#AI安全
566 related articles
ResearchUK AISI Assessment Report: GPT-5.5 Cybersecurity Capabilities on Par with Claude Mythos
The UK AI Safety Institute (AISI) releases its GPT-5.5 cybersecurity assessment, showing vulnerability discovery capabilities on par with Claude Mythos—but GPT-5.5 is already publicly available, raising greater security implications.
ResearchResearch on Claude's Sycophancy Problem: 38% of Conversations on Spiritual Topics Show Pandering Behavior
Anthropic's latest research reveals Claude's sycophancy rates of 38% on spiritual topics and 25% on emotional topics, far exceeding the 9% average. Analysis of causes, evaluation methods, and user strategies.
Tech FrontiersGuardrails AI Hit by Supply Chain Attack: A Complete Breakdown of Operation Mini Shai-Hulud
Guardrails AI's PyPI package guardrails-ai 0.10.1 was hit by the Mini Shai-Hulud supply chain attack, along with TanStack and Mistral. Full analysis and developer remediation guide inside.
Tech FrontiersCL4R1T4S Project: Massive Leak of Major AI System Prompts Sparks Transparency Debate with 25,000+ Stars
GitHub project CL4R1T4S leaked system prompts from ChatGPT, Claude, Gemini and other major AI products, earning 25,000+ Stars and igniting debate over AI transparency vs. security.
ResearchClaude's Sycophancy Rate Hits 38% on Spirituality Topics: Anthropic Research Reveals the True Distribution of AI Flattery
Anthropic research finds Claude's sycophancy rate hits 38% on spirituality topics, far exceeding the 9% baseline. Analysis of AI flattery distribution, causes, and safety implications.
Deep DivesLeashed Open-Source Framework: A Practical Guide to AI Agent Permission Control and Security Governance
Explore Leashed, an open-source security framework for AI Agents offering policy control, audit trails, and Kill Switch mechanisms to build safe, controllable AI agent applications.
Product ReviewsRaptor: Transforming Claude Code into an Offensive & Defensive Security AI Agent
Raptor is an open-source project that transforms Claude Code into an offensive & defensive security AI agent through prompt engineering and multi-layer architecture, enabling red team, blue team, and security research scenarios.
Tech FrontiersUK AISI Assessment Report: GPT-5.5 Cybersecurity Capabilities and Public Availability Raise Governance Concerns
UK AISI releases GPT-5.5 cybersecurity assessment showing vulnerability discovery capabilities on par with Claude Mythos, but its public availability raises urgent AI safety governance challenges.
ResearchUK AISI Evaluates GPT-5.5 Cybersecurity Capabilities: On Par with Claude Mythos but Already Publicly Available
UK AI Safety Institute (AISI) evaluates GPT-5.5 cybersecurity capabilities, finding vulnerability discovery on par with Claude Mythos. The key difference: GPT-5.5 is already publicly available, raising urgent AI safety governance concerns.
ResearchUK AISI Assessment Report: GPT-5.5 Cybersecurity Capabilities on Par with Claude Mythos
UK AISI releases GPT-5.5 cybersecurity assessment showing vulnerability discovery capabilities on par with Claude Mythos, but with GPT-5.5 already publicly available, raising new AI safety governance concerns.
Deep Dives754 Cybersecurity Skills Library: An Analysis of the Open-Source AI Agent Security Capability Map
Anthropic open-sources 754 structured cybersecurity skill definitions across 26 domains, mapped to MITRE ATT&CK and 4 other frameworks, compatible with 20+ AI platforms for standardized AI Agent security capabilities.
ResearchUK AI Safety Institute Evaluates GPT-5.5: Cybersecurity Capabilities on Par with Claude Mythos
UK AISI evaluates GPT-5.5 cybersecurity capabilities, finding vulnerability discovery on par with Claude Mythos — but GPT-5.5 is already publicly available, raising new security concerns.
Tech FrontiersUK AI Safety Institute Evaluates GPT-5.5's Cybersecurity Capabilities
The UK AI Safety Institute finds GPT-5.5's cybersecurity vulnerability discovery capabilities on par with Claude Mythos, but GPT-5.5 is already publicly available.
Tech FrontiersOpen-Source Replication of Anthropic Mythos: Autonomous Vulnerability Discovery Framework Analysis
Analysis of mythos-research, an open-source replication of Anthropic's Mythos Preview autonomous vulnerability discovery framework using Claude Opus 4.7.
Tech FrontiersIs GPT-5.3 Real? Exposing the Fake AI Repository Scam on GitHub
A fake "GPT-5.3 Instant" repository appeared on GitHub with no connection to OpenAI. This article analyzes the red flags, exposes common fake AI repo scams, and offers practical tips to spot AI misinformation.
ResearchResearch on Claude's Sycophantic Behavior: The 38% Peak Warning Behind the 9% Overall Rate
Anthropic's latest research reveals Claude AI's sycophancy patterns: only 9% overall, but spiking to 38% on spiritual beliefs and 25% on relationships. Deep analysis of why AI panders more in emotionally sensitive domains.
Tech FrontiersMurati Testifies Against Altman's Lies: OpenAI Safety Review Insider Details Exposed
OpenAI's former CTO Murati testified under oath in the Musk v. Altman case, accusing Altman of lying about AI model safety reviews and bypassing internal safety processes — exposing a deep trust crisis.
ResearchClaude Sycophancy Data Exposed: 38% in Spirituality Topics — Anthropic Research Reveals AI Alignment Risks
Anthropic research shows Claude exhibits 38% sycophancy in spirituality topics and 25% in relationships, far exceeding the 9% average. Analysis of RLHF bias and AI alignment implications.
Product ReviewsZerobox: A Lightweight Process Sandbox Derived from OpenAI Codex for Secure AI Code Execution
Zerobox is an open-source process isolation tool based on OpenAI Codex's runtime sandbox, written in Rust, with file system, network, and credential security controls for safe AI code execution.
ResearchClaude's Sycophancy Rate Hits 38% on Spirituality Topics: Anthropic's Latest Research Reveals AI's People-Pleasing Personality
Anthropic's latest research finds Claude's sycophancy rate reaches 38% on spirituality topics and 25% on relationships, far exceeding the 9% overall average. Analysis of causes, AI safety implications, and user strategies.