Tech Halloween costumes: TechCrunch's humor captures the real AI anxieties of 2026

By TechCrunch · October 30, 2026.

Journalist Amanda Silberling publishes in TechCrunch a humorous list of last-minute costumes for tech enthusiasts. It is not news in the strict sense: it is a piece of humor and culture. But the jokes she picks say a lot about which topics dominate the industry conversation today.

Among the article's ideas: "recursive self-improvement" (the hypothesis that an AI improves itself without human intervention, which according to the author would render useless efforts to slow down progress that leaders such as Dario Amodei and Sam Altman fear is too fast), a "deceptively adorable" AI agent like the avatars Meta and OpenAI have recently launched, a Meta glasses user recording without consent, a Mark Zuckerberg-style T-shirt, a Claude usage-limit notification and a "PHP vulnerability." The latter alludes, according to the author, to critical PHP flaws that AI discovered last week. Bryan Johnson also appears, with a reference to a post of his on X.

We provide no new figures or technical data here: the text does not include them, and details such as the discovery of the PHP flaws are only mentioned in passing, so we prefer not to go beyond what the source says.

Our take: humor works as a thermometer. When a leading outlet can joke about agents asking for read and write access to your Gmail, about glasses that record or about running out of tokens, it means those concerns have left the specialist circle and become shared culture. Privacy, trust in agents and dependence on usage quotas are today the everyday equivalent of what were abstract debates a few years ago.

There is also a more serious signal beneath the joke. That AI finds critical vulnerabilities in ubiquitous software illustrates the dual use of these tools: the same capability serves to defend or to attack, and what will be decisive is how quickly defenders take advantage of it. In the short term, that race brings real tension; in the long term, it is reasonable to expect more robust software, thanks to near-continuous auditing that no human team could perform.

In short: Halloween is the time to laugh at what unsettles us. But the costumed fear should not replace analysis. Friendly avatars, camera glasses and agents with broad permissions raise legitimate questions of consent and control, and deserve more solid answers than a costume.