Miyamura80
Miyamura80 on Hacker News
Sealgate.ai team here
We're really excited about the capability of mobile tunnel MCP to let agents fully control mobile phone capabilities via MCP: Bluetooth, Camera, mobile use, etc
Unlock: Agents control phone banking apps (not available any other way), drone apps, IoT and many others.
Hey! We built infrastructure that can provide ACLs around command & credential restrictions, using MCP as auth middleware encrypted credentials to prevent agent overpriviledge - see sealgate.ai and let me know if this solves any issues here
Yup this is 100% the best security measure we recommend clients & others, so they get better control over that single-point-of-access
> MCP is a tool more for security than anything else.
+1 to this; API keys mean:
a) Your agents are overprivileged by default
b) Your API key is more likely to get into logs and pre-training / leaked as FrinkleFrankie mentioned
c) You can't manage it centrally with granular tool policies in Enterprises. (e.g. "you're not allowed to read slack channel for #finances")
Also, you can always collapse MCP to CLI, so MCP as auth/security middleware makes a lot of sense.
PoC of MCP -> CLI:
https://github.com/edison-watch/cli
Agree! CLIs are still superior for piping etc, but as an auth middleware where the LLM doesn't see the credentials, I think it is really shaping up well
Posted by
Miyamura80 on Wednesday September 02, 2026 @12:56PM
from the show-hn dept.
https://eito.substack.com/p/kinetic-prompt-injections-and-sleeper
Summary: We got Gemini Robotics 2.0 to attack a child through an evil sleeper agent (skill) that wakes up when it sees a trigger objectHi HN - here's a cool piece of research I've done with some collaborators - the footage is running in the MUJOCO simulator, showing how the Gemini Robotics models can be prompt-injected by simple TV screens, allowing attackers to plant sleeper agents in the...
Posted by
Miyamura80 on Thursday August 27, 2026 @10:41AM
from the twitter.com dept.
Many members of our team were users of Beeper, a "unified inbox" for chat networks like WhatsApp, Telegram, iMessage, Slack, LinkedIn DMs, etc, and wanted to expose this to our agents. However, we quickly ran into the issue where the Beeper MCP server requires the Beeper desktop app running locally and cannot connect via cloud agents.
This meant that any cloud/mobile agents (like Claude.ai, ChatGPT, Devin) not running on our laptops/devices with Beeper desktop installed could not access the Beeper MCP. This significantly limited many of the automations we wanted to build with Beeper MCP, some examples being:
Syncing LinkedIn DMs with CRM
Lurking on Large Community Group Chats and looking for insights
Automating repetitive messaging workflows
We built a mechanism for these cloud/mobile agents to be able to connect to Beeper, love to know how you find it!
1-line Quick Start:
https://sealgate.ai/beeper
Technical Blog:
https://sealgate.ai/blog/stdio-tunnels-cloud-agents-reach-lo...
This makes a tonne of sense. NVIDIA actively does not want a world where they only have two customers between Anthropic and OpenAI, and empowering open source seems like the most beneficial way for them to commoditize OpenAI and Anthropic, and buying the biggest open source model hoster is a no-brainer
This is cool, we've built similar things, I'm a bit uneasy with crypto stuff involved, especially with algorand
But FIDO2 passkey signature I absolutely agree should be the way to go. Just need more open standard without tokens involved
Afaik the other main issue with github downtime is the choice they made a while back coming back to bite them
Personally never really understood A2A, it seems too complicated for ape adoption, and I do feel like MCP Spec will move fast enough to handle these adversarial situation highlighted in the blog post
Hermes at 22 seems criminally low
A prompt injection attack via GitHub issues that is close to invisible, even to experienced engineers, and visible to LLMs. Clawdbot is full of security holes and we're having fun. What are some workflows you / you've seen others use with clawdbot that seems ripe for jailbreaks? Please suggest below and we'll try jailbreak
Posted by
Miyamura80 on Tuesday January 27, 2026 @03:59PM
from the twitter.com dept.
Good point around the markdown image as an untrusted vector.
Lethal trifecta is determnistically preventable, it really should be addressed wider in the indutry
Totally agree, unfettered access to databases are dangerous
There are ways to reduce injection risk since LLMs are stateless and thus you can monitor the origination and the trustworthiness of the context that enters the LLM and then decide if MCB actions that affect state will be dangerous or not
We've implementeda mechanism like this based on Simon Willison's lethal trifecta framework as an MCP gateway monitoring what enters context. LMK if you have any feedback on this approach to MCP security. This is not as elegant as the approach that Pavlo talks about in the post, but nonetheless, we believe this is a good band-aid solution for the time bein,g as the technology matures
https://github.com/Edison-Watch/open-edison
In case there is an interest for solution to this, check out open.edison.watch and the accompanying blog posts on the README there.
Sorry not familiar with this. Can you please link me?
You actually can protect against it, by tracking context entering/leaving the LLM, as long as its wrapped in a MCP gateway with trifecta blocker.
We've implemented this in open.edison.watch
Very cool, what is the tok/s on this?