
Building @bento_guard | Proud member @SuperteamVN, @SuperteamJapan, @sqrDAO | Prev cofounder @Ensofi_xyz
Joined September 2014
- Tweets2.3K
- Following3.3K
- Followers2.3K
- Likes11.5K
Pinned Tweet
All agents on Solana need a safety layer
We gonna work with wallets, agent framework, AI payment infra, or any protocol that let agents interact with them
What @bento_guard does as safety layer? you would avoid:
• Paid the wrong recipient
• Paid 10x the market rate for an API call
• Paid twice for the same service
• Sent money to a fake/forked on-chain service you've never used
• A compromised or hallucinating agent spending your funds
"Hey, this agent has never sent money to this address before, and it's spending 3x your daily average at 2am today and BENTO blocks it"
And it learns your patterns. Over time it stops asking about the things you always do, and flags the things you never do.
I'm not saying agents are dangerous. I'm saying agents with wallets and no guardrails are a disaster waiting to happen.
If you're building agent infra on Solana and haven't thought about security yet, hit me up. Let's talk. 🤙
everyone's shipping agent security.
permissions. identity. control planes. governance.
all four answer "is this agent allowed to do this?"
nobody answers "did you ask for this"
Your AI can sign a perfectly valid transaction that still drains you.
We wrote down how @bento_guard stops that, and what it can't stop yet.
The BENTO LLM Design Report
2 LLMs in parallel. 12 risk categories. 1 verdict before every tx.
Dree Do retweeted
Replying to @solana
Dree Do retweeted
Next picks from the @colosseum Crypto World’s Fair. This time: Infrastructure, Security & Privacy.🛡️Here is my personal pick 👇
Security:
@coldstar_sol - turns any USB drive into an offline signer for Solana. Your keys exist for a split second, then they are gone
@hackhackai - AI security audits built by Solana auditors and checked with real exploit tests
@ProofOfExploit - smart contract audits for early Solana teams that are affordable and can be trusted
@iris_intel_app - shows who is really behind a wallet or a token, and follows stolen money across chains
Stakeward by @alexandr_zhibul - keeps your staked SOL safe, even if someone steals your wallet key
Privacy:
@VisionSwapSol - a private pool, live on Solana mainnet. Send and trade with no link back to you
@TiraiProtocol - private, quantum-safe transfers between Solana and Ethereum for institutions
Puddle by @staycloakedxyz - makes onchain privacy as easy as a normal crypto transfer
p!pon by @Yu_Ayato - prove facts from your ID card, like “over 20”, without showing your name or birthday
@arcane_finance - privacy and compliance in one stack for company payments, payroll and treasury
@Valiron_grade - identity and reputation for AI agents, so businesses know which agents to trust
@akcanetwork - a VPN for people under censorship.
Developer Infrastructure:
@nori_zk - Solana programs can read Ethereum data with ZK proofs, no oracle or multisig needed
SPIN by @payritHQ - payments without internet. Phones with bank apps in Africa connect over Bluetooth
@CatScope_io - fights MEV centralization on Solana with an open market for validator access
@predictefy - one API for 16+ prediction markets, including Polymarket and Kalshi
@pantahq - build your own prediction market app through one API
@tuwa_io - open-source toolkit for wallet login and tracking transactions on Solana and EVM
@soda_labs_ - any Solana wallet or program can own an address on other chains, with no bridge
Ballista by @jacobdotsol - save complex multi-step Solana transactions as templates you can use again
@raptor_trade - real-time Solana market data, very fast and much cheaper
Honorable Mention:
@bento_guard
@fhenix
@SolanaTestForge
@stardotfun
Which project should be on this list? Tag it below 👇
Dree Do retweeted
I was murdered at the Ecosystem Call Halloween Party. Who did it?!
My prime suspects are @afscott and @hanami_foresee.
Dree Do retweeted
YC is looking for agent infrastructure.
We’re building the missing security layer for agents moving money onchain.
@bento_guard is applying this batch!
Wish us luck 🤞
At @ycombinator, we're excited to fund more crypto startups solving financial needs including AI capital raising, stablecoins, trading, and agents.
Here's our latest crypto request for startups. Our next application deadline is Monday.
Dree Do retweeted
Solana moves fast. Keeping up shouldn't take hours.
Tired of jumping between X, news sites and dashboards?
Solana Pulse is a free daily briefing covering the news, markets, DeFi and opportunities that actually matter.
Subscribe free => link in our bio.
Take a look inside ↓
Dree Do retweeted
I found myself LP manually is addicted 😗
So I add a filter to filter my manual position on the portfolio page for @lpagent_io
We made it. Wish us luck!
Ape with AI. Safely. 🍱
We just submitted Bento to @colosseum Crypto World's Fair, @SuperteamJapan 🇯🇵 & @SuperteamVN track
Your AI trades on Solana. Bento checks every trade is what you meant before your wallet signs.
Dree Do retweeted
WHY BENTO: どういう仕組みで動き、なぜBENTOを使うべきなのか
Your AI can sign a perfectly valid transaction that still drains you.
We wrote down how @bento_guard stops that, and what it can't stop yet.
The BENTO LLM Design Report
2 LLMs in parallel. 12 risk categories. 1 verdict before every tx.
We spent months researching, and building to make your agents safer
Ask: “Swap 2 SOL for this new token through Jupiter. Is it safe?”
Agent: BENTO checks available token signals like liquidity, trading volume, holder concentration, and mint or freeze authority.
It also inspects the swap route and requested permissions, then allows, flags, or blocks the transaction before your wallet signs.
Your AI can sign a perfectly valid transaction that still drains you.
We wrote down how @bento_guard stops that, and what it can't stop yet.
The BENTO LLM Design Report
2 LLMs in parallel. 12 risk categories. 1 verdict before every tx.
Dree Do retweeted
New look. Same mission. 🍱
We’ve given Bento a fresh home as we prepare for our next chapter.
Built for agents. Designed for trust.
Take a look → bentoguard.xyz
Road to @colosseum continues.
Dree Do retweeted
Road to @colosseum world fair | Week #2
This week:
✓ End-to-end demo
✓ Risk logic improved
✓ Demo Day @dev3pack + @solana Office Hours
✓ Thanks for all the valuable feedback: @JosipVolarevic2 @muse_jp_sol @HiroYuske @decentra1ized_
Next: we're taking Bento into the wild.
Something is cooking 🍱
Dree Do retweeted
“How do we know an agent is doing exactly what we asked, and nothing more?”
Mastercard asked this recently when introducing Verifiable Intent with Google.
And I've been thinking about this question a lot while building Bento.
We know how to verify a signature.
We know how to secure a wallet.
We know how to check whether a transaction is valid.
But how do we verify that the transaction is actually what the human intended?
> @AnthropicAI is talking about agents misreading user intent.
> @Microsoft Research is looking at what happens when autonomous agents get access to payments and assets.
> @Mastercard is building a trust layer around verifiable intent.
Different approaches, same signal:
As AI starts controlling money, security has to look beyond the transaction.
This is exactly the rabbit hole we're going down with @bento_guard.
Road to @colosseum, still cooking 🍱
Dree Do retweeted
LP morning and chit chat with friends 🥰
@lpagent_io bro plz host some LP livestreams
Dree Do retweeted
Let's Build in Public like this:
Road to @colosseum | BENTO
Week #1
Building trade, prediction & perps inside AI chat, with Bento checking every action before funds move.
Sharing our progress below 👇
BENTO is coming to Crypto World's Fair @colosseum
Road to @colosseum | BENTO
Week #1
Building trade, prediction & perps inside AI chat, with Bento checking every action before funds move.
Sharing our progress below 👇
Dree Do retweeted
Before you build, start here! 🛠️
@SuperteamVN official Kick-Off Workshop is happening right now. Don't build in the dark; come learn what makes a winning submission and how Road to Colosseum will support your journey.
Join the Google Meet now: 👉 meet.google.com/jgw-jcyb-rhn
#PROOFofBUILD @colosseum
🤖 Made with AI
Dree Do retweeted
I'm turning 28 soon
Builder, marketer & founder from Vietnam 🇻🇳, based in France 🇫🇷
Would love to connect with more founders, AI builders, and security researchers.
Dree Do retweeted
One line really stood out to me:
"The ability for AI to act must be matched by the ability for defensive AI to stop the breach."
A few things I believe:
• Capability will always outpace human review.
• MCP expands what agents can do but trust determines what they should do.
• The real control point is execution. That's where security should be enforced.
• Humans define the policies. Defensive AI enforces them at agent speed.
If AI can act autonomously, defensive AI has to move just as fast, with clear guardrails and human oversight where it matters
That's why we started Bento. 🍱
I read @DarioAmodei's essay calling on the labs to pace the frontier. @sama agreed. The frontier will move at whatever speed it moves. The rest of the world will not slow down. Our job in the cybersecurity community is to make sure it moves securely and safely.
Here’s what I see from the front lines:
1. The unit of threat is no longer the hacker. It's an autonomous campaign. I call it the Agent-state. We see coordinated AI agents executing attack campaigns at machine speed.
2. Sophistication is dead as an attribution signal. AI gives every criminal and lone actor elite execution. Identity, infrastructure, and intent tell you who's behind an attack. Skill doesn't.
3. Runtime is the control point. Endpoints, cloud workloads, and SaaS are the battleground. Governance documents don't stop an agent in motion. Enforcement at machine speed does.
4. Every AI agent is a privileged identity. Least privilege, short-lived credentials, traceable actions, and a kill switch. Permissions never expand because an agent decides it needs more.
5. Defense has to be autonomous but also bounded. Machine-speed response, tiered by consequence, with humans owning the high-impact calls.
6. Every failed attack should make every defender smarter. Feed what we block back into detection, across customers and models, with privacy intact. This is what CrowdStrike and NVIDIA introduced with SafeMind: an agentic, always improving model and harness protection system built for defenders.
7. The AI industrial base is critical infrastructure: weights, training clusters, APIs. Call it what it is and protect it like it is.
Pacing what comes next doesn't secure what's already here. The credible path is to deploy with proof: board-level accountability for AI security, independent external red teaming, incident disclosure, secure defaults, and controls that work in production - not on paper. Anthropic and OpenAI just committed to embedding independent evaluators with employee-level access. CrowdStrike will bring what we see from the front lines to that table.
The ability for AI to act must be matched by the ability for defensive AI to stop the breach.




























