Thomas Friedman calls Craig Mundie his “technology tutor.” He’s said it publicly, repeatedly, for over a decade. Perhaps it’s meant to sound endearing. It’s actually a sad confession. The most influential foreign affairs columnist in America openly outsources his entire understanding of technology to a single person. Why?
That person’s track record deserves closer examination within context of a larger institutional failure rated as success.
Microsoft Failure Man
In 1982, Mundie co-founded Alliant Computer Systems, a maker of vector-parallel mini-supercomputers. He became CEO. The company filed for bankruptcy in 1992.
He joined Microsoft that same year to run the Consumer Platforms Division. Here is what he built and championed:
| Year | Product / Initiative | Outcome |
|---|---|---|
| 1992 | Windows CE | Dead |
| 1990s | Pocket PC | Dead |
| 1990s | Auto PC | Dead |
| 1997 | WebTV Networks ($425M acquisition) | Dead |
| 1990s | Interactive television | Dead |
| 2000s | Digital rights management strategy | Dead |
When Bill Gates stepped back from daily operations in 2006, Mundie and Ray Ozzie were appointed to fill his visionary role. Mundie became Chief Research and Strategy Officer. During his tenure in that position, Google beat Microsoft to self-driving cars. Apple beat Microsoft to voice recognition with Siri. Microsoft missed mobile. Missed search. Missed social. Missed cloud computing until Satya Nadella arrived and reoriented the entire company away from the strategy Mundie had been overseeing.
To be fair, Ballmer overruled strategy recommendations, and search and social blinders may have been other divisions’ calls. But show me the wins. By late 2012, Mundie was moved to “Senior Advisor to the CEO,” the corporate equivalent of a quiet pasture upstate where his opinions wouldn’t be heard anymore. He retired in 2014.
Selling Invisible Pants to Elites
What happens to a technology executive whose products all failed but whose rolodex thrived? He becomes a bogus sage. Mundie landed on the National Security Telecommunications Advisory Committee under three presidents. Obama’s PCAST council. The Bilderberg Group steering committee, which he attended every year from 2003 to 2019 except one. The World Economic Forum. And then the capstone: co-authoring Genesis with Henry Kissinger and Eric Schmidt, a book about AI published in November 2024.
Kissinger.
The architect of the secret bombing of Cambodia. The man who backed the coup in Chile and the Indonesian invasion of East Timor. Who enabled the Pakistani genocide in Bangladesh. Who treated civilian populations as abstractions to be managed through force. This is who Mundie chose as his co-author on the governance of a technology that will reshape civilian life everywhere.
That tone-deaf choice tells you everything about the Mundie gambit: power managed by the powerful, consequences borne by everyone else. Talk down and ignore reality.
The technology product arc of disasters was foreshadowing. Build things that fail. Accumulate institutional access along the way. Pivot from practitioner to advisor once the failure pattern becomes undeniable. The access persists because Davos doesn’t audit skills like product launches. It rewards gravitas in presence.
The Column
This week Friedman published a column about Anthropic’s Claude Mythos announcement. His source for interpreting its significance was, naturally, Craig Mundie. The column treats Mundie’s analysis as authoritative. It contains Mundie’s three-step framework for responding to the threat. It quotes Mundie at length. It cites no other technical source.
Friedman writes:
[Mundie is] a former director of research and strategy at Microsoft, a member of President Barack Obama’s President’s Council of Advisors on Science and Technology and an author, with Henry Kissinger and Eric Schmidt, of a book on A.I. called Genesis.
Every word of that sentence is true. Every word of it is also designed to obscure the fact that Mundie’s actual technology career was a sequence of expensive bets that all lost.
The credential list substitutes for any actual performance record.
The Friedman column is similarly bad at placing bets. It describes AI-powered vulnerability discovery as if no one had ever heard of fuzzing, static analysis, or red team operations.
OMFG.
He illustrates the threat with a scenario where children accidentally take down a power grid. This is the Computer Fraud and Abuse Act (CFAA) debate of 1984, recycled. Congress passed it in 1986 after WarGames convinced legislators that teenagers with modems could launch nuclear weapons. Before that, Captain Crunch and the phone phreakers were going to launch nukes, shut down power or destroy the telephone system.
Did he just wake up from 1986?
By early 1999 I personally had reported clear text authentication on the American bulk power grid in five states. Thousands of routers vulnerable to trivial destruction. Did I break anything? The old geezer in a suit panicking that “kids will break everything” has been the go-to move for people who want to centralize control of new technology for forty years. Friedman presents it as an original insight, as if nobody remembers the 414s.
He proposes US-China cooperation on AI governance in the same week the US is actively restricting chip exports to China to prevent exactly the AI capability development he’s now asking them to collaborate on. He doesn’t notice this dumb contradiction because he’s transcribing, not analyzing. He would do far better to trust AI than the hallucinations of Mundie.
Meatspace
A columnist who calls someone his “tutor” on a subject is telling you he cannot independently evaluate what that person says. Friedman admits no second opinion on technology, and he lacks the technical literacy to know what questions would surface one. Mundie tells him this is unprecedented. Friedman writes that it’s unprecedented. Mundie says it requires US-China cooperation. Friedman writes that it requires US-China cooperation.
This is how a man who got WebTV, Windows CE, the Pocket PC, interactive television, and digital rights management wrong becomes the person explaining artificial intelligence threats to New York Times readers.
The technology tutor model has an obvious flaw. The student can’t evaluate the false tutor, because the student can’t evaluate the subject. He can only evaluate the tutor’s confidence, which is deeply ironic. Just like the real danger with a bad AI chatbot, confidence is the one thing Craig Mundie has never lacked.
The threat of AI is the relationship Friedman has with Mundie, not the AI.