Things that caught my attention
Last week, the UK Information Commissioner's Office (ICO) published guidance on AI-drafted subject access requests. Subject Access Requests, or SARs, are a legal right under UK GDPR that let an individual ask an organisation for a copy of the personal data they hold about them.
The ICO guidance is pretty good, and it recognises both that there is an increase in the use of AI in filing requests, and that this sometimes makes them harder to respond to as they've become longer, repetitive, broader or unclear in scope; but it is also very clear that the use of AI does not make a SAR invalid. "AI is another form of assistance", says the guidance, and I agree.
There's a lot of attention to this topic, which is somewhat linked to the "arms race" problem I pointed to in Quantum #672 – agentic flooding: there's a general impression that there's so much AI-generated filing of applications, complaints, requests, that organisations may be overwhelmed and unable to address those requests properly. It is a topic that matters to me, in my current job: as in any organisation that drives engagement from webchats and phone lines, increased AI access to these facilities must be closely looked at so that there are no issues in terms of so much overload that we can't answer contact requests from human users, and also regarding data protection, accuracy of advice, etc.
Is agentic flooding a real problem, and how much of an impact is it having? Academia is looking at it, mostly in terms of risk management and how to limit exposure. In August, Tom Loosemore of Public Digital published a briefing note suggesting that pressure is real but still early, so organisations have time to prepare.
Obviously it's for us to monitor and adapt. One important aspect is to think about whether we could use APIs to better serve agentic requests. As Orsola De Marco argues here, this means improving the reliability and discoverability of public (open) data. "The reader is changing", she writes. "Much of the open data ecosystem still treats AI readiness as a future-state question. It isn’t. Every dataset published today needs to be designed for agent-era reading." I totally agree with this. The other way of adapting, though, is to use all this as a strong hint as to what processes cause citizens to come for advice, to file complaints, or to provide documents, and fix them before they need to ask for support. This is what reimagining service thanks to AI means.
For a good look at agentic flooding stats, look at this interactive website linked to the paper above by Chris Schmitz et al., from which the illustration below is taken.

'till next week,
Giuseppe @puntofisso.bsky.social