EVE AI Lab, Anthropic test AI agents in persistent world; Sui pitches its chain
In brief
- EVE AI Lab and Anthropic studied AI agents in a persistent world with scarce resources.
- No direct partnership ties Sui to the study, according to recent coverage.
- Sui's materials say one programmable transaction block (PTB) can execute up to 1,024 Move functions.
- Sui's July 4 livestream test peaked at 6,086,766 TPS in a controlled demonstration.
Why a world that never resets matters
The case for persistent environments starts with how agents get tested today. Crypto Briefing put it bluntly:
Most AI testing happens in sandboxes where a bad decision costs nothing and everything resets on command.
Research involving EVE AI Lab and environments such as EVE Online shifts the weight to long-term planning and decision-making under scarcity. It's not the only effort in this lane. EVE Online announced a partnership with Google DeepMind in May 2026 to explore AI research in non-resetting economies, and Agent EVE launched a new season on September 20, 2026 (one that promotes networking and trading among agents, with permanent records of what they do).
There's no direct partnership tying Sui to the EVE AI Lab and Anthropic study in recent coverage.
What Sui is pitching to agent builders
Sui Foundation documentation describes an object-centric architecture where each asset is its own object with an owner. Programmable transaction blocks (PTBs) bundle many actions into one transaction that either fully succeeds or fully fails. A single PTB can execute up to 1,024 Move functions, according to Sui's materials.
The foundation says that setup lets agents interact with shared liquidity or state while minimizing re-entrancy risks. Scoped permissions matter too. An agent can be handed limited authority instead of full control.
The rest of the stack is spread across a few tools. Walrus handles decentralized storage for agent memory, and the sui-agent-kit covers identity and payments management. Frameworks such as WAYE.ai give developers additional resources to assemble agent systems.
A big throughput number, with a caveat
On July 4, 2026, Sui ran a public livestream experiment that peaked at 6,086,766 transactions per second, using programmable tunnels that settle transactions on mainnet. That figure came from a controlled demonstration, not ordinary network traffic. Sui also points to sub-second finality and low fees as part of its performance claims.
So where does that leave Sui? Its connection to the EVE and Anthropic research is one of fit, not of formal collaboration. Sui's materials make that case; the study itself doesn't.
Frequently asked questions
Is Sui partnered with EVE AI Lab or Anthropic on the persistent-world study?
No. Recent coverage shows no direct partnership tying Sui to the EVE AI Lab and Anthropic study. Sui's connection to that research is described as one of fit, not formal collaboration, and its own materials argue the dynamics explored can be reproduced on its chain.
What did Sui's 6 million TPS test actually show?
On July 4, 2026, Sui ran a public livestream experiment that peaked at 6,086,766 transactions per second, using programmable tunnels that settle transactions on mainnet. The figure came from a controlled demonstration, not ordinary network traffic.
How do programmable transaction blocks on Sui work for AI agents?
A programmable transaction block (PTB) bundles many actions into one transaction that either fully succeeds or fully fails. According to Sui's materials, a single PTB can execute up to 1,024 Move functions, and the foundation says the setup lets agents interact with shared liquidity or state while minimizing re-entrancy risks.
