iiinova — where your digital self keeps evolving

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

We also show hand-crafted attacks can still bypass CFI built on even the most precise of these call graphs. (8/8) Joint work with Fabian Monrose and Michalis Polychronakis (@mikepo.bsky.social)! ✳️ Paper: cs.brown.edu/people/vpk/p... 💾 Code: github.com/yufeidu/SoK-... #woot26 #browncs #brownssl

000

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

Every tool is also supposed to fall back to single-layer analysis on failure, but none of the released implementations actually do this, and even after we add that fallback ourselves, every multi-layer approach still loses to the baseline in soundness once compiler optimizations are enabled. (7/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

Prior evaluations measured precision only over indirect calls a tool successfully resolved, quietly excluding the 20-60% of calls where analysis just failed! (Count those properly and the reported gains mostly disappear.) (6/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

Our work runs the first systematic study of five state-of-the-art MLTA-like approaches against the single-layer baseline, and finds the promise doesn't hold up. (5/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

"Multi-layer type analysis" (MLTA) promised a fix, adding struct-hierarchy reasoning to sharpen precision without sacrificing soundness, and follow-up work refining the idea has since appeared at multiple top security venues. (4/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

The state-of-the-practice, type analysis, is fast but crude -- it just matches pointer types, which is way too imprecise in real codebases. (3/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

Security tools like CFI and directed fuzzing all depend on accurate call graphs, but indirect calls resolve their targets only at runtime, so static tools have to infer them ahead of time. (2/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

Earlier this month, Yufei Du presented our work, "SoK: Multi-Layer Indirect Call Analysis in the Real World", at the USENIX Association WOOT Conference on Offensive Technologies (WOOT) 2026! (1/8) @wootsecurity.bsky.social

111

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

000

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

Our work runs the first systematic study of five state-of-the-art MLTA-like approaches against the single-layer baseline, and finds the promise doesn't hold up. (5/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

The state-of-the-practice, type analysis, is fast but crude -- it just matches pointer types, which is way too imprecise in real codebases. (3/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 27/08/2026

Security tools like CFI and directed fuzzing all depend on accurate call graphs, but indirect calls resolve their targets only at runtime, so static tools have to infer them ahead of time. (2/8)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 21/07/2026

✳️ Paper: cs.brown.edu/people/vpk/p... 💾 Code: github.com/meghna-panch... #DIMVA2026 #browncs #brownssl

000

Vasileios Kemerlis @vkemerlis.bsky.social · 21/07/2026

Joint work with Andreas D. Kellas, Kostis Kaffes, @stevebellovin.bsky.social, and Simha Sethumadhavan! Thanks to Sotiris Ioannidis and Michalis Diamantaris for organizing a great conference! It was a real pleasure attending and catching up with so many friends and colleagues along the way. (6/6)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 21/07/2026

This lets Santa build tight, workload-specific syscall allow-lists on the fly, without paying the full cost of heavyweight hardening in steady-state traffic. (5/6)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 21/07/2026

When the production service hits a syscall outside its current policy, Santa re-runs the request against the hardened oracle -- if it's benign, the policy is safely expanded; if it's bogus, an attack is caught and blocked. (4/6)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 21/07/2026

Our work, Santa, resolves this by running two versions of an app side-by-side: a fast production version and a "hardened" security-oracle version (instrumented, for example, to catch memory-safety and concurrency errors). (3/6)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 21/07/2026

Cloud teams securing microservices are usually stuck choosing between heavy isolation that hurts latency, or system call filters that are either too loose (based on static analysis) or too brittle (based on dynamic tracing). (2/6)

100

Vasileios Kemerlis @vkemerlis.bsky.social · 21/07/2026

Huge congrats to Meghna Pancholi for presenting our paper, Santa -- a language-agnostic approach to system call policy learning for microservices -- at #DIMVA2026 (the Conference on Detection of Intrusions and Malware & Vulnerability Assessment | @dimvaconf.bsky.social) in Chania, Crete! 🇬🇷 (1/6)

121

Vasileios Kemerlis @vkemerlis.bsky.social · 27/03/2026

Excited to be co-chairing RAID 2026 (raid2026.org) alongside Tiffany Bao this year! Looking for a security venue to submit your research? The deadline is just around the corner—April 16, 2026. CFP: raid2026.org/call.html Looking forward to seeing you in Lancaster this October (11–14)!

144

Vasileios Kemerlis @vkemerlis.bsky.social · 21/10/2025

Contributing to the community—through both research and reviewing—has been one of the most fulfilling aspects of my academic career. Many thanks to the organizers, colleagues, and students who make CCS such a vibrant and rigorous forum for computer security research! #acm_ccs #browncs #brownssl 🚀

010

Vasileios Kemerlis @vkemerlis.bsky.social · 21/10/2025

🏆 Top Reviewers Award, recognizing service and contributions to the CCS community. I'm especially grateful for this honor, as it marks the third consecutive year (2023, 2024, and 2025) that I've received a service award from CCS — a tradition I'm proud to continue.

120

Vasileios Kemerlis @vkemerlis.bsky.social · 21/10/2025

🏅 Distinguished Artifact Award for our paper "PickleBall: Secure Deserialization of Pickle-based Machine Learning Models" (bsky.app/profile/vkem...).

120

Vasileios Kemerlis @vkemerlis.bsky.social · 21/10/2025

🎉 @acm.org CCS 2025 in Taipei, Taiwan was a blast! I had a great time connecting with colleagues and friends at ACM SIGSAC's flagship security conference — a week filled with inspiring research and thoughtful discussions. I was also deeply honored to receive two awards this year:

250

Vasileios Kemerlis @vkemerlis.bsky.social · 21/10/2025

Joint work with Neophytos Christou (Brown University), Columbia University (Junfeng Yang, Penghui Li), Purdue University (Jamie Davis, Wenxin Jiang), Technion (Yaniv David), and Google (Laurent Simon). ✳️ Paper: cs.brown.edu/~vpk/papers/... 💾 Code: github.com/columbia/pic...

000

Vasileios Kemerlis @vkemerlis.bsky.social · 21/10/2025

This work continues our broader effort to secure deserialization across ecosystems -- building on our earlier research presented by Yaniv David at NDSS 2024 (tinyurl.com/mbcevsv6), and Neophytos Christou and Andreas Kellas at BlackHat USA 2025 (tinyurl.com/bdvny4w7).

100

Vasileios Kemerlis @vkemerlis.bsky.social · 21/10/2025

#PickleBall is a static analysis framework that derives and enforces safe deserialization policies for pickle-based ML models. It infers permissible object types and load-time behaviors directly from ML-library code and enforces them through a secure, drop-in replacement for Python's pickle module.

100

Vasileios Kemerlis @vkemerlis.bsky.social · 21/10/2025

📢 Last week, Andreas Kellas presented our work on secure deserialization of pickle-based Machine Learning (ML) models at @acm.org CCS 2025! #pickleball #mlsec #mlsecops #brownssl #browncs

230

Vasileios Kemerlis @vkemerlis.bsky.social · 19/05/2025

👏 Kudos to Marius Momeu (leading author) who did a terrific job presenting our paper -- joint work Alexander Gaidis (Brown University) and Jasper von der Heidt (TU Munich). ✳️ Paper: cs.brown.edu/~vpk/papers/... 💾 Code: github.com/tum-itsec/iu... (coming soon) #brownssl #browncs #ieeesp2025

020

Vasileios Kemerlis @vkemerlis.bsky.social · 19/05/2025

#IUBIK leverages memory tagging (MTE) and pointer authentication (PA), available in #ARM CPUs, to efficiently and effectively isolate attacker-controlled input from security-critical data in the kernel heap. #iubik #mte #pac #arm #brownssl #browncs #ieeesp2025

100

Vasileios Kemerlis @vkemerlis.bsky.social · 19/05/2025

📢 Last week, Brown Secure Systems Lab (SSL, gitlab.com/brown-ssl) was at the IEEE Symposium on Security and Privacy (S&P) 2025, where we presented our latest work on hardening OS kernels against attacks that (ab)use heap-based memory-safety vulnerabilities. #brownssl #browncs #ieeesp2025 🧵

120

Vasileios Kemerlis @vkemerlis.bsky.social · 12/05/2025

#IUBIK hardens OS kernel code against attacks that (ab)use memory errors via means of: (1) attacker-controlled input isolation, (2) memory tagging (ARM MTE), and pointer encryption (ARM PA). ✳️ cs.brown.edu/~vpk/papers/... 💾 github.com/tum-itsec/iu... (soon) #iubik #mte #pac #arm #brownssl #browncs

000

Vasileios Kemerlis @vkemerlis.bsky.social · 12/05/2025

If you're attending the IEEE Symposium on Security and Privacy 2025 and interested in OS/kernel (self-)protection, please stop by the "Memory Safety" track today (Session 3, Track 3, 1PM--2:30PM in Bayview AB) to hear more about #IUBIK from Marius Momeu! #ieeesp2025 #iubik

210

Vasileios Kemerlis @vkemerlis.bsky.social · 12/05/2025

Thank you, Xing Gao and the University of Delaware CIS department for the warm welcome, thoughtful discussions, and the tour of the acclaimed CAR (www.thecarlab.org) lab! #binwrap #sysfilter #nibbler #brownssl

000

Vasileios Kemerlis @vkemerlis.bsky.social · 12/05/2025

If this area of research interests you, you might also find our recent work on Quack (hardening PHP code against deserialization attacks, NDSS 2024: 📄 cs.brown.edu/~vpk/papers/... 💾 github.com/columbia/quack) worth a look.

cs.brown.edu

110

Vasileios Kemerlis @vkemerlis.bsky.social · 12/05/2025

✳️ sysfilter (RAID 2020: 📄 cs.brown.edu/~vpk/papers/..., 💾 gitlab.com/brown-ssl/sy...) -- Automated system-call policy extraction and enforcement in binary-only applications. ✳️ Nibbler (ACSAC 2019: 📄 cs.brown.edu/~vpk/papers/..., 💾 gitlab.com/brown-ssl/li...) -- Shared-library code debloating.

cs.brown.edu

100

Vasileios Kemerlis @vkemerlis.bsky.social · 12/05/2025

I also shared a few highlights from our research efforts over the past five years re: supply-chain security: ✳️ BinWrap (ACM ASIACCS 2023, Distinguished Paper Award 🏆: 📄 cs.brown.edu/~vpk/papers/..., 💾 github.com/atlas-brown/...) -- HW-assisted (via Intel MPK) sandboxing of native Node.js add-ons.

cs.brown.edu

110

Vasileios Kemerlis @vkemerlis.bsky.social · 12/05/2025

My talk, titled "Hardening the Software Supply Chain: Practical Post-Compilation Defenses", was part of the SAVES workshop at IEEE MOST. I discussed both the pressing open problems in this evolving field and the next-gen. challenges of protecting critical infra. from software supply chain attacks.

100

Vasileios Kemerlis @vkemerlis.bsky.social · 12/05/2025

📢 Last week, I had the pleasure of visiting the beautiful University of Delaware (@udelaware.bsky.social) to speak about supply chain security, and reconnect with friends and colleagues! #brownssl #browncs 🧵

100

Vasileios Kemerlis @vkemerlis.bsky.social · 06/04/2025

Thank you to my host Charalampos Papamanthou and the Yale CS department for the warm welcome and thoughtful discussion!

010

Vasileios Kemerlis @vkemerlis.bsky.social · 06/04/2025

It was a real pleasure catching up with friends, colleagues, and students. And with the spring weather fully cooperating, I couldn't resist snapping a few photos of Yale's beautiful campus in the early morning light.

110

Vasileios Kemerlis @vkemerlis.bsky.social · 06/04/2025

If this area interests you, you might also find BeeBox (strengthening eBPF against transient execution attacks, USENIX Security 2024: 📄 cs.brown.edu/~vpk/papers/..., 💾 gitlab.com/brown-ssl/be...) and IUBIK (leveraging ARM MTE+PA to isolate attacker-controlled data, IEEE S&P 2025) worth a look.

cs.brown.edu

120

Vasileios Kemerlis @vkemerlis.bsky.social · 06/04/2025

✳️ EPF (USENIX ATC 2023: 📄 cs.brown.edu/~vpk/papers/..., 💾 gitlab.com/brown-ssl/epf) — Exploiting the (e)BPF sub-system for bypassing modern protections and ways to fix this. #epf #brownssl (Joint work with @mikepo.bsky.social, Marius Momeu, Vaggelis Atlidakis, Di Jin, and Sergej Proskurin.)

cs.brown.edu

110

Vasileios Kemerlis @vkemerlis.bsky.social · 06/04/2025

✳️ SafeSLAB (ACM CCS 2024: 📄 cs.brown.edu/~vpk/papers/..., 💾 github.com/tum-itsec/sa...) — Kernel heap hardening through memory tagging. #safeslab #brownssl

cs.brown.edu

110

Vasileios Kemerlis @vkemerlis.bsky.social · 06/04/2025

✳️ xMP (IEEE S&P 2020: 📄 cs.brown.edu/~vpk/papers/..., 💾 github.com/virtsec/xmp) — Selective intra-kernel memory isolation using hardware-assisted virtualization. #xmp #brownssl

cs.brown.edu

110

Vasileios Kemerlis @vkemerlis.bsky.social · 06/04/2025

In this "tin anniversary" edition, I reflected on how OS kernel exploitation and defense have evolved over the past decade, and shared highlights from some of our recent work in the field over the last five years:

110

Vasileios Kemerlis @vkemerlis.bsky.social · 06/04/2025

📢 Honored to return to Yale University last week to speak at the Department of Computer Science colloquium on Operating Systems security -- exactly 10 years after my first talk there on the same topic! #brownssl #browncs 🧵

250