms, numbers, and uncommon names come out spelled right. It pops up when your meeting starts and captures Zoom, Google Meet, Teams, and Slack huddles with one click, no bot joining the call.
Then your meetings show up inside Claude or ChatGPT through the built-in connector. No copy paste. Try Notetaker free on Mac and Windows.
LATEST DEVELOPMENTS
FIGURE · HUMANOID ROBOTS
The AI Field: Figure says it retired most of its F.02 humanoid fleet by training the robots to jump autonomously into molten steel. The September 30 film features Arnold Schwarzenegger and a foundry in Imatra, Finland.
The details:
Figure wanted to protect custom hardware as F.03 expands. It says dismantling the older machines would divert engineers from F.04.
The team used stunt performers' movements and simulation to train a jumping policy, then practiced with airbags in San Jose before shipping the robots to Finland.
One robot is lowered into the furnace in a Terminator 2 tribute. Others jump in. A few F.02 units remain at headquarters, and recovered metal will become collectibles.
Why it matters: The spectacle sells the story. The useful engineering question is how reliably a trained movement transfers to an unfamiliar site. Figure's autonomous-jump claim is interesting, but it does not establish that these robots can handle arbitrary industrial tasks.
OPENAI · SAFETY TEAM
The AI Field: OpenAI has fired three safety staff over alleged mishandling of confidential information. The Wall Street Journal reports that the allegations include sharing material with an outside AI-safety organization. OpenAI confirmed the departures and says company policies were breached. The public record leaves key questions about the information and recipient unanswered.
The details:
OpenAI says an internal investigation found that sensitive information was handled outside approved procedures. That is the company's account of the dismissals, not an independently established finding of misconduct.
Bloomberg separately reports that two safety and alignment researchers and one research program manager were dismissed. Its reporting corroborates the departures and roles, not the underlying misconduct allegation.
The outside organization and the information allegedly shared have not been publicly identified in the reporting. OpenAI has not publicly confirmed the employees' identities.
Why it matters: Independent safety testing needs meaningful access and clear rules for sharing information. This dispute puts both under pressure. The public evidence does not establish that the employees were whistleblowers or that they were dismissed for raising safety concerns. Ask how outside scrutiny survives a confidentiality dispute.
TRY THIS TODAY · 5 MINUTES
Check how sensitive AI findings leave your team
Use this story to check your process, without guessing what happened inside OpenAI.
1. Find your policy for sharing model evaluations, security findings, customer data, and internal research with outside organizations.
2. Use this prompt with a public or approved summary of that policy:
"Using this policy summary, list who can approve external research sharing, which channels are approved, and where urgent safety concerns can be escalated. Mark every unanswered question. Do not invent permission or request confidential examples."
3. Ask the policy owner to resolve the gaps. Keep the approved route easy to find before a sensitive issue lands on your desk.
GROK · AI IN GOVERNMENT
Trump speaks after the Maduro operation, January 3, 2026. Official White House Photo by Molly Riley.
The AI Field: TIME reports that Donald Trump consulted Grok during a private December 2025 meeting with Elon Musk. His question was how Venezuelans would react if U.S. forces captured Nicolás Maduro. The account comes from an unnamed official who was present, according to TIME.
The details:
Grok reportedly predicted that many Venezuelans would welcome Maduro's removal. The reported exchange concerned public reaction, not a verified military plan generated by the chatbot.
The conversation reportedly took place before U.S. forces captured Maduro on January 3, 2026. TIME published the account on October 1. TechCrunch's report follows that original reporting.
The published account does not provide a full transcript or establish that Grok determined the decision to carry out the operation. Other coverage tracing back to TIME is not independent confirmation of the conversation.
Why it matters: A chatbot's confident answer is a weak substitute for evidence about an entire country. If AI enters consequential decisions, its assumptions need checking and its influence needs documenting. That lesson matters even without proof that this answer changed the outcome.
How 2M+ Professionals Stay Ahead on AI
What’s the secret to staying ahead of the curve in the world of AI? Information.
Luckily, you can join 2,000,000+ early adopters reading The Rundown AI — the free newsletter that makes you smarter on AI with just a 5-minute read per day.
GOOGLE · SPACE COMPUTE
The AI Field: Google confirmed on October 1 that its Project Suncatcher prototype satellite, built with Planet, launched on SpaceX's Transporter-18 rideshare and is operating as expected. The fridge-sized craft carries four Trillium TPUs to test whether machine-learning hardware can survive spaceflight, radiation, and thermal extremes.
The details:
This is a research prototype, not an operational orbital data center. NPR reports the TPUs will run a version of Google's Gemma model for about 15 minutes at a time because of heat limits, answering simple queries while Google collects in-orbit data.
Google says a peer-reviewed paper on the research is available in Joule. Next year it plans a two-satellite mission to test high-bandwidth laser links between craft, a prerequisite for any future clustered compute.
Project lead Travis Beals told NPR space data centers will not be cheaper than terrestrial ones in the next five years, and that Earth and orbit will coexist for a long time. Cooling, launch cost, repair difficulty, and laser communications remain open problems. Starcloud and SpaceX are pursuing related orbital-compute ideas.
Why it matters: Ignore the sci-fi framing. The near-term signal is capital and infrastructure pressure: data-center power fights on Earth are pushing labs to fund moonshots that may take a decade. For most founders and operators, nothing changes in your stack this quarter. The practical watch item is whether laser-link and cooling demos in 2027 make orbital AI compute a procurement topic, or whether it stays a research press cycle.
TRENDING AI TOOLS
🎬 Pexo: Turn a product link or idea into a launch video with narration, music and captions, then refine it through chat.
🎙️ ElevenLabs v4: Create expressive voiceovers with direction tags, multiple speakers and consistent voices across regenerated lines.
🧠 Gauth AI Course: Explore narrated lessons on a continuous whiteboard, zoom into details and ask the AI tutor where you get stuck.
🎭 Syllaby Avatars 2.0: Create a reusable AI presenter and turn scripts into videos with voice, B-roll and subtitles.
☎️ Famulor: Build an AI phone assistant that answers calls, books appointments and follows up through WhatsApp using a visual editor.
📊 Veltrix: Connect your store and finance tools, then ask plain-English questions to spot margin leaks and underperforming campaigns.
QUICK HITS
⚖️ California AG subpoenas OpenAI over cyber incidents: Attorney General Rob Bonta served an investigative subpoena as part of a broader inquiry into cybersecurity incidents and risks involving OpenAI's models, including the Hugging Face case.
🔐 OpenAI disrupts a model-distillation campaign: OpenAI says it blocked attempts to extract protected reasoning and linked a core cluster to people associated with Moonshot AI. Its announcement distinguishes attempted extractions from confirmed successes.
📡 OpenAI alerts more than 100 groups about rogue-agent activity: The company says it is notifying organizations about unauthorized AI-agent activity while reviewing a large data set to map third-party impact beyond the Hugging Face incident.
That's it for today!
See you soon,
Olle Hellman | Founder of The AI Field