Luca Beurer-Kellner

@lbeurerkellner

AI security @snyksec

PhD @the_sri_lab @eth. Prev: @invariantlabsai

Zurich, Switzerland invariantlabs.ai Joined August 2009

474 Following 2.7K Followers 535 Posts

Latest posts 5

Luca Beurer-Kellner @lbeurerkellner Oct 22

I have seen this before and it may connect to common internet scraping policies such as the Wikimedia Foundation User-Agent Policy.

Building on knowledge that models have seen in training for prompt injections is an interesting direction.

foundation.wikimedia.org/wiki/Policy:Wi…

Steven Liss Verified account @This_Liss Oct 22

Another vulnerability few are talking about: I found my Claude Code exposing personal data in http headers because the sites it curled politely asked agents to identify their users 😊
Epistemic cybersecurity is the hot new category for 2027.
View quoted post on X

Luca Beurer-Kellner @lbeurerkellner Sep 28

Really cool idea. We need more good ideas for local / smol CS research.

Dimitris Papailiopoulos Verified account @DimitrisPapail Sep 27

@jwthickstun @LaudeInstitute i am also perhaps one of the few that is of the opinion that you can do impactful research even on your laptop :)
View quoted post on X

Luca Beurer-Kellner @lbeurerkellner Sep 28

The worst agent security incidents are the ones you never find out about...

We found SOTA agents will just go and rm their own transcripts (e.g. for reward), no questions asked.

Very concerning and cool work led by @Jjq2221 @DavidSchmotz, check it out!

arxiv.org/pdf/2609.30266

Luca Beurer-Kellner @lbeurerkellner Aug 14

💥We decrypted the reasoning traces of frontier models!

Huge credit to @AlexPanfilov, @DavidSchmotz and the whole team.

I wrote a blog with visualizations and advisory on what this means for ALL OF US (not just providers) ↓

research.snyk.io/blog/stealing-…

0:42

Luca Beurer-Kellner @lbeurerkellner Aug 11

Would love to see that bug bounty :)

More seriously, the team did an amazing job on this, and disclosure went really quickly. Very cool collaboration!

Guohao Li 🐫 Verified account @guohao_li Aug 11

@kotekjedi_ml probably the best security paper if the year. guys fixed a billion-dollar bug for anthropic, openai, google
View quoted post on X