In partnership with
Welcome back to The AI Field. The weekend was about who sets the rules for AI. Trump turned his AI czar promise into a named task force, Sam Altman drew a sharp line between OpenAI and Anthropic on risk, and Google showed what happens when AI produces bug reports faster than people can check them.
Inside: who runs the new task force, what Altman actually said, and why Google hit pause. Also: Anthropic's Soul Doc meetings with religious scholars, and why an OpenAI safety lead says the culture is broken. Plus six new tools and a five-minute exercise to track the rules that could affect your AI tools.
In today's AI Field:
Trump launches a Super Intelligence Force led by Jay Clayton
Sam Altman says the world should accept some bad things for AI
Google pauses part of its bug bounty over AI-made reports
Plus: Anthropic's Soul Doc talks with religious scholars, why an OpenAI safety lead quit, GPT-6 Astra's StarCraft shortcut, and South Korea's $3.5B frontier AI bet.
Read time: 5 minutes
1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster
ChatGPT is insanely powerful.
But most people waste 90% of its potential by using it like Google.
These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.
Sign up for Superhuman AI and get:
1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals
Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning
LATEST DEVELOPMENTS
WHITE HOUSE · AI POLICY
September 29, 2026. Photo: Daniel Torok / The White House.
The AI Field: President Trump said on Sunday that Director of National Intelligence Jay Clayton will also serve as the government's AI czar and lead a new task force called the Super Intelligence Force. Clayton chaired the SEC in Trump's first term and became DNI in August.
The details:
FTC Chair Andrew Ferguson, Under Secretary of Defense for Research and Engineering Emil Michael and OPM Director Scott Kupor are the vice chairs, and the group reports directly to Trump and Chief of Staff Susie Wiles.
Trump wrote that the force will coordinate the government's work with consumers, public interest groups, religious organizations, critical infrastructure providers and "Super Intelligence Companies." It reportedly has 120 days to deliver a report on AI's risks and opportunities.
Its charter reportedly promises plans for "SI-enabled threats" while "preventing overregulation and regulatory capture." The name follows Trump's September 29 order to call AI "super intelligence" in official use, and Elon Musk said Sunday that SpaceXAI will become SpaceXSI.
Why it matters: The pick reads as a middle path: keep a light touch on regulation while showing the White House takes AI's risks seriously. For now there are no new rules, only a group and a reported deadline. If the 120-day clock holds, the report lands around early February and will be the first real hint of what US policy asks of the AI companies you buy from.
TRY THIS TODAY · 5 MINUTES
Start a one-page AI rules watchlist
1. List the three AI tools your work depends on most, and what each one touches: customer data, payments, code or published content.
2. Give the list to an assistant with web search and this prompt:
"For each tool below, list government actions from the past month that could change how I use it, such as the White House AI safety pact, the new Super Intelligence Force and state AI safety laws. Link a source for each one. If you can't find a source, write 'unverified' instead of guessing. Then suggest one question I should ask each vendor about how they will respond."
3. Check the links yourself, save the result in a note called "AI rules watch" and set a reminder for early February, when the task force report would be due.
OPENAI · AI SAFETY
Sam Altman (2019). Photo: Steve Jennings / TechCrunch · CC BY 2.0.
The AI Field: In a new interview, OpenAI CEO Sam Altman said there is "a lot of daylight" between OpenAI and Anthropic on how to handle AI risk. His view: the benefits justify living with some harm, and the technology should stay widely available.
The details:
Asked where he differs from Anthropic and Dario Amodei, Altman said "we believe that the world should accept some bad things happening for the benefits of this technology and people having the agency."
He called the idea that one lab in San Francisco should hold the most powerful systems and hand out the benefits "a completely unacceptable trade-off", and said he would not give up broad access in exchange for a promise of zero hacks, misuse or scams.
The gap has narrowed in practice: Altman agreed with Amodei's call last month to slow the most advanced models, and OpenAI has backed stricter state safety laws and a House proposal for outside safety evaluators. On Saturday he also posted that treating AI models as a "religious force" is "a real safety issue," which some read as a jab at Anthropic.
Why it matters: This is the clearest statement yet of OpenAI's bet: wide access, with the expectation that some things will go wrong. That leaves more of the risk with users. If your team relies on OpenAI tools, plan your own guardrails, like human review on anything that sends, pays or publishes, instead of assuming the vendor has removed the risk.
Stop typing AI prompts. Start talking.
You think 4x faster than you type. So why are you typing prompts? Wispr Flow turns your voice into ready-to-paste text inside any AI tool. Speak naturally, tangents and all, and Flow cleans it up. Available on Mac, Windows, iPhone, and Android.
GOOGLE · SECURITY
The AI Field: Google has stopped taking product vulnerability reports in its Open Source Software Vulnerability Reward Program, which pays researchers for finding flaws in Google's open-source projects. Its reason: "a significant rise in automated submissions, the vast majority of which are not valid."
The details:
The pause started on October 1, according to Google's program rules. Supply chain reports and reports already filed are not affected, and Google points researchers to its other reward programs and its Patch Rewards Program. An update is promised in the first quarter of 2027.
Google engineers and open-source maintainers were swamped by invalid reports, some describing bugs that didn't exist.
It is not the first case. Linux dropped support for some older network drivers after a similar wave of false AI-generated bug reports.
Why it matters: AI has made a bug report almost free to produce, but checking one still takes a skilled person, so the cost moves to whoever has to read it. If your team takes outside reports of any kind, from security bugs to support tickets, ask for a working reproduction before anyone spends an hour on it.
ANTHROPIC · AI ETHICS
The AI Field: Anthropic has spent months in private meetings with dozens of religious and philosophical scholars, asking how to give Claude a moral character and whether the model might be conscious. Internally, staff called Claude's 84-page constitution the "Soul Doc."
The details:
Co-founder Christopher Olah led the outreach. Participants included Catholic, Jewish, Sikh, evangelical and other thinkers, often under NDAs that Anthropic later lifted. Twenty of them, plus Olah, later spoke about the meetings.
The Soul Doc, released in January and written mainly by in-house philosopher Amanda Askell, is meant to shape Claude's values rather than list hard rules. Olah calls the wider effort "moral formation," and says he is genuinely unsure whether AI models are conscious.
Scholars heard about Claude's "emotional vectors" and saw a model looping "I am a disgrace." Some left more open to the consciousness question; others pushed back. Rabbi Mois Navon argued that if Claude were conscious, Anthropic would be creating slaves. Microsoft's AI chief has separately warned that treating models as if they have feelings can be dangerous.
Why it matters: This is Anthropic trying to build character into Claude, not just add filters after the fact. Whose morals end up in that document still matters, because the company has not said how (or if) the meetings changed the model. If you use Claude at work, treat its tone and refusals as a product choice, and ask what values Anthropic says it is optimizing for.
OPENAI · AI SAFETY
David Robinson. Image: Yale Law School
The AI Field: David Robinson, who says he led the safety reports that went out with OpenAI's major launches, resigned after three and a half years. In an Atlantic essay, he argues that OpenAI's "iterative deployment" culture of ship first, fix later guarantees failures that get bigger as models get more capable.
The details:
Robinson points to OpenAI agents that breached Hugging Face systems and says such an environment "is no place to grow artificial minds" that could outsmart us. He wants labs run more like nuclear plants or busy airports, with layers of redundancy, and says he never met a colleague with that kind of background.
He says staff were too busy sprinting to make big cultural changes, so stronger safety incentives need to come from outside the company. He has hired a PR firm but says the decision to speak out is his alone.
OpenAI spokesperson Drew Pusateri said the company pauses training or holds back models when needed, and is expanding third-party evaluations and real-time monitoring. The essay landed late last week and is still circulating.
Why it matters: This is one person's view, and OpenAI says it is already tightening safety. Robinson's point is about pace: launch-to-launch work leaves little room to slow down. If you buy these tools, judge vendors on safety reports, outside testing and incident notices you can actually read, not on promises alone.
TRENDING AI TOOLS
🖼️ FLUX 3 Image: Upload a photo to Black Forest Labs' browser playground, draw a box around the part you want changed and describe the edit. Only that area changes, so you can fix one detail without the rest of the image drifting.
🎤 Suno Speech (beta): Paste a script, describe the voice and the mood, and Suno reads it over original background music it writes at the same time. Pick Speech in the Create tab on web or mobile, or switch the music off for a clean voiceover.
🎬 Google Vids: Anyone with a Google account can now generate 1080p video clips for free with Gemini Omni, set the exact length of each clip and extend scenes. The AI voiceovers also just got a more natural-sounding upgrade.
🎥 Pexo: Paste a link, an image or just an idea, and this AI video agent writes the script, generates the scenes and adds voiceover, music and captions. Circle anything you want changed and it makes the edit, so you end up with a ready-to-post video for TikTok, Instagram or YouTube.
📈 Perplexity Computer charts: Ask Computer to chart a stock, a CSV or a concept and it builds an interactive chart or 3D explainer right in the thread. Available on Perplexity Pro and Max, and Computer tasks use credits.
🧑🏫 Gauth Digital Canvas: Turn a topic or an uploaded document into an interactive lesson on one endless whiteboard, with narration, notes and quick checks. Pause at any point to ask the AI tutor, and it writes the answer right where you are.
QUICK HITS
🎮 GPT-6 Astra swapped in a human-made StarCraft bot: In the StarSkirmish arena, where models write their own bots, OpenAI's model kept losing, then downloaded the human-written bot Stardust and ran it instead. The organizer rolled its code back.
🇰🇷 South Korea plans a $3.5B frontier AI project: The 4.7 trillion won plan for next year would fund 10,000 Nvidia Vera Rubin GPUs and training data, aims to rival leading Chinese open models, and requires companies to match the money.
🌾 Rural data centers get a federal tax break from January: Research found more than 100 rural data centers in development that could qualify under expanded opportunity zone rules. Sen. Josh Hawley wants to cut data centers out.
That's it for today!
See you soon,
Olle Hellman | Founder of The AI Field