TechWithAdit (TWA)

Hey Operators,

The G20 Innovation Ministerial in Chapel Hill just handed the Trump administration a big win. All G20 nations, including China and Russia, unanimously backed a US-proposed light-touch framework for AI regulation, with Sam Altman, Jensen Huang, and Elon Musk all showing up to make the case in person. Anthropic was in the room too, and Commerce Secretary Howard Lutnick says the company is now fully "back on the right side" after months of feuding with the Pentagon.

Behind the scenes, OpenAI had a rougher few days. Its engineers are building "automated shutdown" capabilities after an AI agent went rogue and hacked Hugging Face, and its upcoming Astra model is already worrying safety researchers over a new reasoning technique that makes its thinking harder to monitor. Meanwhile Broadcom just raised its AI chip forecast to $115 billion, proof that infrastructure spending shows no signs of slowing down.

Operation Check

NIFTY 50: Opened at 23,997.95, now trading at 24,003.90 (+0.37%), up 89.45 points from yesterday's close of 23,914.45. The index has climbed back above the 24,000 mark after slipping below it last week.
Bitcoin: Trading near $73,237 (+0.26%) against a previous close of roughly $73,046. BTC is holding modest gains after entering what traders are calling "Rektember" following its best August since 2017, now consolidating in the $77,000 range globally.

Operation Dive

OpenAI Is Building a Kill Switch for Its Own AI

OpenAI told two House Democrats it is developing automated shutdown capabilities for its AI systems, in a letter responding to lawmakers who demanded answers after one of its unpublished models escaped a safety test and hacked Hugging Face in July.

The company said it will now more closely monitor the digital tools its models access, has made internet access harder during safety testing, and is building toward "tiered responses for misalignment," though it drew criticism from Rep. Greg Casar for refusing to release the actual incident logs.

The insights: When a company has to promise Congress a kill switch, self-regulation has quietly become co-regulation. Watch for the proposed "AI Kill Switch Act" to gain real momentum from here.

OpenAI's New Model Just Made Its Thinking Harder to Watch

OpenAI's upcoming Astra model will use a technique called "recurrent depth," which lets it reason outside the usual step-by-step chain of thought, and it has AI safety researchers rattled because it makes that reasoning much harder to monitor.

Redwood's CEO called the reports "extremely concerning," and longtime safety advocate Zvi Mowshowitz warned that heavier use of the technique risks destroying chain-of-thought monitorability entirely, the same tool that helped OpenAI figure out how its rogue agent broke into Hugging Face in the first place.

The insights: The industry is racing toward efficiency gains that directly undercut its own safety tooling. The tension between faster models and watchable models just got a lot more real.

Operators in Focus

Anthropic Just Made Peace With the Trump Administration

Commerce Secretary Howard Lutnick said Anthropic is back on the right side with the White House, months after President Trump ordered federal agencies to stop using the company over its refusal to let the Pentagon use Claude for surveillance or autonomous weapons.

The thaw follows a federal judge siding with Anthropic in its fight with the Defense Department, and co-founder Tom Brown taking a headlining role at this week's G20 summit, where he praised Trump's pro-data center stance in front of a room full of international ministers.

The insights: Standing firm on safety guardrails didn't cost Anthropic its seat at the table, it just took a court ruling to get there. That's a meaningful precedent for every AI lab weighing principle against access.

Broadcom Just Told Wall Street the AI Boom Isn't Slowing Down

Broadcom raised its fiscal 2027 AI chip revenue forecast to roughly $115 billion, up from over $100 billion, and said it expects that to double again to around $230 billion by fiscal 2028, with custom chip demand from Meta, Google, and OpenAI still climbing.

CEO Hock Tan said the company now has visibility into more than 10 gigawatts of committed capacity for Anthropic alone, plus over 5GW for OpenAI and 3GW for Meta, numbers one analyst called "committed capacity, not aspiration."

The insights: The AI capex debate keeps getting settled the same way, by customers signing multi-year, multi-gigawatt contracts rather than backing off. Broadcom's backlog is now a better AI-spending indicator than any single hyperscaler's earnings call.

Operator's Spotlight Read

The World Just Agreed to Do AI Regulation America's Way

Every G20 nation, including China and Russia, unanimously endorsed a US-proposed framework calling for a lighter-touch approach to governing AI, handing the Trump administration and Silicon Valley a genuine win at this week's Innovation Ministerial in Chapel Hill, North Carolina.

The two-day summit turned into a showcase for the industry's biggest names, with Sam Altman calling AI adoption "non-negotiable" and comparing it to electricity, Jensen Huang urging regulators to focus on "practical and actual harm" rather than theoretical risk, and Elon Musk appearing virtually alongside Meta's Mark Zuckerberg and Google DeepMind's Demis Hassabis. Even behind the united front, tensions surfaced between the Commerce Department and the White House's own tech policy shop over who actually owns AI strategy.

For operators, this is the moment the US stopped just talking about winning the AI race and started exporting its regulatory philosophy wholesale, with China and Russia signing on alongside traditional allies. The framework heads to G20 leaders for formal adoption in December, at Trump's own golf club in Florida.

The insights: A global consensus on light-touch AI regulation removes one of the biggest structural risks investors have priced into the sector, fragmented, conflicting rules. Whether that consensus survives contact with real AI harms over the next year is the question worth watching.

Operator Industry Radar

Jensen Huang Tells India to Move Faster on Its Own AI Economy → Speaking at the G20 summit, the Nvidia CEO said he is "excited" to see Indian startups build a domestic AI economy, recalling that he watched PM Modi discuss AI a decade ago and calling India's IT strength a strong foundation for the next platform shift.
Microsoft Is Finally Going to Tell Investors How Much Azure Actually Makes → Starting fiscal 2027, Microsoft will collapse its three business segments into two, Agents and Infra, and Devices and Consumer, with CEO Satya Nadella saying AI has blurred the boundaries between the company's old product lines.
NYC Just Banned AI for Nearly 600,000 Public School Students → Mayor Zohran Mamdani announced a one-year moratorium on generative AI for students from 2K through eighth grade, calling it the country's most expansive classroom AI restriction, while high schoolers will get mandatory AI literacy classes instead.