We don't need to hack your AI Agent to hack your AI Agent …and we don't need an AI agent for that either :)

Via a large enterprise's AI assistant, we obtained access to several million Entra identities and all chat logs including attachments — no prompt injection or model tricks required.

For all we know, the poor agent was not at fault and may not have even been able to witness what was happening.

https://srlabs.de/blog/hacking-ai-agent

#AI #AIhacking #VulnerabilityDisclosure #ResponsibleDisclosure

Pokémon Go players thought they were catching Pikachus.

They were actually building the nervous system for robot civilization.

500M humans. 30B images. Zero consent forms.

The game was the harvest.
https://www.technologyreview.com/2026/03/10/1134099/how-pokemon-go-is-helping-robots-deliver-pizza-on-time/

RE: https://mastodon.social/@Gargron/1

10 years today.

What's the state of digital sovereignty for our academic landscape?

Inspired by a similar post looking at digital sovereignty of municipalities, I explored what messaging infrastructure universities rely on. Sadly, many have switched to hyper scalars but few large universities keep running their own email infrastructure. Germany, Austria, France does not look too bad and lead by example.

[Note that the assessment is based on a simple MX records comparison against a list of known scalars, I don't yet check SPF records or guesstimate the SMTP software/version, this may be done in a future version.]

Check out the interactive map: https://nebelwelt.net/gannimo/unimx/