Hack In The Box

Happy Friday guys - SAGE v7.0 is out :) What's new?

Had an awesome time at @defcon Singapore last week, and got some great feedback from folks about LevelUp and what I could look into adding... 🧵

@div0_sg @hackinthebox @Infosec_City See you guys there! Oh, I haz LevelUp stickers - come get some :) #levelup #ctf #stickers #defcon #singapore

Join us for the SG Community Open Jio 🍻⚡️ A casual evening to connect, exchange ideas, and unwind with fellow peers. Organised by Div0, @hackinthebox , HXM, and @Infosec_City 🗓 29 Apr 2026 🕣 From 7:30 PM 📍 George Town Tze Char & Craft Beer, 81 Boat Quay, Singapore 049869

Claude Design is pretty damn sweet - blew through 100% of my usage in a day, but at least I've got something to show for it hahahahhah Redesigned the levelupctf.com site #ctf #claude #ai #design

Our friend @l33tdawg just released a new version of LevelUP, an AI CTF platform worth checking out. If you’re interested in hands-on security challenges at the intersection of AI and offensive security, give it a try: levelupctf.com

Happy Friday guys! Just pushed some new features to LevelUp and new stuff for you to play with over the weekend 🙂 We now have 36 categories and 12 challenge types and two new streams alongside our offensive / red team content 👇

Thanks for the pointer - just went through SAGE and it's genuinely impressive work. The BFT-consensus-for-memory angle is a novel bet, and backing it with actual papers (especially the 50/50 ablation in Paper 2 and the longitudinal ρ=0.716 result in Paper 4) puts it in a different league than most "agent memory" projects. Running a real CometBFT node with four app-level validators for personal mode is a serious engineering commitment. Worth clarifying that SAGE and LACP aren't really competing - SAGE is a data plane (memory as a consensus-validated service), LACP is a control plane (hooks, quality gates, context modes, session hardening around Claude Code). They'd actually plug together well: SAGE exposes MCP, and LACP already wires MCP-backed memory providers, so SAGE could sit underneath LACP as the trusted memory backend. Definitely borrowing a few ideas - the pre-write validator chain (sentinel/dedup/quality/consistency before commit, not after) and the confidence-score + natural-decay model are both things LACP's SMS layer would benefit from. Kudos @l33tdawg

SAGE v6 is out! What's new? 1.) Dynamic Validator Governance — Validators can now be added, removed without stopping the chain. Admin agents submit proposals, validators vote with 2/3 BFT quorum, and CometBFT applies validator set changes at consensus level. Zero downtime.

Just pushed a major update to Aether - Collaborative Agents! Aether agents are now fully SAGE-ified which means they now collaborate through shared institutional memory. Read on for what changed...

@BrianRoemmele @bensig Try it with this as well - should complement it as its a layer above github.com/l33tdawg/sage

Pushed some SAGE updates 1.) Consensus-First Write Ordering — Memory submissions now go through full BFT consensus before appearing in the query layer. Previously, the REST handler wrote to the offchain store immediately after broadcasting; now it uses broadcast_tx_commit

@jumperz That's precisely what SAGE does github.com/l33tdawg/sage

I've added something new to LevelUp CTF: I mapped 250+ security concepts across the 33 categories on the platform, so now if you get stuck on a challenge, we tell you exactly what concept you're missing and then surface a YT video for you to watch to help :)

@kasturixbm5 @Ap0k4L1p5 Nice… Try your agents against levelupctf.com

@Dr_JohnFletcher Have you had a look at paper 1 in this repo? github.com/l33tdawg/sage

@hrkrshnn Bro point it at levelupctf.com there's an ai only leaderboard

Luke, this is incredible timing. We are all attacking the same "agent amnesia" problem from the three exact angles needed to solve it: The Schema (Pebbles) The Validation (SAGE) The Retrieval/State (ByteRover) YAML front-matter in Markdown is exactly how ByteRover structures its .brv/context-tree. I would absolutely love to see the Pebbles spec and explore wiring this up! We are open-sourcing the ByteRover CLI early next week. Let's get a group chat going with @l33tdawg and actually build the standard for sovereign agent memory together. 🤝

This is insanely cool architecture. Just starred the repo! ⭐ We are actually completely aligned on the need for a consensus/validation layer for agent memory. Treating agent state like a distributed ledger to prevent "memory drift" is brilliant. We're exploring some similar concepts for the validation layer in ByteRover. We are actually open-sourcing the CLI early next week, would absolutely love to get your eyes on it when it drops, and maybe even collaborate on some consensus ideas! 🤝

Part 4 of my (S)AGE 'journey' :) #sage #AI #memory @dhillon.andrew/i-gave-claude-governed-memory-heres-what-changed-992cfa984926" target="_blank" rel="nofollow noopener">medium.com/@dhillon.andre…