Algorand Foundation has launched AC2, an open protocol designed to make AI agent approvals more secure by keeping private keys under the user’s control. In the foundation’s public announcement, AC2 is described as a standard for direct, encrypted communication between users and AI agents that lets agents request signing actions without handing over wallet access.

The launch centers on a familiar problem in AI-driven workflows: agents may need to sign payments, code commits or other digital actions on a user’s behalf, but the surrounding messaging tools do not provide cryptographic verification or scoped approval. Algorand says AC2 is meant to close that gap by allowing the user to review and approve each requested action through their own wallet interface.

How AC2 is described to work

According to the foundation, AC2 establishes an end-to-end encrypted WebRTC connection between a user’s wallet or app and an AI agent. When the agent needs authorization for an action such as a payment, git commit or API request, it sends a signing request through AC2. The user then approves the action directly, while the private key remains with the user.

The announcement also says the protocol is blockchain-agnostic and open source, with both the specification and a reference implementation now available. Algorand framed AC2 as a way to support broader “agentic” workflows without requiring users to surrender full control of their accounts.

The foundation said the protocol can be extended for different message types and signing formats, which it says would make it usable in settings where agents need limited, user-approved authority rather than unrestricted access.

For now, the launch is best understood as a security and communications layer for AI agents rather than evidence that AI payments or agent-driven commerce are already widespread. The key change is narrower but concrete: users are meant to approve exactly what an agent can sign, instead of relying on chat-based instructions that can be easier to spoof or misread.

The post Algorand launches AC2 to let AI agents request user approvals without exposing private keys appeared first on The Cryptocurrency Post.

Token Terminal’s agentic payments explorer shows AI agents initiating 73.0 million stablecoin transfers over the past 180 days, with USDC accounting for virtually all of the activity.

The explorer data also points to a concentration of transfers on a small number of networks. Base led with 38.7 million transfers, while Polygon followed with 26.1 million, according to Token Terminal.

The figures describe transfer activity initiated by AI agents, not a broader measure of stablecoin adoption across the market. The available data also does not by itself explain who the agents were, what applications they were using or whether the transfers reflect sustained operational usage beyond the measured period.

Still, the numbers offer a concrete snapshot of how agent-driven payments are appearing in onchain data. In this case, the activity is largely tied to USDC and concentrated on Base and Polygon, which makes the breakdown more specific than a generic claim about AI and crypto payments.

The post Token Terminal data shows AI agents driving 73 million stablecoin transfers in 180 days appeared first on The Cryptocurrency Post.

Fetch.ai said a winning LA Hacks project called MultiEval used its specialist AI agents to test how multi-agent systems should delegate work, with the system running a 50-agent simulation of the US economy overnight.

50 specialized AI Agents ran a simulation of the US economy overnight.

A winning @LAHacks project, MultiEval, used @Fetch_ai‘s specialist agents to test how multi-agent systems should delegate work.

This system autonomously improved another Agent harness and was used to run a…

— Fetch.ai (@Fetch_ai) August 20, 2026

In a post on the company’s official X account, Fetch.ai described MultiEval as a project that used its specialist agents to evaluate other agents and improve an agent harness autonomously. The company said the setup ran through ASI:One and framed the process as “agents evaluating agents.”

How the LA Hacks project was described

According to Fetch.ai, MultiEval was used to test delegation inside multi-agent systems, with specialist agents helping run the benchmark and improve the harness used for the simulation. The company did not provide additional technical details in the post about the exact structure of the evaluation or the full scope of the overnight run.

The post adds to Fetch.ai’s broader LA Hacks presence, but the central claim is limited to the project’s use of specialist agents, the 50-agent simulation and the ASI:One workflow described by the company.

Fetch.ai also has a LA Hacks event page on its official site, which places the project in the context of its hackathon participation and agent-focused materials.

The post Fetch.ai says LA Hacks project MultiEval used specialist agents to test multi-agent delegation appeared first on The Cryptocurrency Post.