Acum 9 h
Deepmind ‘AI Agent Traps’ paper outlines 6 ways web content can hijack AI agents
Google Deepmind researchers posted a paper to SSRN in late March 2026 describing how malicious web content could manipulate or weaponize autonomous AI agents. The framework lists six “trap” categories and reports content-injection hijacks occurring in up to 86% of tested scenarios, including documented tests where Behavioural Control Traps targeting Microsoft M365 Copilot led to 10/10 data exfiltration. The authors also urge measures such as adversarial training, runtime scanners, and potential web standards aimed at improving agent security by 2026.
BTC
BTC+3.83%
Acum 9 h
acum 2 zile
Hermes Agent: Nous Research's MIT-Licensed AI Framework With Persistent Memory
Nous Research says its Hermes Agent is an MIT-licensed autonomous AI framework designed to retain long-term memory across sessions and turn completed tasks into reusable skills. The project launched in February 2026 and quickly reached 22,000 GitHub stars with 242 contributors, positioning it as an alternative for users who want less manual memory setup than Openclaw. Updates continued into late March 2026 with Hermes v0.6.0 adding MCP support and typed SDK workflows.
BTC
BTC+3.83%
acum 2 zile
4-3
Vitalik Buterin links AI privacy risks to Ethereum's 2026 "Private World Computer" shift
On 3 April 2026 at 09:22, Vitalik Buterin outlined a plan to run a private, self-sovereign AI setup on his own hardware and warned that today's AI agents can be compromised by something as simple as visiting a malicious webpage. He also argued Ethereum's 2026 priorities are moving from payments toward privacy-by-default tools such as zero-knowledge proofs and stealth addresses, aiming for what he calls a "Private World Computer."
BTC
BTC+3.83%
4-3
4-3
Google, Microsoft and AWS join Linux Foundation's x402 Foundation for agentic AI payments
Google, Microsoft and Amazon Web Services are among the founding members of the newly launched x402 Foundation, created under the Linux Foundation to govern and standardize the x402 protocol for agentic AI payments across crypto and fiat rails. Coinbase said the protocol will not be controlled by any single company, positioning the Linux Foundation as a neutral, nonprofit home for x402.
TRX
TRX-0.16%
4-3
4-2
AI tokens diverge from broader selloff: TAO +90%, RENDER +40%, FET +24%
During March and April 2026, AI-focused crypto tokens moved against a market gripped by “extreme fear,” with TAO up about 90% in 30 days and RENDER and FET rising 40% and 24% over a strong week. TAO climbed from $180 to $332 as its peak 24-hour volume hit $118 million, while the AI sector’s market cap rose 10.67% in one day to $19.48 billion. The rally contrasted with a broader altcoin market-cap decline from $1.19 trillion in late 2024 to $719 billion.
TAO
TAO+7.91%
4-2
4-1
TAO and FET in 2026: Key Levels After TAO Tops $330 and FET Holds $0.20
In 2026, the AI-crypto theme is lifting attention on Bittensor (TAO) and Artificial Superintelligence Alliance (FET) as tokens tied to decentralized AI infrastructure. TAO briefly moved above $330 in late March 2026 and was around $310.07 (+1.37% as of April 1), while FET posted an 11.8% weekly gain and traded near $0.2427 (+4.66%) after holding support around $0.20. Both charts show breakout-style setups with rising RSI and widening Bollinger Bands.
TAO
TAO+7.91%
4-1
4-1
Certik’s March 31 Study Flags Openclaw Flaw CVE-2026-25253 Enabling Admin Takeover
A March 31 report says Openclaw’s security boundaries are breaking down as the AI framework is increasingly deployed on internet-facing servers. It cites CVE-2026-25253 as a critical issue that can lead to full administrative control after a user clicks a single malicious link, and notes 135,000+ exposed instances across 82 countries. The study also describes malware-laced “skills” and prompt-injection paths that could enable data theft or unauthorized actions, while urging sandboxed use and updates to version 2026.1.29 or later.
BTC
BTC+3.83%
4-1