The Weekly Waypoint, Issue #22
June 22, 2026
For two years, the AI industry has been selling a promise: this changes everything, and it changes everything for the better. This week, reality pushed back.
A Pew Research study found only 16% of Americans think AI will have a positive impact on society. KPMG pulled a report about AI usage because the AI it used to write the report hallucinated. A UK police officer is under investigation for using AI to "create evidence" in multiple cases. PwC published research showing AI is making medical bills higher, not lower. And Microsoft -- the company that bet its future on AI -- had to turn to AWS to keep GitHub running because its own infrastructure couldn't handle the load.
The gap between what AI companies say and what AI delivers has never been wider. And people are noticing.
Noam Shazeer Jumps to OpenAI
The AI talent war just claimed its biggest scalp yet. Noam Shazeer, co-lead of Google's Gemini project and founder of Character.AI, announced he's joining OpenAI. Shazeer is one of the most influential figures in AI -- he was an author on the 2017 "Attention Is All You Need" paper that basically created the transformer architecture every modern AI is built on. Google losing him to OpenAI is a gut punch, and it signals that OpenAI is still the destination of choice for top talent despite all the drama at both companies.
Only 16% of Americans Think AI Is Good for Society
Pew Research dropped a study this week that should make every AI executive uncomfortable. Just 16% of Americans believe AI will have a positive impact on society. Wall Street is pouring hundreds of billions into AI. Main Street isn't buying it. The gap between investor enthusiasm and public sentiment is becoming a strategic risk for the entire industry.
KPMG Pulls AI Report Over Hallucinations
KPMG published a report on AI usage, then quietly pulled it after readers noticed apparent hallucinations. The irony is painful: a report about AI adoption couldn't survive contact with AI. This is the trust problem in miniature -- if one of the world's biggest consulting firms can't use AI reliably for a report about AI, how can anyone trust it for anything that actually matters?
Police Officer Used AI to Fabricate Evidence
A Derbyshire police officer in the UK is under investigation for using AI to "create evidence" in multiple cases. This is the nightmare scenario for AI accountability: not a hypothetical risk, but an actual officer allegedly using AI to manufacture evidence that could have sent people to prison. The case is still under investigation, but it exposes a regulatory vacuum -- there are no clear rules about AI use in law enforcement, and the technology is moving faster than any framework to govern it.
AI Is Making Your Medical Bills Higher
PwC published a report this week showing that AI is increasing medical costs rather than reducing them. The promise was that AI would streamline healthcare administration, improve diagnostics, and lower bills. The reality, according to PwC, is the opposite: AI tools are adding complexity and cost to the billing process without delivering savings to patients. When the consulting firm that benefits from AI adoption admits it's making things more expensive, that's worth paying attention to.
Microsoft Turns to AWS as GitHub Buckles Under AI Load
Microsoft bought GitHub in 2018 with a clear plan: integrate it with Azure, make Microsoft's cloud the backbone of the world's code. This week, Microsoft started adding AWS capacity to keep GitHub running because AI-driven coding activity overwhelmed Azure's infrastructure. The company that wants to own the AI revolution can't keep its own developer platform running without help from its biggest cloud competitor. That's not a good look.
The Pentagon Is Proudly Using AI to Write Congress-Mandated Reports
The Department of Defense boasted this week that 1.5 million users across the Pentagon are using AI (specifically Google Gemini) to write reports that Congress requires. They're not hiding it -- they're bragging about it. Whether that's efficiency or an admission that compliance reports are so meaningless that AI can write them is an exercise left to the reader.
The Anthropic Saga Continues
WIRED reported this week that SK Telecom -- South Korea's largest wireless carrier and an Anthropic investor -- was at the center of the Mythos export controls controversy. The White House asked Anthropic to revoke SK Telecom's access after officials raised concerns about the company's potential China connections through its parent conglomerate SK Group. Anthropic complied immediately. Meanwhile, AWS reportedly raised concerns about Anthropic's models before the government crackdown, and the WSJ profiled Nicholas Carlini -- the researcher Anthropic sent to Washington to calm government nerves about AI safety. The Fable 5 shutdown aftermath is turning into a sprawling story about corporate influence, national security, and who gets to decide what AI is too dangerous to deploy.
The Takeaway
This week wasn't about one big story. It was about a pattern. The public doesn't trust AI. The consulting firms promoting AI can't use it without embarrassing themselves. The infrastructure supporting AI can't keep up. The government is both shutting down AI models and proudly using them to write its own paperwork. And the talent behind AI is still consolidating at a few companies while everyone else fights for scraps.
The AI industry has a trust problem, a cost problem, and an infrastructure problem -- and it's running out of time to fix them before the gap between promise and reality becomes a chasm.
This was the free weekly edition. Paid subscribers get the full deep dive -- a comprehensive analysis of AI's reality check moment, what the data says about public trust, why the infrastructure is buckling, and what it all means for the future of the industry.
See you next Monday. -James
The AI can write your reports, your evidence, and your medical bills. Whether it should is a different question entirely.