Massively viewed AI-generated fake news: policy-compliant!

In this issue: the power of GAFA; reviewer openness; software catastrophes: will we ever learn?; three days until the deadline for VERIFAI.

What governments can do against Facebook: whine

There were worse items in the news last week, but still, this one can get you thinking about where the seats of power are today.

In a recent townhall-like meeting, the French president, Emmanuel Macron (it does not matter for this story whether you like him or not, we are talking about the function, not the man), tells the audience that he received a worried call from an African head of state inquiring about his whereabouts. The reason: a fake news report with 15 million views on Facebook; the reporter, against a civil-war-style background, explains that there has been a coup d'état in Paris and a colonel has taken power. A follow-up video doubles down by showing purported mass demonstrations by the colonel's supporters. All very well done, with real-looking and real-sounding journalists. All fake.

Then, as Macron explains (see the whole story, with the speeches dubbed in English, on France 24 here), someone from his staff calls Facebook to ask that the video be removed. Denied. Then, believing he would have more sway, he intervenes himself, to no more avail. The video, says Facebook, does not contravene its policies.

All that is left is for him to whine about the powerlessness of governments.

I am well aware of the subtleties involved and no reasonable person wants a government who can suppress news at will. This is something else, however. Non-state actors have in a few years gained enormous influence, unheard of in previous states of technological progress, which enables them potentially to disrupt entire countries or, who knows, start wars. The GAFA (Google, Apple, Facebook, Amazon — did it ever strike you how close this acronym is to MAGA, whose bidding they increasingly perform?) can do more or less whatever they want.

It is fashionable to rail about the horrible Brussels bureaucracy which stifles innovation and forces us to click “essential cookies only” seventy-seven times a day. Against the dominant players' blatant abuses of power, however, something must be done.

Reviewers exposed

(From an article in Science, through ACM Technews.) A security breach, taking advantage of a bug in the software, has revealed the identities of the reviewers for over 10,000 articles.

This was bound to happen (it has happened before, but the cases, at least those reported, were of a smaller scale).

I am not fretting: I walked out of the supposedly-secret-reviewer a long time ago. I always sign my reviews and turn them in conditional on my identify being revealed. I explained the rationale many years ago in this note. (The text is a bit heavy and I would write it more concisely today, but the ideas stand.)

There are very, very few cases justifying anonymity. Named-reviewer refereeing should be the default. We should have the courage of our opinions. And leaks will be one fewer thing to worry about.

One sure way to advance software quality

I have sung the following song more than once, in blogs and opinion pieces galore, but it is worth intoning again. Two weeks ago this newsletter had a note about the Cloudflare bug, with some elements of analysis from the available information. Have you heard more? Probably not. It is remarkable how software catastrophes make headlines, and then we forget. No lessons are learned.

Our model should be aviation. Travel by airplane has gone from the most dangerous to by far the safest of all modes of transportation. To quote myself from these earlier (obviously ineffectual) exhortations, it did not get there by accident. It got there by accidents.

Any airplane crash or near-crash gives rise to a painstaking inquiry, which must produce an exhaustive report so that lessons can be learned. And over the years the lessons have been learned, each time leading to improvements which, cumulatively, have worked wonders.

We should do the same. Any large-scale software mishap, particularly if it involves public money or causes significant public damage, should be required by law to cause a systematic review, paid for by the responsible company, with an obligation to reveal the software factors related to the bugs, and to publish the results. The Ariane crash in 1996 led to such a process (see the 1997 article by Jean-Marc Jézéquel and me for an analysis), but very few other cases did.

Pesky EU bureaucrats, are you listening?

Deadline approaching for VERIFAI submissions

Three more days for submitting proposals (which can be just abstracts) to the VERIFAI workshop on The Interplay Between Artificial Intelligence and Formal Verification (8-11 March 2026, near Toulouse in France). The workshop page with links to the EasyChair submission page is at https://www.laser-foundation.org/verifai-26/.

Cover photo: Rhapsody in Blue, White and Blond (window of a glass souvenir shop, Prague).