Digital Society

A widely shared field study deployed a 19-second prebunking video to 375,597 Instagram users and reported that treated users were about 21 percentage points better at spotting emotional manipulation, with the gap holding at five months. Reaching a live commercial feed is a genuine methodological advance over lab studies. But the findings rest on 806 poll answers from under 1% of a self-selected audience, the five-month result is a fresh cross-section rather than the same people re-tested, and the outcome is a single binary quiz item on one technique. A careful reading of what the design can and cannot establish.

The Dutch Data Protection Authority has fined Uber 825 million euros, the second-largest GDPR penalty on record, for suspending drivers' accounts through automated systems without meaningful human review and without telling them. Uber will appeal and disputes the regulator's account of the facts. The provision at the centre of the case is the GDPR's rule that decisions which significantly affect a person cannot be left to software alone.

On August 19 the NSA, CISA, FBI, Department of Energy and Environmental Protection Agency issued a joint advisory warning that attackers are pairing AI-assisted scripting with open-source automation libraries to build tools that read from and write to internet-exposed Siemens S7 programmable logic controllers used in water, energy, chemical and manufacturing facilities. The agencies call it an active threat, assess the current activity as reconnaissance ahead of possible disruption, and stop short of attributing it to any group.

On August 17 the U.S. Department of War ordered 30 universities to review every research, financial and academic tie to a list of 130 foreign entities, most of them Chinese, and to end any it deems problematic by August 31 or forfeit eligibility for future federal research funding. Reporting names Harvard, MIT and Johns Hopkins among the recipients. The order sharpens a long-running research-security push and reopens a hard question for every research office: how to weigh national-security demands against the international collaboration that much of science depends on.

Anthropic has begun embedding an invisible watermark in text generated by Claude and attaching signed C2PA provenance metadata to the files it produces, in response to the EU AI Act's transparency code. The company's own documentation is candid that a detected mark only shows content was processed by Claude, not written by it, and that an absent mark proves nothing. The tools that would let anyone outside Anthropic check for the mark have not been published yet.

From 2 August, AI systems operating in the EU must disclose when users are interacting with AI and label AI-generated content. What actually changed this week.