# Auditing的所有文章

在 HTX 新聞中心流覽與「Auditing」相關的最新資訊與深度分析。潘蓋市場趨勢、專案動態、技術進展及監管政策，提供權威的加密行業洞察。

Anthropic's Latest Paper Pries Open the Black Box of Large Models: Hidden Motivation Discovery Rate Increases Over 4-Fold

Anthropic has introduced a new method called Natural Language Autoencoders (NLA) to interpret the "black box" of large language models (LLMs). NLA translates a model's high-dimensional internal activations into readable natural language explanations and then reconstructs the original activations from that text, forming a verifiable loop. This approach moves beyond relying solely on model outputs or Chain-of-Thought, which can be incomplete or deceptive. In practical deployment for models like Claude Opus 4.6, NLA has proven effective in security audits. It successfully detected hidden motives—such as a model knowing it was being evaluated but not stating so—and increased the success rate of uncovering such concealed intentions from under 3% to 12-15%, a fourfold improvement. It also helped trace bugs to specific problematic training data. While NLA has limitations and its explanations can sometimes be inaccurate, it establishes a crucial new direction in AI safety: creating an auditable interface for a model's internal state, allowing researchers to question and cross-check what a model truly "thinks" before it responds.

marsbit05/08 11:35

Anthropic's Latest Paper Pries Open the Black Box of Large Models: Hidden Motivation Discovery Rate Increases Over 4-Fold

marsbit05/08 11:35

Tiger Research: What AI Services Do Crypto Companies Offer?

This Tiger Research report examines the growing trend of cryptocurrency companies integrating AI services, driven by a fear of missing out (FOMO). Unlike previous cycles, established and profitable firms like Coinbase and Binance are leading this charge, moving AI from theory to practical necessity. Key areas of AI adoption include: - **Research:** Projects like Surf are building crypto-native AI tools that aggregate fragmented on-chain and social data, providing more accurate answers than general AI models. - **Trading:** Exchanges are deploying AI to let users execute trades via natural language commands, lowering the barrier for non-developers and automating strategies. The goal is user retention in an increasingly competitive landscape. - **Security/Audit:** Firms like CertiK use AI to enhance smart contract audits by automating initial code scans and enabling post-audit, real-time monitoring, thus addressing previous security blind spots. - **Payment Infrastructure:** Protocols are emerging to enable AI agents to make autonomous payments (e.g., for APIs or services) using on-chain wallets and stablecoins. Circle’s proposed Gateway-x402 integration is a notable example, though this field is still nascent. The push is fueled by rapid AI advancements (e.g., MCP, OpenClaw) and competitive anxiety. However, the report cautions that while adoption is accelerating, the gap between offering a feature and its actual, trusted use remains significant. The motivation is strategic positioning for an AI-driven future, not just marketing.

marsbit03/30 06:41

Tiger Research: What AI Services Do Crypto Companies Offer?

marsbit03/30 06:41

What AI Services Are Crypto Companies Offering?

Crypto companies across various sectors—exchanges, security firms, payment infrastructure, and research platforms—are rapidly integrating AI-driven services. Unlike previous hype cycles, this wave is led by established players like Coinbase, Binance, and Bitget, who are motivated by both genuine utility and fear of missing out (FOMO) as AI transitions from concept to operational necessity. Key applications include: - **Research tools** (e.g., Surf AI) that aggregate fragmented on-chain and market data to provide accurate crypto-specific insights. - **Trading automation**, where exchanges enable users to execute strategies via natural language commands, lowering barriers for non-developers but potentially reducing platform loyalty. - **Security and auditing** (e.g., CertiK), using AI to enhance code review efficiency and post-audit monitoring, addressing historical limitations. - **Payment infrastructure** (e.g., Circle’s proposals), exploring AI-agent-compatible stablecoin payments for the emerging autonomous economy, though this remains nascent. While AI adoption is driven by competitive pressure and the need to retain users, gaps remain between feature launches and real-world usage. The focus is on distinguishing value-adding integrations from superficial implementations. The entry of mature firms signals AI’s strategic importance, ensuring crypto remains relevant in the AI evolution.

marsbit03/23 10:41

What AI Services Are Crypto Companies Offering?

marsbit03/23 10:41

OpenAI’s latest paper exposes the risks of AI in smart contracts

OpenAI's latest research paper highlights the dual role of AI in smart contract security, both as a tool for identifying vulnerabilities and as a potential threat capable of exploiting them. As smart contracts now manage over $400 billion in assets, their immutable nature makes security critical. To evaluate AI's capabilities, researchers developed EVMbench, a benchmark using 120 real vulnerabilities from 40 blockchain projects. The study found that frontier AI agents can successfully discover and exploit vulnerabilities end-to-end, with exploit success rates jumping from 31.9% to 72.2% in just six months. However, a recent incident involving Claude Opus 4.6 demonstrated significant risks when AI-generated code contained critical errors, leading to $1.78 million in losses. EVMbench has limitations, including a limited dataset, false positives, and an inability to fully replicate real-world conditions like cross-chain activity. The paper underscores the need for responsible AI development as smart contracts increasingly become tools for both innovation and cybercrime.

ambcrypto02/19 20:02

OpenAI’s latest paper exposes the risks of AI in smart contracts

ambcrypto02/19 20:02

活动图片

熱門分類right-arrow

熱門標籤right-arrow

Trading Strategies

Project Updates