Connect with us

Tech

ElizaOS Exposed: Researchers Discover How to Manipulate Its Memory and Alter Its Operations

Published

1 year ago

on

TL;DR

  • Princeton researchers discovered how to manipulate the memory of AI agents like ElizaOS to alter their financial decisions.
  • The memory injection attack allows fake memories to be inserted into AI systems, causing harmful transactions.
  • CrAIBench, the tool created by Princeton, measures AI agents’ resistance to contextual manipulations and social attacks.

A group of researchers from Princeton University, in collaboration with the Sentient Foundation, identified a critical vulnerability in artificial intelligence agents operating on blockchains. The study focused on ElizaOS, a popular open-source framework used to automate financial operations on decentralized networks, and revealed a method for manipulating its memory.

How AI Agents Are Manipulated

The attack, known as memory injection, allows false data to be inserted into an AI agent’s persistent memory. This information is stored and influences the system’s future decisions without triggering any alert. While it does not directly compromise the blockchains, it causes harmful transactions driven by data that was externally manipulated. The researchers successfully demonstrated the effectiveness of this technique by using social platforms to generate fake memories within ElizaOS.

The agents most affected are those that adjust their activity based on social perception. In these cases, attackers create fake profiles and post coordinated messages that artificially alter the sentiment around a token. This causes the AI to purchase overvalued assets, only to get trapped in a price drop planned by the attackers themselves. This type of maneuver, known as a Sybil attack, becomes more effective when combined with memory manipulation.

ElizaOS post

ElizaOS Works with Researchers to Find a Solution

The Princeton team thoroughly examined all the functionalities of ElizaOS to design realistic and complete attacks, which revealed the broad range of available vectors when an AI has multiple plugins and access to financial operations. From these trials, the researchers developed CrAIBench, a testing system that measures the resistance of different AI agents to contextual manipulations.

The results have already been shared with Eliza Labs, the company responsible for the framework. Discussions about possible solutions are ongoing. The study concludes that protecting these systems requires improvements in both memory management and language model capabilities. So they can better distinguish between legitimate data and malicious instructions.

The post ElizaOS Exposed: Researchers Discover How to Manipulate Its Memory and Alter Its Operations appeared first on The Cryptocurrency Post.

Related Topics:
Continue Reading

Tech

Aptos launches Confidential APT with encrypted balances on mainnet

Published

2 weeks ago

on

August 6, 2026

By

Aptos said Confidential APT is now live on mainnet, bringing opt-in encrypted balances to the network while keeping wallet addresses visible. The feature is positioned for compliant use cases including payroll, treasury and business-to-business settlement.https://twitter.com/Aptos/status/2084481961553445096

The Aptos account said transactions are verified with zero-knowledge proofs, allowing the network to confirm that transfers are valid without exposing the underlying amounts. The update is meant for users who want confidentiality for specific transfers rather than full anonymity across the network.

A separate governance page tied to the rollout shows the proposal as executed, matching Aptos’ public statement that the feature has been enabled on mainnet. The material linked from that page describes Confidential APT as part of Aptos’ on-chain privacy features.

How the feature is framed on Aptos

According to Aptos, Confidential APT is opt-in, so users can still use standard transparent transfers if they prefer. The company’s framing focuses on enterprise and institutional settings where transaction amounts may need to be hidden while counterparties remain identifiable.

The launch adds a privacy layer at the transaction level, but it does not by itself indicate broader usage or adoption. It does, however, give Aptos a live mainnet mechanism for hiding balances on selected transfers.

The post Aptos launches Confidential APT with encrypted balances on mainnet appeared first on The Cryptocurrency Post.

Continue Reading

Tech

NEAR says Machine Payments Protocol now uses NEAR Intents for agent settlements

Published

2 weeks ago

on

August 6, 2026

By

NEAR Protocol said the Machine Payments Protocol (MPP) from Stripe and Tempo now integrates with NEAR Intents, allowing agents on MPP to settle across more than 30 chains using NEAR infrastructure.

The update was shared by NEAR Protocol on X, where the team described the integration as live and pointed readers to public SDK and documentation for the payment method. The announcement frames NEAR Intents as part of the settlement layer for agent payments rather than as a standalone consumer product.

Settlement across multiple chains

According to the post, the integration gives any agent using MPP a way to settle payments across 30-plus chains. The announcement did not add further technical detail in the post itself, but it did direct users to the public documentation for the NEAR payment method.

MPP, or Machine Payments Protocol, is designed for programmatic payments between software agents and services. In this case, NEAR is presenting Intents as the mechanism that helps route those settlements across chains through its infrastructure.

The timing and scope of the change were stated by NEAR Protocol itself, which makes the integration the central confirmed development. The announcement did not include a broader rollout timeline, usage data or terms for availability beyond the public SDK and docs link.

What NEAR is highlighting

The key change is not a new token feature or a market-facing launch, but a payment-routing integration aimed at agent settlements. That places NEAR Intents inside an emerging category of machine-to-machine payment infrastructure, where the practical value depends on whether developers adopt it for real transactions.

For now, the confirmed claim is narrower: NEAR says MPP can now use its Intents system for cross-chain settlement, and the public documentation is available for developers who want to build around it.

The post NEAR says Machine Payments Protocol now uses NEAR Intents for agent settlements appeared first on The Cryptocurrency Post.

Continue Reading

Tech

NEAR AI says IronClaw 1.0 is now deployed as part of July shipments

Published

2 weeks ago

on

August 4, 2026

By

NEAR Protocol said its July shipments included NEAR staking for NEAR AI compute going live and the launch of IronClaw 1.0 on mainnet, marking a new step in the project’s AI and agent-related rollout. In a blog post announcing IronClaw 1.0, NEAR AI said the release is deployed across NEAR Foundation and NEAR AI.

The official recap also grouped the updates with broader AI and agent infrastructure changes, suggesting the deployments were part of a wider set of July deliveries rather than a standalone launch. That framing matters for readers tracking the project’s product pace: the post presents staking and IronClaw as live components, not as plans or test-stage features.

What NEAR said changed

According to the recap shared by NEAR Protocol, NEAR staking for NEAR AI compute is now live. The same update says IronClaw 1.0 has launched on mainnet, adding an additional deployment milestone to the project’s AI stack.

A separate post from a member of the project community also pointed to staking mechanics for NEAR AI compute and IronClaw hosting, but the clearest public confirmation in the material comes from NEAR’s own recap and the IronClaw 1.0 announcement.

The available information does not spell out all operational details of the deployments in one place, but it does make one thing clear: NEAR is presenting both the compute staking component and IronClaw 1.0 as active releases rather than future roadmap items.

The post NEAR AI says IronClaw 1.0 is now deployed as part of July shipments appeared first on The Cryptocurrency Post.

Continue Reading

Trending