ShieldFont misleads scrapers while OpenAI ads can outlive a stop command; GitHub exposes rule bypasses, and young people demand a voice from the start of AI governance.
Artificial Intelligence··Midday
Readers see the page, scrapers get the decoy
ShieldFont, developed by Isaque Seneda and Gabriel Abrucio, displays the right words to a person looking at a web page while substituting other words in its source code. The ligature-based system changes an average of 24.5 per cent of all words on a page and 45.8 per cent of its content words. Over three months, the developers built a dictionary of nearly 12,000 common words and assigned three separate replacements to each entry. In tests against six public scraper pipelines, more than 90 per cent of pages that would ordinarily have been accepted failed the quality filter. Among the small share that passed, about 20 per cent of the words became correctly spelled text carrying false information. The trap raises the cost of collection: a scraper can render the page and apply optical character recognition to bypass the font, but doing so costs 5 to 13 times more than retrieving raw source code. The protection also reaches readers. Search engines, screen readers, copy-and-paste tools and translation software encounter the damaged source text, so a barrier aimed at scrapers can also disrupt accessibility and ordinary web tools.[1]
A stopped ad may run on, while a bypassed rule becomes visible
OpenAI's advertising terms allow a cancelled or changed campaign to keep running for up to one business day, with the advertiser responsible for whatever is delivered during that window. Excel4Business told The Register that its ads were still running about 94 minutes after it paused the campaign. The advertiser reported 60.72 pounds in invalid charges in the first incident and about 6.47 pounds when the problem recurred on 4-5 August. OpenAI initially acknowledged a defect, then declined a refund under the terms. A spokesperson said the one-business-day window was standard practice and that ads were not deliberately run after a campaign was switched off. An organisation-level rules dashboard now in public preview on GitHub makes a different control point visible. Administrators can see allowed, failed and bypassed runs across every repository, filter by status, branch, ruleset and date range, and export results as CSV. The dashboard also ranks repositories with the most bypasses. In one interface, a stop command does not end delivery immediately; in the other, administrators gain a view of where a rule was bypassed after evaluation.[2], [3]
Young people want a seat from the beginning
More than 600 young people gathered at UN headquarters in New York for International Youth Day on 12 August and launched their own declaration on AI policy and governance. UN News says the text will feed into work on the Global Digital Compact. Felipe Paullier, Assistant Secretary-General for Youth Affairs, said consulting young people tokenistically at the end of a process does not amount to meaningful participation; their involvement has to be designed in from the beginning. Data presented at the meeting indicate that children adopt AI 3 times faster than adults. Another figure puts 6.1 per cent of jobs held by people aged 15 to 29 among those most exposed to AI-driven displacement. Organised by the International Organization of Youth, the event moves the control debate beyond software switches. It asks for the generation using the technology faster and living longer with its labour effects to help shape the rules. ShieldFont gives publishers a technical barrier, and GitHub gives administrators visibility, while the advertising case exposes a delay between a button and its outcome. The youth declaration turns to who gets a voice when those controls are designed in the first place.[4]
