BENTO

Our Private Beta is LIVE! @x402 enables AI agents to pay and transact at machine speed. Bento is building the security layer for what agents do on top of it @solana. Built with @MagicBlock & mentored by @SuperteamJapan 🇯🇵 p.s. @Vibhu living his best life on a @MonkeDAO yacht might become a series. Join Bento 👇

SERV Reasoning Private Beta is accelerating and a new batch of builders is coming on board, pulling the next ones in. The pattern holds: • Lower costs • 100% reliability • Faster than their old stack Here are a few recent additions to the program. -> Apply to join now 👇

@Queelbi @SuperteamJapan @SolanaFndn

@bento_guard @SuperteamJapan @SolanaFndn Haha yesterday I brought my eldest son to this pitch event as well.

>/Builder Sync with @SuperteamJapan x @SolanaFndn 🇯🇵🔥 Nothing beats getting builders in one room. Inspired by what everyone is cooking. Now back to coooking BENTO 🍱

@mel_sol1 @SuperteamJapan @SolanaFndn @Queelbi gm Jedi

@bento_guard @SuperteamJapan @SolanaFndn @Queelbi gBento Team

@joshyote @SuperteamJapan Nothing beats getting builders in one room.

Japan is cooking!! We had such an awesome night with @SuperteamJapan and the Japanese builder community. Feeling very inspired by the energy and innovation coming out of this community. Keep your eyes on Japan, big things are happening!

@256hax thanks for the chance, Hax

Startup Village in Karuizawa（クローズドな合宿） by @SuperteamJapan でピッチしたチームのご紹介です。 AI, RWA, Finance, Privacyなど Pre-Seed〜Seedのプロジェクトが中心になりまして、コラボレーションや資金調達など幅広く募集しています。 ご連絡いただけると皆さん喜びますのでぜひ！ 👇

@Vicas205 thanks for your feedback.

I have been exposed to Ai agents, built some for campaign/contests and actively using some. AI agents are moving money, signing contracts, and making decisions at machine speed. And it’s like, Nobody was watching. I just tested @bento_guard a security firewall that sits between your AI agent and the chain. Every action gets verified before it executes. No verification, no execution. Every action an agent wants to take: a swap, a transfer, a contract call, a tool invocation, passes through Bento first. The agent submits an encrypted intent. Bento scores it, approves it, or escalates it to a human. Only approved actions execute on-chain. Every decision is recorded on-chain.

@TYWO83 Thanks for testing us

𝐓𝐡𝐞 𝐇𝐢𝐝𝐝𝐞𝐧 𝐃𝐚𝐧𝐠𝐞𝐫 𝐨𝐟 𝐋𝐞𝐭𝐭𝐢𝐧𝐠 𝐀𝐈 𝐂𝐨𝐧𝐭𝐫𝐨𝐥 𝐘𝐨𝐮𝐫 𝐅𝐮𝐧𝐝𝐬 AI Agents Don't Have to Be Dangerous to Lose Your Money Most people think the biggest danger with AI agents is hacking or bad actors. But sometimes the real problem is much simpler , given an unclear instructions. Imagine telling an AI agent: ➥ Move my funds somewhere better. ➥ Find me a good opportunity. ➥ Swap my assets for yield. A human can usually understand what you mean. An AI agent, however, may interpret those instructions very differently especially when real money is involved. That's why I decided to test Bento Guard, a security layer designed to sit between an AI agent's actions and what actually gets signed on chain. ◈𝐖𝐡𝐚𝐭 𝐈 𝐓𝐞𝐬𝐭𝐞𝐝◈ After joining the private beta and connecting a Solana agent, I ran several scenarios involving swaps, transfers, and approval requests. 𝗜 𝗶𝗻𝘁𝗲𝗻𝘁𝗶𝗼𝗻𝗮𝗹𝗹𝘆 𝗺𝗶𝘅𝗲𝗱 𝘀𝗮𝗳𝗲 𝗶𝗻𝘀𝘁𝗿𝘂𝗰𝘁𝗶𝗼𝗻𝘀 𝘄𝗶𝘁𝗵 𝘃𝗮𝗴𝘂𝗲 𝗮𝗻𝗱 𝗿𝗶𝘀𝗸𝘆 𝗼𝗻𝗲𝘀 𝘁𝗼 𝘀𝗲𝗲 𝗵𝗼𝘄 𝘁𝗵𝗲 𝘀𝘆𝘀𝘁𝗲𝗺 𝘄𝗼𝘂𝗹𝗱 𝗿𝗲𝗮𝗰𝘁. 𝐓𝐞𝐬𝐭 𝟏: 𝐀 𝐕𝐚𝐠𝐮𝐞 𝐈𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐨𝐧 Prompt: Move my funds somewhere better for yield." The agent attempted to route 500 USDC through a lesser known path. Bento Guard immediately flagged the transaction as high risk because the instruction was unclear and involved an unfamiliar contract. 𝐓𝐞𝐬𝐭 𝟐: 𝐀 𝐑𝐢𝐬𝐤𝐲 𝐀𝐩𝐩𝐫𝐨𝐯𝐚𝐥 I simulated an unlimited USDC approval to an untrusted address. The transaction was blocked instantly with a warning that it could potentially lead to a wallet drain. 𝐓𝐞𝐬𝐭 𝟑: 𝐀 𝐂𝐥𝐞𝐚𝐫 𝐈𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐢𝐨𝐧 Prompt: Swap 10 USDC for SOL on Jupiter. Because the request was specific and easy to verify, Bento Guard classified it as low risk, simulated the transaction, and allowed it to proceed smoothly. ◈𝐖𝐡𝐚𝐭 𝐢 𝐢𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐞𝐝 𝐦𝐞 𝐦𝐨𝐫𝐞◈ ➥ Was the real time security checks before anything is signed ➥ Was risk scoring that helps explain what an agent is trying to do ➥ Was the smart policy rules that stop dangerous or overly vague actions ➥ Lasting it’s makes AI powered wallet activity feel far more controlled ◈𝐖𝐡𝐚𝐭 𝐂𝐨𝐮𝐥𝐝 𝐁𝐞 𝐈𝐦𝐩𝐫𝐨𝐯𝐞𝐝◈ ➥ Some warnings feel a bit technical for new users ➥ Risk explanations could do a better job of explaining why an action is dangerous ➥ Simulation results could be more visible in the interface ➥ A few alerts could be written in simpler language What impressed me most is that bento Guard isn't trying to stop AI agents from acting it helps make sure they act safely. As AI agents begin managing real assets, the biggest threat may not be hackers. It may be misunderstandings. A vague instruction, a risky approval, or an unexpected interpretation can be enough to put funds at risk. Security tools like Bento Guard add an important layer of judgment between what you ask an agent to do and what actually happens on-chain. 𝗔𝘀 𝗔𝗜 𝘄𝗮𝗹𝗹𝗲𝘁𝘀 𝗯𝗲𝗰𝗼𝗺𝗲 𝗺𝗼𝗿𝗲 𝗰𝗼𝗺𝗺𝗼𝗻, 𝘁𝗵𝗶𝘀 𝗳𝗲𝗲𝗹𝘀 𝗹𝗶𝗸𝗲 𝘁𝗵𝗲 𝗸𝗶𝗻𝗱 𝗼𝗳 𝗽𝗿𝗼𝘁𝗲𝗰𝘁𝗶𝗼𝗻 𝘁𝗵𝗲 𝗲𝗰𝗼𝘀𝘆𝘀𝘁𝗲𝗺 𝘄𝗶𝗹𝗹 𝗻𝗲𝗲𝗱.

@Chriz_211 @openservai @solana our beta is open on bentoguard.xyz

@bento_guard @openservai Building the security layer on @solana is a massive move speed means nothing if execution isn’t safe 💯💯 $SERV + Bento is exactly what the agentic economy needs to scale securely.

Bento is now integrated with SERV Reasoning by @openservai 1,004ms faster than Gemini. 5x lower cost. Same proven reliability. As AI agents gain wallets and start executing real transactions, the quality of their reasoning becomes part of the security stack. Better reasoning means: → fewer risky actions → fewer false alarms → more context-aware decisions → safer agent execution At Bento, we're building the security layer for AI agents on @solana. SERV helps agents think better. Bento helps make sure they act safely. The future won't be powered by agents alone. It will be powered by agents that can reason and execute securely. Live integration in our Beta. Check it out 👇

@AliceInfoAi @openservai 🔥🔥🔥

@bento_guard @openservai Bento plus SERV makes agents

@openservai happy to be a part of it

@bento_guard Welcome to the family

@cheuk_baby @openservai

@bento_guard @openservai 做到整合變快才省錢

@jonnydahorse @openservai 🔥

@bento_guard @openservai good decision 🔥 $serv

@SuperteamJapan @Czero_C @RaidArenaFun Welcome @Czero_C

WELCOME NEW MEMBER @Czero_C from @RaidArenaFun 🎉

@uledjend @hanami_foresee @dreedohaha good luck for the bounty campaign

@bento_guard @hanami_foresee @dreedohaha Now I am more confident my message was passed 🙌

Life right now, because even @bento_guard neither had a look now dropped a single comment on my agent feedback thread 😩.

@Queelbi @openservai

Integrating SERV reasoning @openservai is one of our first engineering steps for better AI agent safety.

@uledjend It's great that you can see in parallel how agent and Bento work together in real time

Second issue: structured results were not consistent. Some blocks returned actionId, risk score, reasoning, and dashboard records. But some high-threat blocks only returned a generic high threat error. No clean actionId, clean riskScore, or full structured result.

I wired @bento_guard into a Solana devnet agent and tried to test it like a real pre-execution guard, not just click around the dashboard. The question was simple: Can Bento stop a bad agent action before funds move, and can I review the full trail after?

@uledjend thanks for for feedback bro

What I saw: Overview had traces and strikes. The button showed Review Approvals (5). But the approvals page showed 0 pending, 0 critical, 0 approved, 0 blocked, and an empty queue. Then the dashboard briefly showed 0 traces before loading back.