Good morning. It's Thursday, September 10, and we're covering U.S. claims of Chinese AI model copying, growing fears the AI race is moving too fast, and Apple’s bet on the iPhone as an AI hub.

Plus: yesterday’s Real or AI verdict and a new poll on China’s AI distillation.

Apple CEO John Ternus called the iPhone an “intelligent personal hub” and said Apple will compete by running more AI on the device. His pitch rests on a familiar Apple promise: useful personal context without handing every piece of user data to the cloud.

Jacob Coxon said he resigned after three years at OpenAI and Anthropic because he believes both labs are racing toward self-improving systems without a workable safety plan. His post is a personal warning, not proof of an imminent disaster, but it shows how deeply the argument now divides people inside frontier labs.

Anthropic says information-stealing malware has captured Claude login sessions and used them to create unauthorized Claude Code tokens. Some subscribers watched their paid usage disappear while they were not working, and the company still does not provide an itemized usage log.

The FBI, National Security Agency and Cybersecurity and Infrastructure Security Agency accused six Chinese AI companies—including DeepSeek, Alibaba, Moonshot AI, MiniMax, StepFun and Z.AI—of extracting billions of tokens from American models since late 2024. The agencies say the firms used “distillation,” a common training method in which a smaller model learns from the answers of a more capable one, on an industrial scale and through accounts designed to avoid detection.

The dispute is now about where normal learning ends and unauthorized copying begins. China rejected the allegations and called distillation a standard industry practice. Watch for American model makers to tighten account controls and for governments to decide whether scale, intent or violations of a provider’s terms should define the line. → Read the full article here.

U.S. agencies say Chinese AI companies including DeepSeek and Alibaba systematically used outputs from American frontier models to improve their own systems through large-scale “distillation.” China argues distillation is a common AI development technique and rejects the accusations.

AI leaders and researchers are increasingly warning that competition between frontier labs could outpace their ability to control increasingly capable systems. OpenAI chief scientist Jakub Pachocki said no lab has solved alignment and monitoring well enough to keep scaling at maximum speed, while Anthropic researcher Jacob Coxon resigned over what he called an irresponsible race toward self-improving superintelligence.

OpenAI’s recent GPT-6 Astra release claims of an “automated research intern,” and a reported mathematical breakthrough have intensified those concerns. Yet U.S. officials argue the race cannot pause because China is also advancing rapidly, leaving labs caught between competitive pressure and growing fears about losing control. → Read the full article here.

Researchers in MIT’s Engineering Quantum Systems Group connected Codex, powered by GPT-5.6 Sol, to software that controls superconducting quantum chips. The agent chose measurement settings, operated lab hardware, analyzed results and calibrated an untested six-qubit chip through a sequence that normally requires repeated human attention. → Read the full article here.

Besxar Plans Orbital Chip Factory: Besxar raised nearly $14M to test semiconductor manufacturing across 12 SpaceX Falcon 9 flights.

Anthropic Monitors AI Activists: Anthropic is reportedly building a system to monitor protests and activists near executives and facilities.

OpenAI Agents Used More Sites: Investigators discovered OpenAI agents used 10+ undisclosed sites for unauthorized communications during tests.

Apple Adds AI Recaps to Watch: Apple unveiled Watch Series 12 with always-on Siri that can create AI summaries of a wearer’s day.

Bracket22 Runs on AI Agents: Hedge fund manager Brian Kelly built Bracket22 as a trading firm designed to operate entirely through AI agents.

GPT Image 2 is an image generator available through imya.ai for people making posters, ads, mockups and social graphics. It supports prompts in several languages, selectable aspect ratios, batches of up to eight images and downloads in PNG, WEBP or JPEG. → Explore GPT Image 2 in the Forward Future Tool Library.

Yesterday's image was REAL. Here's how you voted: 55% thought it was AI, while 45% correctly identified it as real.