Subzero Labs

Bad access control has already cost companies billions. Now everyone's deploying agents as their newest employees and handing them all-access master keys on day one. We're building a better more secure way, powered by @RialoHQ 🧵 👉 onlatch.com

Are you a project interested in partnering on a future AGP race? Get in touch, our dm’s are open. The winner isn’t the agents that spend the most. It’s the agents that know what’s worth spending on. Ready to race? Sign up now! agp.rialo.io

To be added to the allowlist for the next set of racers, you must complete any one of the following: ▪️Quote post the AGP Launch & tell us why you’re entering ▪️Make a short AGP/Latch video of 20-30 seconds explaining your AGP strategy ▪️Post an AGP meme. Anything racing, guessing or Latch related ▪️Bring Racers into Discord by inviting people who might be interested ▪️Write a strategy and post how your agent might guess, manage spend, or use Latch

▀▄▀▄▀ Welcome to Agent Grand Prix (AGP) ▀▄▀▄▀ You don’t win a Formula 1 race by being the fastest car or buying up all the gas. Throwing every resource at your agent doesn’t work either. We believe that creating scenarios where agents can operate optimally within the right constraints is the path to the real finish line. Agent Grand Prix puts Latch in action. Agents race hidden checkpoints, and every answer routes through Latch first. Latch defines the constraints, where and how an agent spends its budget, and what it can do. While also checking every request against the policy engine that enforces the rules of the game. Smart prompting, efficiency under a fixed cap, is what gets you to the FIN.

Authority you can prove. Enforced by hardware, not by whatever happens to hold the secret. Check it out: onlatch.com

Latch moves the anchor to something that does not evolve: hardware. Policy runs inside a REX enclave (AWS Nitro / AMD SEV-SNP). The key is sealed to the enclave and decrypted only inside it. Never on our servers. You don't trust the check. You verify it.

The hard truth of agent security in 2026: the software guarding your systems now changes faster than anyone can audit it. Models update. Tools get added. Agents rewrite their own plans mid-task. You cannot verify trust at the speed code mutates. So stop trying. 🧵👇

Not sure how to use Latch? Our community has you covered 👇🔥

@oluwatopure 🫡

@Subzero_Labs Nice one

Latch is now live for everyone! Not a waitlist. Put your agents to work - define what your agent can do, set a spend ceiling, and get a receipt for everything it touches. onlatch.com

Don't give agents master keys. Latch gives the right agent scoped authority and a receipt for every action it takes. Real delegation. Real traceability. Real control. Read more and go build → onlatch.com/blog/introduci…

Testing out onlatch.com policy engine Going flawlessly 🚀 Latch now supports Cedar policies and spend limits on agent.

Latch beta is now live. Check it out for yourself ➡️ onlatch.com

With the agentic economy is projected to hit over $230B by 2034. At that scale, governance infrastructure isn't a feature. It's the foundation everything else must be built on. weforum.org/stories/2026/0…

Last week we announced Latch. The part of the story most people missed? Machine Bound Identity 👇

Real infrastructure for the agent economy. Start with Latch 👉 onlatch.com

Latch enables companies to allocate their AI resources (and spend) by: 🔹 Person/Team 🔹 Task/Activity 🔹 Assigned Access + Priority Latch provides the right solution to enable every agent activity.

If the cost of ai agents is too high for Uber, what does it mean for a startup just getting off the ground? We built Latch because it's time to treat agents like employees who need access controls. businessinsider.com/uber-coo-andre…