Associate Professor at @S3DatCMU @SCSatCMU & Researcher at @AntithesisHQ.
Pittsburgh, PA
Joined September 2018
- Tweets1.8K
- Following547
- Followers2.7K
- Likes4.2K
I'm quite happy to read this. Very promising directions!
With the insane frequency of security incidents we're witnessing, I really do hope verification can get ahead of exploitation sooner rather than later.
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
> “Gang, are you even reading this? be honest.” Honest answer: a language model was, and now so is everyone reading this post. Handwritten prompt injection is now in the autograder’s threat model.
--
Amazing write-up by @shrey_twr!
This semester I am TAing 17-313, the undergrad software engineering course at CMU.
I built an autograder for the course... with AI writing every line of the code. It got me thinking about what's left of software engineering now that coding is automated away.
I jotted down my thoughts here: thelastsoftwareengineer.subs…
Rohan Padhye retweeted
found a nasty bug with @AntithesisHQ. if a backup is happening during a leap second then the backup would be done/completed without doing a backup.
Time-Traveling Agents!
antithesis.com/blog/2026/tim…
The Antithesis Multiverse Debugger is now available via API and CLI :-)
What if saying `claude build me a cloud platform make no mistakes` actually worked??? With the newest Antithesis release we have closed the loop and enabled agents to ship reliable systems: antithesis.com/blog/2026/tim…
Rohan Padhye retweeted
What if saying `claude build me a cloud platform make no mistakes` actually worked??? With the newest Antithesis release we have closed the loop and enabled agents to ship reliable systems: antithesis.com/blog/2026/tim…
This is what I imagine it must feel like for @code_monet every day.
Rohan Padhye retweeted
Hot take: Want to kill paper slop? Brutally cap submissions at 1–3 per author per conference.
Force authors to filter their own noise instead of treating peer review like a lottery.
Bonus: your best work might again get the attention it deserves.
I'm excited for SPLASH/ISSTA 2026 this week!
If you wanna know what's cooking at Antithesis and why academia should tune in, I'll be giving a REBASE Industry Track on Wednesday at 1:30pm.
"Autonomous Testing with Time-Traveling Agents in a Chaotic Multiverse"
Preview:
Rohan Padhye retweeted
arXiv has updated our policy on rate limiting for all submitters.
This update was made to fairly distribute moderator time & support the arXiv community of staff, volunteers, readers & authors.
Please read our announcement to learn more: blog.arxiv.org/2026/10/01/up…
ALT arXiv logo in white over a whiteboard with math symbols.
I'm stoked to author another blog post with @AntithesisHQ !
Distributed Systems + AI + Mutation Testing = Vibes with High Confidence
Bonus: 3 serious rqlite bugs that Fable couldn't even...
Mutation testing is a powerful technique that essentially lets you test your testing. Historically, it's been extremely difficult to mutation test test suites for distributed systems -- we've just solved that with a new Antithesis agent skill that creates deep distributed system bugs, then runs them through our bug zapper for you.
Read on: antithesis.com/blog/2026/mut…
Rohan Padhye retweeted
Mutation testing is a powerful technique that essentially lets you test your testing. Historically, it's been extremely difficult to mutation test test suites for distributed systems -- we've just solved that with a new Antithesis agent skill that creates deep distributed system bugs, then runs them through our bug zapper for you.
Read on: antithesis.com/blog/2026/mut…
Everyone quoting this post, so here goes me.
I'm quite excited for AI + formal methods overall.
Folks pointing out that vibe spec'ing is sketchy, but I'm fine if they're made open.
Perhaps we need Application Verification Interfaces (AVIs) for agents to inspect claims.
I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached.
TLA+ also works well. I sometimes combine Lean and TLA+ to look for issues around data flow, concurrency, and state mgmt.
I don't know either language well, but Claude is excellent at both. This approach is super useful for formally modeling your code and finding bugs that a human probably wouldn't have spotted.
Is formal verification the future of coding (or at least, bug finding)?
Another anecdote. We got *invited* to submit an extended version of a Distinguished (conference) Paper from a April 2025 conference, prepared in 2024. Despite the invitation, it is still under review 1.5 years later, and when it eventually gets published it will be 2+ years old.
Journals today may be rather disorganized. We submitted a paper to a top journal in my field (ACM Transactions on Computer Systems or TOCS) in Jan 27. 8 months passed and it is still "under review." No ETA or any information.
Fortunately, @tylergu_jiawei has graduated with a great PhD thesis so the unresponsiveness doesn't matter.
I don't want to overly generalize it. But, if students need pubs to convince committees, find a more responsive journal or be aware of the risk.
Rohan Padhye retweeted
many people seem to be upset by >50k submissions to ICLR. i think we should celebrate it! more people are doing AI (including AI safety) research - isn't it amazing?
as for how to navigate the growing amount of papers, luckily, Larry Wasserman formulated a solution back in 2012!
Rohan Padhye retweeted
This pains me @sigsoft : but it may be time to revert to de-anonymized review. I want to know who's throwing all these papers with AI-written, incomprehensible thickets of words at us.
I've got half a mind to decline to review for conferences that don't de-anonymize authors
Rohan Padhye retweeted
TMLR has faced a deluge of submissions, necessitating stricter desk rejection policies due to limited reviewer capacity
Co-EiC Nihar Shah reached out to authors of 10 papers slated for desk reject. Could they answer questions about their *own* submission?
medium.com/@TmlrOrg/asking-a…
I've gotten into several debates about SE conferences and peer review recently, so I decided to pen down my thoughts. Feedback or rage welcome.
Article: rohan.padhye.org/posts/peer-…
(Graphic is AI-generated)
💯 - The SE conference experience (attending, reviewing, authoring) has gone totally downhill and my lab is aligning more with PL and Systems these days.
