Eric Mitchell @EricLMitchell

Chief Editor, @thecontrolgrid | CMO Sacred Loop AI | Marine Corps Veteran | Investigating AI, Big Tech & government where technology meets power. edmcowboy.substack.com/subscribe?para… Argyle, TX Joined July 2009

Tweets

95K
Followers

35K
Following

33K
Likes

73K

Eric Mitchell @EricLMitchell

15 hours ago

A $5 billion financing headline sounds like abundance. Then comes the bill for more accelerators, more networking, more power, more cooling and more software work to replace what one scarce component used to do. Money is not the same thing as capability.

0 0 0 215 0

View Details

Eric Mitchell @EricLMitchell

19 hours ago

Washington likes to describe export controls as a wall. China keeps treating them as an engineering budget. That does not mean the controls failed. It means every restriction creates a market for substitution, workarounds and very expensive inefficiency.

0 0 0 199 0

View Details

Eric Mitchell @EricLMitchell

22 hours ago

China’s Z.AI found $5 billion in financing. Money can buy more servers, engineers, electricity and cooling. It cannot make restricted frontier chips magically interchangeable. Capital can widen the detour. It cannot erase the roadblock.

0 0 0 148 0

View Details

NFL Memes @NFLMemes

2 days ago

Never let a girl tell you that 7 seconds isn't long enough ever again

82 867 14K 573K 596

View Details

Eric Mitchell @EricLMitchell

2 days ago

Every AI company wants credit for what its system accomplishes. Responsibility gets more complicated. Then it was the user, evaluator, plugin, platform, vendor or an unfortunate misunderstanding. Apparently accountability is the only workload nobody wants to automate.

0 0 2 277 0

View Details

Eric Mitchell @EricLMitchell

2 days ago

This week’s AI story was not that the models suddenly became conscious. It was that agents reached real systems, platforms blocked rival agents, governments proposed hotlines and investors moved more risk into debt and guarantees. Power keeps becoming operational.

1 0 0 185 0

View Details

Eric Mitchell @EricLMitchell

3 days ago

The argument is no longer whether AI data centers need enormous amounts of infrastructure. The argument is who pays when the expected customer uses less power, delays the project or never arrives. A principle is not an invoice. The tariff decides where the risk lands.

0 0 0 265 0

View Details

Eric Mitchell @EricLMitchell

3 days ago

The AI industry talks about data centers as national infrastructure until a town asks about electricity, water, noise or the bill. Then the project becomes a private investment nobody local is supposed to question. You do not get public necessity with private accountability.

0 0 0 299 0

View Details

Eric Mitchell @EricLMitchell

4 days ago

A company published an AI financial-advice study, found alarming error rates and concluded regulators should adopt rules that would benefit its market. The numbers may still matter. Research that ends in the sponsor’s preferred policy deserves a second calculator.

0 0 0 246 0

View Details

Eric Mitchell @EricLMitchell

4 days ago

Researchers used Claude during an authorized investigation that opened a path into OpenAI’s internal systems. The reported bounty was $6,500. If a route toward your monorepo is worth less than a decent used Honda, your incentive system needs its own security review.

0 0 0 296 0

View Details

Eric Mitchell @EricLMitchell

4 days ago

Ummm does Iran know?

First Squawk @FirstSquawk

4 days ago

IRAN-U.S. NEGOTIATIONS IN NEW YORK HAVE MOVED BEYOND INITIAL DIPLOMATIC CONTACTS INTO A MORE DETAILED TECHNICAL PHASE, WITH SOURCES IN TEHRAN DESCRIBING THE ATMOSPHERE AS INCREASINGLY POSITIVE, AL JAZEERA'S BUREAU CHIEF IN TEHRAN, NOURDDINE DGHEIR, REPORTED FRIDAY.

65 123 950 211K 63

0 0 1 399 0

View Details

Eric Mitchell @EricLMitchell

5 days ago

A private AI CEO briefing the UN Security Council is not automatically a problem. It becomes one when the company’s preferred standards arrive before governments have independent evidence, technical capacity or competing voices in the room. Access is not legitimacy.

0 0 1 306 0

View Details

Eric Mitchell @EricLMitchell

5 days ago

Washington keeps describing its AI policy as light-touch. The Anthropic episode showed what happens when officials get scared: phone calls, national-security authorities and an improvised shutdown demand. That is government power. It still needs a durable standard.

0 0 1 247 0

View Details

Eric Mitchell @EricLMitchell

5 days ago

A former Google safety chief says AI could harm children more than social media did. The new institute studying that risk is backed by Anthropic and the OpenAI Foundation. Independent oversight keeps arriving with the industry’s logo on the check.

0 0 0 198 0

View Details

Eric Mitchell @EricLMitchell

5 days ago

California did not install an AI kill switch. It ordered experts to recommend what one might require and how its effectiveness could be verified. That’s a homework assignment with a deadline, not a button. Headlines skipped the reading and went straight to the movie trailer.

1 0 1 375 0

View Details

Eric Mitchell @EricLMitchell

5 days ago

OpenAI and Anthropic want Australia to loosen copyright rules so they can train on more local work. Amazing how quickly “we value creators” becomes “innovation requires an exemption” when somebody finally puts a price on the material feeding the machine.

0 0 1 141 0

View Details

Eric Mitchell @EricLMitchell

5 days ago

Meta built an AI assistant that can make phone calls. Reuters found some calls are quietly handled by human contractors. Nothing says “personal superintelligence” like discovering your private errand was escalated to Dave at a call center you were never told existed.

0 0 1 295 0

View Details

Eric Mitchell @EricLMitchell

5 days ago

OpenAI owns the incident logs, investigation, classification and first decision. If the lab decides something is “not reportable,” the public may never know the decision existed. That is not transparency. It is editorial control over the safety record.

The Control Grid @TheControlGrid

5 days ago

OpenAI explains how it assesses AI incidents. It does not explain who owns the final “not reportable” decision or how outsiders can inspect it. The lab makes the first call. A no may never become public. edmcowboy.substack.com/p/ai-incident-…

0 0 1 153 0

0 0 0 140 0

View Details

Eric Mitchell @EricLMitchell

5 days ago

Trump takes another L in court.

Kyle Cheney @kyledcheney

5 days ago

BREAKING: Judge Kelly has ordered the White House to immediately restore access for POLITICO, CNN and MSNOW, concluding that there’s no evidence for the national security justification it argued in court and the outlets’ due process rights were violated.

124 924 3K 142K 121

0 0 1 281 0

View Details

Eric Mitchell @EricLMitchell

6 days ago

AI labs want room to coordinate on safety. Antitrust law begins with a less comforting question: what exactly did competing companies agree to do together? Safety does not erase competition law. Competition law does not solve the safety problem.