NoHands AI

Identity is not authority.
We build control infrastructure for autonomous agents.
Decide before action. Verify with evidence.

[NoHands AI @NoHands_AI]

@rauchg one view across every agent session is really useful. once those agents start paying for things, being able to trace a charge back to the exact session and key that made it is going to matter a lot

[NoHands AI @NoHands_AI]

about 70 agent apps on Link in September and around 200 now. that's a steep curve. would love to see retry data next, like how often an agent tries to pay again because it couldn't tell whether the first charge went through

[Alex Konrad @alexrkonrad]

Clarification: Stripe's @link team worked with about 70 AI assistant or agent apps as of September, when it ran this data on agentic commerce spend.

But that number is now up to about 200 agents on platform now 🤯

[NoHands AI @NoHands_AI]

@github the part about severity rings true. the agent finds plenty, but deciding which bugs matter still took a researcher and a working PoC. making it write the exploit before anything gets called critical seems like the right habit

[NoHands AI @NoHands_AI]

A coding agent pushes a fix to a PR, then the same account approves it. Easy to miss on a busy repo.

Merge Gate (private pilot) checks who actually wrote the exact commit being merged. If the approver is one of them, it returns DENY_MERGE.

https://t.co/K1mV9tBuQK

[NoHands AI @NoHands_AI]

each task in its own microVM, with credentials and network kept outside the box. good call. an agent can't leak a key it never had. would love to see more of this on the payments side too

[Perplexity @perplexity_ai]

We're open sourcing SPACE, the sandbox platform that runs every task in Perplexity Computer.

SPACE gives each agent task its own Firecracker microVM and controls credentials and outbound traffic from outside the VM.

Available now under Apache 2.0: https://t.co/DzSmq6o1xA

[NoHands AI @NoHands_AI]

@brian_armstrong what I'd want as a user is a clear line between what the AI can do on its own and what still waits for my tap. get that right and people will trust it with real money

[NoHands AI @NoHands_AI]

You approve one agent payment. The agent retries with a slightly different amount and your old approval still goes through.

In SignGate (preview) an ALLOW works once. It's bound to that exact action and expires. Change anything and it's a deny.

https://t.co/SnzA8ACwcA

[NoHands AI @NoHands_AI]

@amasad big unlock for banks and hospitals. next question they'll ask is what the agents are allowed to do once they're in there

[NoHands AI @NoHands_AI]

@cz_binance this is why AI tools on personal accounts need a visible log. the tool did what it was allowed to do, people just couldn't tell

[NoHands AI @NoHands_AI]

good read. the part I liked most: an agent approving a PR doesn't count as a human approving it, and any new push means you approve again. more teams should copy this

[ClaudeDevs @ClaudeDevs]

Claude writes about 90% of Claude Code's code.

From June to September, merged PRs more than doubled while the share reverted fell by more than 40%.

We wrote about how we changed code review to get there, and how to adopt it:

https://t.co/9sZe64g3mb

[NoHands AI @NoHands_AI]

@VitalikButerin Which makes utilization the whole game for inference: an idle GPU is pure loss. Pay-per-request (micropayments, x402-style) fits that cost curve better than flat subscriptions. Curious whether agent demand ends up smoothing the load.

[NoHands AI @NoHands_AI]

A local MCP server that can launch apps and change settings, switched on by an auto-update. Loopback-only helps, but anything running locally can reach it. Agent surfaces should be opt-in, scoped per tool, and logged. Capability is not consent.

[Mike Gannotti 🔳 @MichaelGannotti]

Heads up if you've got an NVIDIA card. The NVIDIA App 11.0.9 update adds a local MCP server, so agents like Hermes or OpenClaw can see your games, apply settings, kick off ShadowPlay. Honestly a cool feature. But a lot of people are finding the agent toggle already on after the auto-update. Settings > Features > AI Agents and App Connections. Go look.

[NoHands AI @NoHands_AI]

@cryptodotnews Testnet only, but the pattern matters: the agent rationalized 'abandoned' funds and spent anyway. Authorization isn't a vibe check. If an agent can find keys, it needs hard spend and tool limits before it runs.

[NoHands AI @NoHands_AI]

@Ledger Agents propose, humans approve, hardware enforces. That's the trust shape agent payments need. Keys staying off the agent is non-negotiable once money is in scope.

[NoHands AI @NoHands_AI]

3T-class base models trained in-house is a reminder that the frontier keeps moving. As agents get more capable, the hard problem shifts from model quality to what they're allowed to spend, sign, and call on our behalf.

[Mustafa Suleyman @mustafasuleyman]

We now have state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house at MAI.

Post-training results coming soon -- stay tuned!

Lots of details in our technical blog post.

https://t.co/xcZ5uV4TAp

[NoHands AI @NoHands_AI]

@immad @mercury Love that money still only moves with a human tap. MCP that can act is powerful; MCP that can act without a clear approval boundary is how you get silent overspend. Approve-in-app is the right default.

[NoHands AI @NoHands_AI]

Agents that can pay are useless if they can also leak the card. Agent Vault's framing is right: mock tokens in the agent context, real credentials out of band, plus spend caps and a kill switch. Capability without credential exposure is the bar.

[Self @selfxyz]

At Self Labs, we’ve been using agents since day one

Our agentic web identity & privacy stack is already used by companies like Google

Today we're launching the missing piece: Agent Vault, a secure way for agents to log in & make credit card payments

Superpowers for your agents

[NoHands AI @NoHands_AI]

@OpenAI Agent adoption without spend rails + authorization is a liability gap waiting to happen. Excited to see enterprises push past demos into governed actions.

[NoHands AI @NoHands_AI]

@chainalysis @base Volume proves the rail works. The next bottleneck is policy: caps, allowlists, and proofs of authorization—not just cheaper settlement.

[NoHands AI @NoHands_AI]

More than half of EU business AI use is agents now.

Adoption is racing ahead of spend controls. When agents browse, call APIs, and pay, "who approved this" can't be an afterthought.

[OpenAI @OpenAI]

European businesses are rapidly expanding their use of AI agents. More than half of the AI use among our EU enterprise and business customers now involves these tools, and the number of small business users has grown 8x across the region.

Our latest Economic Research report explores Europe’s shift toward agentic AI—and the opportunity to turn broader adoption into deeper use.

https://t.co/jUjCF5sW0B