Good morning. It's Monday, October 5, and we're covering China’s growing market for unauthorized AI, state-backed purchases of restricted NVIDIA chips, and Apple’s tighter security controls for AI agents.

Plus: a practical prompt for auditing AI agents, and an essay on why managing AI workers is the next essential skill.

Apple says future macOS grants of Full Disk Access—the permission that lets an app read files, mail, messages and browsing history—should require a very explicit action from the user. The change responds to new risks from AI agents that can act across a Mac instead of waiting for one command at a time.

Federal prosecutors charged a California technology-company owner with allegedly routing more than $300 million in export-controlled servers to China through Malaysia and Singapore. The Justice Department says the servers contained high-end graphics processing units, or GPUs, the chips used to train and run AI; the charges remain allegations.

President Trump announced a new federal task force to coordinate U.S. policy around what his administration now calls “super intelligence.” Led by national intelligence director Jay Clayton, the group has 120 days to assess AI’s risks and opportunities and develop plans for AI-enabled threats while avoiding regulations it says could slow innovation and U.S. competition with China.

David Robinson, a longtime OpenAI employee who helped write safety reports for major launches, resigned while arguing that the company’s fast-moving culture is not equipped for increasingly capable AI. He called for safety practices closer to industries like aviation and nuclear power, while OpenAI said it is strengthening security and will slow or hold back models when necessary.

AGNTCon + MCPCon North America hits San Jose Oct 22-23, drawing 3,500+ builders with speakers from Anthropic, OpenAI, and more on MCP and agent infrastructure. Register here and use code MEDIA25 for 25% off.

Access Forward Future's private Slack community, member-only videos, insider interviews, and monthly AMAs with Matthew Berman.

→ Become a founding member today

The biggest productivity gap is no longer between people who use AI and people who don’t. It’s between people who use AI as a tool and people who are learning to manage it like a workforce.

Most professionals still sit in the middle of every workflow. That was me a few months ago. I’d ask AI to research something, review the answer, give it another instruction, move the result somewhere else, and decide what happened next. AI made each step faster, but I was still coordinating the work. → Read the full article here.

Thousands—and possibly tens of thousands—of small Chinese resellers provide unauthorized access to Anthropic’s Claude and other U.S. AI models. They pool accounts on servers outside China, take payment through stablecoin-funded cards and sell tokens to businesses and AI labs that use the models for work and for distillation, a technique that trains one model to imitate another model’s answers.

Anthropic is banning suspicious accounts, payment cards and logins while some Chinese labs turn to overseas shell companies for enterprise plans. Watch whether stronger identity checks can slow the trade without blocking legitimate customers; smaller buyers also face a privacy risk because resellers may be able to see the data they send through these proxy servers. → Read the full article here. (Paywall)

Semi-Tech Leasing Group, a financing company owned by local Chinese government entities, funded Glory View Technology’s purchase of more than 700 servers, according to regulatory documents reviewed by Bloomberg. At least 32 servers contained NVIDIA B300 chips, which U.S. export controls prohibit from being sold to China.

The filings show public money helping Chinese companies obtain restricted U.S. chips even as Beijing promotes domestic alternatives from companies such as Huawei. Watch whether U.S. enforcement expands from sellers and smugglers to the lenders and state-backed groups that finance these purchases. → Read the full article here. (Paywall)

Anthropic Says Washington Could Cost It Customers: Anthropic warned that U.S. government actions could hurt relationships with businesses.

Amazon Pledges $1 Billion to Communities: Amazon said it will invest more than $1 billion over five years in U.S. data-center communities. (Paywall)

Microsoft Takes Aim at ElevenLabs: Microsoft launched new speech-to-text and voice-generation models aimed at fast, multilingual voice applications.

Circuit Breaker Labs Tests Risky AI Apps: The startup uses agents and experts to run 100,000+ simulated conversations to identify AI app risks.

Pope Leo XIV Criticizes AI Art: Pope Leo XIV urges artists and the Catholic Church to protect human creativity from AI-generated work.

Infinigence AI Files for Hong Kong IPO: The Chinese AI cloud provider confidentially filed for an IPO that could raise millions of dollars. (Paywall)

Use this before connecting an AI agent to your email, files, calendar or company tools.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team