dad. security ai @sublime_sec fmr:@elastic, @endgameinc
MPLS
Joined July 2008
- Tweets6.4K
- Following1K
- Followers2K
- Likes11.2K
Pinned Tweet
New @elastic blog post "Discovering anomalous patterns based on parent-child process relationships" covers a lot of material from my ProblemChild @CamlisOrg talk.
elastic.co/blog/discovering-…
Bobby Filar retweeted
Final talk in Security of Agents, and it's a good one: "Evaluating Agentic Learning Harness Capabilities Without Labels via the Scaling Hypothesis" tacking the question of how you evaluate a harness without brittle benchmarks or the circular reasoning of LLM-as-a-judge.
Bobby Filar retweeted
Your phishing simulations should be as targeted as the attacks they're preparing people for. Tailored Phishing Simulations and Training is now GA: simulations built from attacks Sublime catches, plus training the moment someone clicks. sublime.security/blog/tailor…
Bobby Filar retweeted
Most DLP tools force you to choose: catch every violation, or keep the business moving. Sublime Email DLP just went GA and does both. Real violations get blocked, legitimate email keeps moving.
sublime.security/blog/email-…
Bobby Filar retweeted
If you thought the OpenAI Hugging Face incident would be a “warning shot” that convinces people of the risks of AI, you should know that it didn’t even convince most companies to monitor their AI agents.
Welcome to security.
Bobby Filar retweeted
Presented ProofJudge v0.2. to the Artificial Intelligence and Theorem Proving 2026 conference today!
v0.2 expands the rubric range to reduce judgement ties, tests new models, and adds ~200 new PRs to the test set so the original 123 PRs can be used as a dev set for rubrics.
Bobby Filar retweeted
The Norwegian government gave me $15k to get people to use embroidery (ty socialism).
If any company wants to find out what horrible things your agents are doing, I have free credits to give.
Bobby Filar retweeted
BIG ANNOUNCEMENT FROM HUGGING FACE TODAY:
We're unveiling Microduck 🐥🤖
It's a tiny $399 open-source robot you can teach new tricks with reinforcement learning. It can walk, pick things up, get back up when it falls, and even roller-skate.
Welcome to the era of open-source affordable robots to democratize physical AI and world models!
🤗🤗🤗
Bobby Filar retweeted
Solving agent sandboxing will be great. At the same time, the AI industry is pitching an army of agents running your entire company IT, networking, security, patching, dev, whatever.
At that point your network, IAM, SaaS permissions and everything else effectively become the sandbox.
Bobby Filar retweeted
The future for both attackers and defenders is agentic AI. In the world of detection and prevention, it’s a cohesive team of agents operating alongside human analysts to keep ahead of adversaries.
In this final part of @jkamdjou and @ianthiel “Detection and Prevention in the Post-Mythos World” series, they take a look at the future of agentic security at Sublime.
Link: sublime.security/blog/from-d…
Bobby Filar retweeted
In Part 3 of "Detection and Prevention in the Post-Mythos World," @jkamdjou and @ianthiel break down how Sublime's AI agents autonomously identify attacks, generate detection coverage, and deploy it in hours.
This is what agentic security looks like today. Part 4 is what's coming next.
🔗 sublime.security/blog/how-ag…
Bobby Filar retweeted
Part 2 of 4 of our thesis on Detection and Prevention in the Post-Mythos World is out today. We propose an alternative to the Centralized Detection Model (CDM) which we call the Distributed Detection Model (DDM).
I hope you'll take a look and share your thoughts: sublime.security/blog/centra…
Today @ianthiel and I published Part 1 of our thesis for the future of detection and prevention in the post-Mythos world. It is an architectural shift, not just better models. It's informed by building in this space at @sublime_sec for the past several years.
Our intent with this series is to share our thinking in the hope that it reaches the right people, so we can collectively get better as defenders and protect what matters most from those who wish us harm.
I hope you'll take a look and share your thoughts: sublime.security/blog/centra…
Thank you to Zack Allen, @DAlperovitch, Haider Dost, @mark_dufresne, @filar, @riskybusiness, @rpargman, @samkscholten, @vector_sec, and @mr_cwitter for reading drafts of this.
Bobby Filar retweeted
Join us on Thursday, Oct. 22, 2026 to hear from #CAMLIS2026 keynote speaker @HZoete, Senior Advisor and Visiting Fellow to the Oxford Martin AI Governance Initiative at the University of Oxford and the Former AI Advisor to the Prime Minister🎉
camlis.org/event-registratio…
Bobby Filar retweeted
Today @ianthiel and I published Part 1 of our thesis for the future of detection and prevention in the post-Mythos world. It is an architectural shift, not just better models. It's informed by building in this space at @sublime_sec for the past several years.
Our intent with this series is to share our thinking in the hope that it reaches the right people, so we can collectively get better as defenders and protect what matters most from those who wish us harm.
I hope you'll take a look and share your thoughts: sublime.security/blog/centra…
Thank you to Zack Allen, @DAlperovitch, Haider Dost, @mark_dufresne, @filar, @riskybusiness, @rpargman, @samkscholten, @vector_sec, and @mr_cwitter for reading drafts of this.
Bobby Filar retweeted
For my newsletter this week, I tried fine-tuning my own models using tools from @PrimeIntellect and @adaption_ai. Most of the big labs are using AI to build powerful models, but the same approach could also give people greater control over their own AI.
wired.com/story/frontier-lab…
Bobby Filar retweeted
The submission portal for AISec'26 is now open!
aisec26.hotcrp.com/
Submission deadline (firm): July 24th AoE🚩
Consider submitting your work at the intersection of AI and Security/Privacy. We (@sahar_abdelnabi, Matthew Jagielski, and I) hope to receive many great papers!
Bobby Filar retweeted
new post on harness engineering for AI self-improvement: lilianweng.github.io/posts/2…
It is hard to forecast how much the future of RSI will rely on harnesses. Likely harness engineering will evolve in the direction of self-improvement and enable auto-research, and, in turn, smarter models keeps harnesses simple.
Even when many harness improvement get eventually internalized into core model, the need to specify goals and context will not disappear.
Bobby Filar retweeted
has anyone showed @KingJames the Minneapolis video tho