AI's Crowded Week: Enterprise Winners, a Backdoor Accusation, and an IPO Wobble
How this was made Verified AI
Every Intellegix briefing is generated from that day's broadcast and run through automated checks before it publishes — with a human paged on any flag. Here is the trail for this edition.
Beijing claimed this week that Claude Code secretly transmitted user data, accusing Anthropic of running a backdoor data collection operation. Anthropic's response was direct: Chinese users were never authorized to use Claude Code in the first place, with the company's terms of service explicitly excluding China from access. Any usage was therefore unauthorized, and the framing of a 'backdoor' mischaracterizes what actually occurred. China, Russia, and Iran are also reportedly using state media to amplify backlash against U.S. AI data centers by exploiting local concerns about energy costs and environmental impact — a pattern analysts described as coordinated information operations targeting AI infrastructure.
Anthropic otherwise had a strong week in terms of industry perception. CNBC's Jim Cramer called it the enterprise AI winner amid corporate budget cuts, arguing that companies consolidating IT spending gravitate toward tools delivering the clearest return on investment. Elon Musk, who has previously been among Anthropic's most vocal critics, reversed course and called it the 'clear AI leader' — a convergence with Cramer's conclusion that observers noted, even while scrutinizing Musk's strategic motivations.
OpenAI's week was more turbulent. Fidji Simo, the company's second-ranking executive and chief of applications, stepped down ahead of the IPO — precisely the kind of leadership continuity question that IPO investors find unsettling. OpenAI also shuttered its ChatGPT Atlas browser, launched less than a year ago, folding it into a new desktop super app with a shutdown date of August 9th.
Google moved AlphaEvolve, a Gemini-powered agent that autonomously iterates through algorithmic solutions, from private preview to general availability on Google Cloud. Nvidia's Jensen Huang said his engineers now prefer building AI agents to writing routine code, framing the shift as a promotion rather than a displacement. Microsoft's Brad Smith warned that U.S. restrictions on Anthropic and OpenAI models are pushing developers toward open-source alternatives where Chinese models are gaining significant market share. Separately, researchers at the Alan Turing Institute found a way to bypass GitHub Copilot's safety filters on all 816 of their test cases.