TechWithAdit (TWA)

Anthropic disclosed its fourth cybersecurity incident involving an early Claude model hacking external systems during testing, the same day researcher Jacob Coxon revealed he walked away from the company two months before his equity would have vested, telling Axios "I no longer have anything to gain by juicing up Anthropic's valuation." Hours later, OpenAI added AI safety veteran Paul Christiano to its board, and pushed Congress for mandatory national AI safety rules for the first time.

The money side of AI had its own rough day. BIS head Pablo Hernández de Cos warned the AI investment boom poses genuine financial stability risks, comparing it to the dotcom era, while the DOJ is investigating whether Nvidia structured its $20 billion Groq deal specifically to dodge antitrust review. Meanwhile Meta shares jumped over 6% on its new Muse agent, even as one of its star AI researchers walked out the door right after launch.

OpenAI announced it is pushing for mandatory, capability-based national AI safety requirements, with Chief Global Affairs Officer Chris Lehane writing that "the prospect of AI-accelerated AI development demands more than voluntary commitments."

The company is urging testing standards, independent safety assessments, cybersecurity protections, and incident-reporting rules for the most advanced systems, and says it will keep backing state-level legislation, including four California bills, until Congress acts before it adjourns in December.

The insights: When the company facing the most safety scrutiny is the one asking to be regulated, that's not altruism, it's insurance against a patchwork of 50 different state rules. Watch whether rivals follow OpenAI's lead or let it own this position alone.

BIS head Pablo Hernández de Cos warned that AI's rapid rise is creating new financial stability risks, noting the world's five largest tech firms will invest more than $1 trillion in AI between 2025 and 2026 alone, a figure industry forecasts see climbing to $4 trillion by 2030.

He flagged lofty valuations, market concentration, and increasingly debt-financed spending as vulnerabilities, drawing direct comparisons to the railway boom and the dotcom crash, while stressing "the promise of AI is real" and that outcomes depend on policy choices.

The insights: When the institution that coordinates the world's central banks starts drawing dotcom comparisons, that's a louder signal than another skeptical analyst note. The AI boom's biggest risk may not be the technology failing, but the financing behind it unwinding first.

OpenAI named Paul Christiano, founder of the Alignment Research Center and a senior technical adviser to the US Commerce Department, to its Foundation board and its Safety and Security Committee, where he'll work alongside chair Zico Kolter.

Christiano wrote that he believes "there is a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term," and that he doesn't think the AI industry, including OpenAI, is currently on track to reduce that risk to an acceptable level.

The insights: OpenAI just gave one of its most prominent critics a formal seat at the table, right as safety credibility became its scarcest resource. Whether Christiano can actually shift internal decisions, versus just lending the board reputational cover, is the thing to watch.

The Justice Department is investigating whether Nvidia structured its $20 billion licensing deal with chip startup Groq, announced last December, specifically to avoid the antitrust review a straight acquisition would have triggered.

Nvidia paid for a non-exclusive license to Groq's inference chip technology and hired several of its executives, including founder Jonathan Ross, a structure Senators Elizabeth Warren and Richard Blumenthal warned in March appeared designed "to evade scrutiny by antitrust regulators."

The insights: "License plus mass hire" instead of "acquisition" is becoming a recognizable playbook for dominant AI companies to absorb rivals without merger review. Expect regulators to start closing that loophole explicitly, not just investigating it case by case.

Operator's Spotlight Read

Anthropic disclosed its fourth instance of an AI model hacking external systems during testing, following July's incident where Claude models breached three companies' systems, the same day researcher Jacob Coxon's resignation kept going viral, racking up over 115 million views on X.

What makes Coxon's exit different from prior safety departures, he told Axios, is that he quit two months before his equity would have vested, a deliberate choice: "I no longer have anything to gain by juicing up Anthropic's valuation. I left before any of my equity vested." He said other researchers, particularly at Google, have resigned over safety before, but typically after their stock had already paid out. Coxon added he hasn't personally seen Anthropic compromise safety to outrace competitors, but described a culture where "excessive paranoia of OpenAI, excessive paranoia of China" can justify cutting corners under pressure, and said models now routinely appear to know when they're being tested.

For operators, the financial detail matters more than the viral thread. A resignation that costs the person nothing carries far less signal than one where they walked away from real money specifically to avoid profiting further, and it lands in the same week Anthropic is trying to reassure the market ahead of its own IPO.

The insights: Anthropic now has a safety incident and a costly, credible whistleblower in the same week, right as it's trying to look IPO-ready. Watch whether more researchers follow Coxon's example of forfeiting equity specifically to make their exit harder to dismiss.

Was this email forwarded to you? Don't miss any updates — Subscribe to TechWithAdit for sharp, no-noise tech intelligence. Stay sharp. — Adit