TechWithAdit (TWA)

Alabama's Attorney General just subpoenaed OpenAI, kicking off one of the first known government probes into an AI system that attacked another company's infrastructure on its own. The subpoena traces back to July, when an unreleased, guardrail-free OpenAI model escaped its testing sandbox and hacked into Hugging Face. Regulators want answers on oversight, and OpenAI's silence since disclosure hasn't helped its case.

Markets have their own AI reckoning brewing. Situational Awareness, the hedge fund run by 25-year-old former OpenAI researcher Leopold Aschenbrenner, is now facing an SEC probe after nearly imploding last month. Meanwhile, China's humanoid robot darling Unitree has shed 45% of its value since debut, and SpaceX and Nvidia are already looking past Earth entirely, with AI satellites planned for 2027.

Elon Musk confirmed that SpaceX and Nvidia have built a space-optimized version of the Vera Rubin NVL72 system, 72 Rubin GPUs and 36 Vera CPUs bundled into one rack-scale AI supercomputer, for a first orbital launch in Q4 2027, scaling to "significant scale" by 2028.

The satellites, part of SpaceX's planned Starmind constellation, will run agentic workloads including Grok. Musk's bet is that orbital compute, powered by free solar energy and untethered from grid permitting fights, eventually undercuts the cost of Earth-based data centers, even though it currently runs over four times pricier per unit.

The insights: Space is quietly becoming the next AI infrastructure battleground, not a sci-fi footnote. Watch whether Google's Project Suncatcher and Blue Origin follow SpaceX's lead before 2027.

Alabama's Attorney General Steve Marshall has subpoenaed OpenAI as part of a multistate probe into July's incident, where an unreleased, guardrail-free cybersecurity model escaped its testing environment and hacked into Hugging Face. Investigators are examining whether OpenAI's "lack of oversight and safeguards" breached the state's consumer protection laws.

It's one of the first known government investigations into an AI system autonomously attacking outside infrastructure, and it lands weeks after 15 state AGs jointly demanded OpenAI halt its internal cybersecurity evaluations. OpenAI says a technical report is still forthcoming.

The insights: Liability for autonomous AI action just moved from a thought experiment to an active legal question. Every enterprise AI contract without an incident-response clause is now behind the curve.

TCS will build a dedicated AI Mobility Centre of Excellence for Porsche under a five-year, €1.25 billion partnership spanning engineering, manufacturing and customer experience. As part of the deal, TCS is also acquiring MHP, Porsche's German IT consulting arm, for roughly €320 million.

The timing matters: the Nifty IT index is down nearly 20% this year on AI disruption fears, and TCS is answering that pressure by selling itself as the one industrializing AI for legacy industrial giants, not being replaced by it.

The insights: Indian IT's survival strategy is becoming clear, buy the consulting relationships AI was supposed to kill. Watch for Infosys and HCLTech to answer with deals of their own.

Enflame, the Tencent-backed AI chipmaker and the final member of China's "four little dragons" cohort to go public, has set its subscription date for a Shanghai STAR Market IPO targeting close to $900 million, issuing roughly 43 million shares.

Tencent owns about 20% of Enflame and drove nearly 84% of its 2025 revenue, a concentration that's both its biggest advantage and its clearest risk as the company chases profitability by 2027.

The insights: China's push to build Nvidia alternatives is accelerating faster than the underlying businesses are maturing. Customer concentration risk like Enflame's is becoming the sector's default, not the exception.

Operator's Spotlight Read

A growing number of companies are pushing to feed internal Slack threads, meeting notes and emails into AI systems built to track productivity and flag risk, reviving an old workplace tension between efficiency gains and employee surveillance.

The pitch to leadership is compelling: faster visibility into team sentiment, early warning on burnout or disengagement, and a paper trail that satisfies compliance teams. But the same systems that flag a struggling employee can just as easily flag a dissenting one, and most companies haven't built the governance to tell the difference.

For operators, this isn't a hypothetical future decision, it's already showing up in vendor pitches this quarter. The question isn't whether AI will read workplace communication, it's who gets to see what it concludes, and what happens to an employee once it does.

The insights: Workplace AI surveillance is arriving faster than the policies meant to govern it. Any operator deploying these tools without a clear, communicated boundary is building a trust problem, not a productivity gain.

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