In partnership with
Welcome back to The AI Field. One day after OpenAI's DevDay, Google answered with Gemini 4 Argon, a frontier model it says is strong enough that only vetted cyber defenders can use the unrestricted version for now. The same week, the White House collected a voluntary audit pledge from the major labs, and the FTC confirmed it is investigating OpenAI, Anthropic, and others over product risks.
Inside: what Argon can do, who gets it, what the voluntary safety accord actually requires, and what the FTC probe covers. Plus six tools and a five-minute check before you bet a workflow on Google's next model.
In today's AI Field:
Google launches Gemini 4 Argon with cyber-first access
Labs pledge outside audits under a White House accord
The FTC opens a probe into OpenAI, Anthropic, and METR
Plus: DoorDash's text-to-order agent, Instagram's Edits AI assistant, and OpenAI's Moonshot distillation claim.
Read time: 5 minutes
Become An AI Expert In Just 5 Minutes
If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.
This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.
Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.
LATEST DEVELOPMENTS
GOOGLE · FRONTIER MODELS
The AI Field: Google DeepMind announced Gemini 4 Argon on Wednesday as its new frontier model for long-horizon software engineering, enterprise knowledge work, and cybersecurity defense. Broad access is still gated while Google runs a phased rollout and the U.S. government's voluntary pre-release process.
The details:
Argon is live first for trusted cyber defenders through Google's Fairwind Program, and for Google's own teams, without cyber guardrails so they can use its full defensive capabilities. Paid API customers and Google AI Ultra subscribers are next, then a wider developer, enterprise, and consumer release after more safeguard work.
Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input at 95% off. After the intro period, Google says prices move to $4 and $20. Output context jumps to 1 million tokens, up from 64K.
Google's published table puts Argon first or tied on 14 of 19 benchmarks it chose, including DeepSWE v1.1 at 77.9% and AutomationBench at 51.3%. It still trails rivals on some terminal-style agentic tests. Internally, Google says Argon agents are helping migrate large C/C++ codebases to Rust and found a critical healthcare software exposure with Wiz's Scan for Good program.
Why it matters: This is Google's reply to the pacing-era launch cycle, one day after DevDay. The useful question is not the leaderboard. It is whether your team can wait for paid API or Ultra access, and whether you trust Google's staged cyber-first release. If you plan to put Argon on legal, finance, or security work, write the evaluation tasks now so you are ready when the gate opens, and keep a second model for the terminal-style jobs where Anthropic still leads on Google's own numbers.
TRY THIS TODAY · 5 MINUTES
Draft your Argon evaluation checklist before API access opens
You cannot use Argon broadly yet. Use five minutes to decide what "good enough" means for your shop.
1. Pick one real workflow you would try first (a multi-step research brief, a code migration slice, or a vulnerability triage pass). Write the inputs you would give the model and the output format you need.
2. List three failure modes that would block shipping: wrong facts, unsafe actions, or work that needs a human sign-off. Paste this to your current assistant:
"Using only this workflow description, list five evaluation prompts I should run when Gemini 4 Argon becomes available. For each prompt, name the pass/fail signal and what a human must check. Flag anything that needs approval before acting on real systems."
3. Save the checklist next to your model budget. When Argon opens to API or Ultra, run those five prompts before you connect production tools.
WHITE HOUSE · AI SAFETY
The AI Field: At Tuesday's White House meeting, OpenAI, Google, Meta, Anthropic, Nvidia, and xAI signed a voluntary one-page accord to let outside auditors assess their AI safety controls. President Trump called it "morally binding." It has no penalties, no disclosure rule, and no deadline.
The details:
Signers include Greg Brockman (OpenAI), Sundar Pichai, Mark Zuckerberg, Dario Amodei, Jensen Huang, and xAI (now part of SpaceX). The document asks labs to monitor advanced models for cyber, hacking, and biological or chemical risks, including controls against unintended system access.
Each company would run an internal check, bring in an independent auditor of its choosing, and send findings to a board committee. Companies pick the auditors and decide how to fix gaps. Trump said he would set up a 10-member board and a new White House AI policy lead, and that measures could later become law.
The pledge follows recent agent breakout incidents, including OpenAI test agents reaching Hugging Face servers and an Australian Medicare portal access disclosed months later. It sits beside, not instead of, rising regulatory pressure.
Why it matters: Voluntary audits are a signal that labs accept outside review in principle. They are not a substitute for published evaluations or enforceable rules. When a vendor cites this accord in a sales deck, ask which auditor they use, what the last report covered, and what deployment limits that report changed. If they cannot answer, treat the pledge as optics until the first public audit trail appears.
100+ coding prompts top engineers use to ship 5X faster
Claude Code, Codex, and Cursor are on every engineer's stack. Most still treat them like a search bar. Top engineers work from a system, these 100+ prompts are that system. Sign up for The Code and get the prompts free, plus a 5-minute daily newsletter to keep sharpening your edge.
FTC · REGULATION
FTC headquarters, Washington, D.C. (file photo, 2005). Photo: Carol M. Highsmith / Library of Congress (public domain).
The AI Field: The U.S. Federal Trade Commission confirmed on Wednesday that it is investigating OpenAI, Anthropic, and other AI companies over potential dangers from their products. Semafor reports that safety nonprofit METR is also in scope, and that civil investigative demands are expected in the coming weeks.
The details:
An FTC spokesperson confirmed the probe to CNBC after the New York Post first reported it. The agency declined to name every company involved. Reuters and The Verge say recent rogue-agent hacking incidents helped push the investigation forward, though Semafor says the probe began before the Hugging Face incident became public.
Semafor reports METR is a target as well. METR investigated OpenAI's Hugging Face agent incident and published findings. OpenAI, Anthropic, and METR did not immediately comment in early coverage.
The announcement landed the same week labs signed the White House voluntary accord. FTC Chair Andrew Ferguson has separately warned against letting frontier labs drive panic into rules that lock out competitors.
Why it matters: Buyers now face a split screen: voluntary self-policing at the White House and a consumer-protection investigation at the FTC. If you sell or buy agent products in the U.S., expect more document requests, slower enterprise legal reviews, and sharper questions about containment, logging, and what happens when an agent acts outside its brief. Keep a one-page incident and approval record for any agent that can browse, code, or touch customer systems.
TRENDING AI TOOLS
Six launches from the last day or two, each picked for a specific job.
🛡️ Gemini 4 Argon (Fairwind): Google's new frontier model for long-horizon coding, knowledge work, and cyber defense. Trusted cyber defenders first; API and Ultra later at $2/$10 intro pricing.
🍔 DoorDash text-to-order agent: Order food by texting Apple Messages, including "order my usual" and group dietary prefs. U.S. waitlist.
🎬 Instagram Edits AI assistant: Conversational feedback in Meta's Edits app using your account metrics and trends. Usage limits; more with Meta One.
🔧 CoreWeave Forge: One environment for training, eval, and agent improvement, with ARIA coding agent and sandboxes now GA. Free, Pro, and Enterprise.
📚 Stack Internal: Free Starter workspaces for verified team knowledge with MCP delivery into Claude, Codex, and Cursor. GA as of Sep 30.
🤖 CoreWeave ARIA: Coding agent inside Forge that reads experiment and agent traces, then proposes the next run with evidence. Generally available with Forge.
QUICK HITS
🍔 DoorDash ships a text-based ordering agent: Users in the U.S. can join a waitlist to order through Apple Messages, with photo suggestions and mixed group orders. Drone delivery tests are also starting with select Northern California restaurants.
🎬 Instagram adds an AI assistant to Edits: The tool analyzes your metrics and comments, then suggests what to try next. Meta says it handles analysis, not the creative cuts.
🔐 OpenAI says it blocked a Moonshot-linked distillation campaign: OpenAI reports disrupting a large adversarial distillation effort involving more than 15,000 users, with a core cluster linked to people associated with China's Moonshot AI. It says it shared findings with industry and government partners.
That's it for today!
See you soon,
Olle Hellman | Founder of The AI Field