AI might take over the internet but it cannot crack a cryptic crossword

Key points

• OpenAI's CEO and Elon Musk both called for slowing AI development, as reported by the Guardian on 13 September.
• The Trump administration has pushed back against calls for AI guardrails, drawing criticism from Congress as of 15 September.
• Letter writers to the Guardian this week identified two awkward gaps in how the public talks about AI risk.
• One reader tested AI against the Guardian's cryptic crossword and found it useless.

The conversation around AI danger has a peculiar quality right now. We're told the technology might seize control of the entire internet and outpace human judgment entirely. And yet, one Guardian reader reports, it cannot solve a cryptic crossword.

John Hougham, writing from Reading in Berkshire, made that observation after the Guardian reported that both OpenAI's chief executive Sam Altman and Elon Musk had added their voices to calls for putting the brakes on what they described as reckless AI development. Hougham had apparently put the technology to the test. Result: rubbish at cryptic clues.

Small detail. Funny, too. But it captures something real about the distance between the most alarming claims about AI and what the systems actually do reliably today.

Does the comparison to Covid tell us anything?

A second letter makes a sharper point. Garry Wynne, from Norley in Cheshire, noted that President Donald Trump is now downplaying the dangers of AI, just as Congress pushes harder for safety guardrails. His challenge: didn't Trump say much the same thing about Covid?

A fair thing to put on record. A leader's track record of dismissing expert warnings about emerging risks matters when the next warning arrives. The comparison doesn't settle anything about AI specifically, but it invites scrutiny of the judgment being used to wave the concern away.

We reported on 14 September that Trump had called AI safety scrutiny a "SICK conspiracy" on Truth Social, without citing a single concrete example to back that up. Wynne's letter is, in effect, a reader doing the same cross-referencing.

Should you worry about any of this?

For most people, day-to-day AI is nowhere near the apocalyptic end of the spectrum. It drafts emails, suggests recipes, fails at cryptic crosswords. The existential-risk debate happens mostly among researchers and politicians.

But that debate shapes the rules governing the AI tools everyone uses. When governments decide how much oversight to apply, those choices eventually affect what AI can do to your job and your data.

The crossword joke is actually a useful anchor here. Both things can be true at once: the technology has real limits today, and those limits are not permanent. Watching who dismisses safety concerns, and whether they offer genuine technical evidence or just personal confidence that it will all be fine, is practical reading. It is not political theatre.