Microsoft Word: AI Copilot Exploited for Attacks
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
A New Vulnerability in Microsoft Word
Håkon Måløy has recently uncovered a new attack method involving prompt injection targeting Microsoft Word. This innovative technique transforms attacks into self-replicating worms, thereby increasing their potential for propagation.
The attack involves inserting hidden instructions into a Word document, which is then used as source material by Copilot for Word. The latter can interpret these instructions as part of the user's request, which can manipulate the document being drafted or edited. Furthermore, Copilot can copy these hidden instructions into the resulting document, turning it into a new propagation vector.
A Concerning Propagation Vector
If this new vector is used in another workflow assisted by Copilot, the instructions can trigger again and spread to other documents, even in the absence of the original document from the attacker. This ability to self-replicate is a first in the realm of prompt injection attacks.
Microsoft's Response
The vulnerability was responsibly reported to Microsoft, which had 144 days to develop a fix. However, no comprehensive solution has yet been implemented to cover the entirety of this class of attacks. This situation leaves Word users exposed to potentially serious security risks.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.