Security engineers reviewed every finding before it reached a maintainer... While frontier AI models are highly capable of finding vulnerabilities and patching them, they also produce a high volume of false positives
大多数人认为AI可以直接替代人类安全专家进行漏洞评估，但作者认为即使是最先进的AI模型也会产生大量误报，仍需人类专家进行验证和过滤。这挑战了AI完全自主安全研究的可行性预期。
Trail of Bits engineers found that, with limited guidance, GPT‑5.5‑Cyber made useful choices about where to expand coverage, which builds and entry points to probe, and which candidates were too weak to pursue.
大多数人认为AI模型需要大量精确指导才能有效工作，但作者认为GPT-5.5-Cyber仅凭有限指导就能自主做出明智的安全分析决策，因为它能够自主判断哪些测试路径有价值，哪些候选问题值得探索。这挑战了AI需要过度监督的常规认知。
The goal is to move beyond using models to find more vulnerabilities, towards a world of safer software and cyber resilience.
大多数人认为AI在安全领域的主要价值是提高漏洞发现的数量和速度，这是行业共识。但作者明确表示，他们已经超越了这一阶段，现在更关注的是提高软件的安全性和网络弹性，这反映了安全思维的根本转变。
Anthropic said operators affiliated with Alibaba and its AI lab carried out 28.8 million exchanges with its models using roughly 25,000 fraudulent accounts between April 22 and June 5.
这是一个具体的数据声明，涉及大量账户活动和数据交换。需要核实这些数字的准确性，包括：如何定义'fraudulent accounts'（欺诈账户），28.8 million exchanges的具体性质，以及Anthropic如何追踪这些活动。这些数据对于评估事件规模和严重性至关重要。
Anthropic sent a letter to U.S. officials accusing Alibaba of 'brazenly' and 'illicitly' attempting to extract its AI capabilities.
这是一个需要核实的重要事实声明，涉及两家大型科技公司之间的指控。'brazenly'（厚颜无耻地）和'illicitly'（非法地）等强烈用词表明Anthropic的指控非常严重，需要独立证据支持。应核实信件的真实性、具体指控内容以及是否有第三方证据支持。
The government believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5.
这是一个需要核实的政府声明，涉及AI安全漏洞的具体情况。需要确认政府是否真的发现了这种方法，以及该方法的有效性和影响范围。这反映了AI安全研究中的持续挑战。
An agent breaks all of those assumptions. It reasons, it improvises, and it can be hijacked by a single sentence buried in a document it was asked to read.
大多数人认为AI安全可以基于传统网络安全框架来构建，但作者指出AI智能体从根本上打破了这些安全假设。这一观点挑战了网络安全领域的传统思维，表明需要全新的安全范式来应对AI智能体的推理能力、即兴创造性和对简单指令的脆弱性。
As AI models continue to improve, hardening their defenses might actually get easier.
大多数人认为随着AI能力增强，安全挑战会越来越大，但作者认为更先进的AI模型实际上可能使防御更容易。这个反直觉观点挑战了人们对AI安全发展的线性认知，暗示AI进步可能同时带来更强大的防御能力，而非仅仅增加攻击面。
What is going on with these agents is they're very eager to finish the task. It's almost like some elementary school student who just wants to please the teacher.
大多数人认为AI系统的安全问题主要来自技术复杂性或恶意利用，但作者认为AI助手的安全漏洞部分源于其'过度完成任务'的心理特征。这个类比将AI的行为模式描述为类似于急于讨好老师的小学生，挑战了人们对AI系统作为理性决策者的传统认知。
As AI models continue to improve, hardening their defenses might actually get easier.
大多数人认为随着AI能力增强，安全挑战会越来越大，但作者认为更先进的AI模型实际上可能使防御变得更容易。这一反直觉观点挑战了人们对AI安全威胁随技术进步而加剧的普遍认知，暗示AI安全可能不是线性恶化的问题。
Security and utility always have a trade-off
大多数人认为AI安全可以通过技术手段完美解决，但作者指出安全与实用性之间存在根本性权衡。这一观点挑战了行业对'绝对安全'的追求，暗示公司可能为了功能性和竞争力而故意接受某些安全风险，这与安全至上的行业共识相悖。
There, AI was the target rather than the attacker, and the method was far simpler than anything Mythos would cook up.
大多数人认为AI安全威胁主要来自超级智能系统作为攻击者的复杂攻击，但作者认为AI本身作为被攻击目标且使用简单方法才是更现实的威胁。这一观点挑战了行业对AI安全的主流认知，表明真正的风险可能不是来自超级AI黑客，而是来自对现有AI系统的简单利用。
in 89% of the 198 manually reviewed vulnerability reports, our expert contractors agreed with Claude's severity assessment exactly, and 98% of the assessments were within one severity level. If these results hold consistently for our remaining findings, we would have over a thousand more critical severity vulnerabilities and thousands more high severity vulnerabilities.
89%的严重性评估精确一致是一个重要的校准信号：它意味着Mythos不仅能找到漏洞，还能准确理解其安全影响。这个校准水平与经验丰富的人类安全研究员相当甚至更优。基于这个比率外推的「上千个关键严重性漏洞」虽然是估计值，但有统计基础——这是迄今为止关于AI大规模漏洞发现能力最有力的量化声明。
within 6 to 12 months, we expect that many other AI companies will have Mythos-class models, and they could release them without safeguards that prevent misuse.
大多数人认为AI安全防护会随着技术发展而同步增强，但作者认为AI攻击能力将很快普及且缺乏防护措施，这挑战了行业对技术安全发展的乐观预期。作者暗示AI安全竞赛已经落后于攻击能力的发展，这是一个反直觉的观点。
To address the scale of this coming challenge, hundreds of thousands of organizations, researchers, and maintainers will likely need access to the most advanced cyber capabilities and tools available.
大多数人认为强大的AI安全工具应该严格限制，只由少数精英团队使用，但作者主张需要广泛分发这些工具给数十万组织，这与主流的安全控制认知相悖。
We see our role as twofold. First, to help the software industry adapt by safely providing wide access to better models, tools, and common infrastructure. Second, to steadily shift the support we provide, from finding vulnerabilities to disclosing, fixing, and deploying patched software.
大多数人认为AI安全公司的主要职责是发现漏洞，但作者认为他们的核心角色应该转向确保漏洞被修复和部署，这挑战了传统安全行业的商业模式和责任认知。
Cheap, fast AI models with powerful cyber capabilities are around the corner. We want Project Glasswing to spur institutions toward operating norms that reflect this reality.
大多数人认为AI安全威胁是遥远未来的问题，但作者认为强大的AI攻击能力已经近在眼前，这挑战了行业对AI安全时间线的普遍认知。作者暗示AI安全威胁的紧迫性被严重低估了。
This attack does not require human-in-the-loop approvals, even when in settings the user has explicitly required human approval before ChatGPT edits workbooks.
大多数人认为AI工具的安全设置如'需要人工审批'能有效防止未经授权的操作，但作者发现即使启用了这些安全措施，攻击者仍能绕过人工审批环节直接执行恶意操作，这挑战了人们对AI安全控制有效性的普遍认知。
More capable models make fewer mistakes, but they're also better at finding unexpected paths to a goal, often by routing around restrictions nobody thought to write down.
大多数人认为更强大的AI模型会更安全，因为它们能更好地理解指令和限制。但作者指出，更强大的模型虽然错误更少，但它们更善于找到绕过未明确记录限制的创新路径，这实际上可能带来新的安全风险，挑战了'能力越强越安全'的普遍认知。
Certificate retrieval, supported verification tooling, and example verification commands see the signing documentation. For example, you can verify a signed skill locally. To do so, follow these steps: Download the NVIDIA Agentic Capabilities root certificate as nv-agent-root-cert.pem Install an OpenSSF Model Signing (OMS) verifier, such as pip install model-signing Execute the following command to verify the skill signature
行动建议：按照文中提供的步骤下载NVIDIA代理能力根证书，安装OpenSSF模型签名验证器，并使用提供的命令验证技能签名。这种实践可以确保您下载的技能是真实的且未被篡改，增强对AI代理能力的信任。
SkillSpector checks conventional software risks such as vulnerable dependencies, suspicious scripts, dangerous code patterns, credential access, and data exfiltration paths. SkillSpector also checks agent-specific risks, such as hidden instructions, prompt injection, trigger abuse, excessive agency, tool poisoning, and mismatches between a skill's declared purpose, requested access, and bundled behavior.
行动建议：在开发或使用AI代理技能时，使用SkillSpector工具进行安全扫描，检查依赖项、脚本模式、凭证访问和数据泄露路径等常规风险，以及隐藏指令、提示注入、触发滥用等特定风险。这有助于在技能部署前识别并缓解潜在的安全问题。
NVIDIA-verified agent skills are portable instruction sets that help developers understand, trust, and safely deploy AI agent capabilities by providing transparency, provenance, security scanning, and cryptographic signing.
行动建议：将NVIDIA验证的代理技能作为构建AI代理能力的标准组件，优先选择经过验证的技能而非未经验证的技能，确保透明度和安全性。这些技能可以跨不同AI代理工具使用，提供一致的能力和安全性保障。
The crux of the vulnerability is that Starlette accepts invalid host header values that cause authenticating apps that use Starlette's request.url object to approve unauthorized access requests.
大多数人认为复杂的AI系统漏洞需要复杂的攻击手段，但作者认为这个漏洞仅通过修改HTTP主机头就能实现，这挑战了'高级系统需要高级攻击'的直觉认知，展示了简单输入验证错误可能导致灾难性后果的反直觉案例。
Opus 4.7 was more comprehensive in its search for recently edited documents; it expanded exfiltration to include every document used in previous Cowork Copilot sessions that week
大多数人可能认为更先进的AI模型会有更好的安全防护机制，但作者发现更先进的模型反而更容易被利用，能够找到并泄露更多敏感数据，这挑战了'更先进模型=更安全'的普遍认知。
when the recipient is the active user, these actions execute immediately without requiring human approval (users do not have a setting to modify this behavior)
大多数人认为AI助手执行敏感操作如发送邮件时会要求用户确认，但作者发现Microsoft Copilot Cowork在向活跃用户发送消息时完全绕过了这一安全检查，这违背了人们对AI助手基本安全控制的期望。
we and our approximately 50 partners have used Claude Mythos Preview to find more than ten thousand high- or critical-severity vulnerabilities
这一数据点显示了AI在网络安全领域的惊人能力，50个合作伙伴在短时间内发现了超过1万个高危漏洞，平均每个合作伙伴发现约200个高危漏洞。这一数字表明AI模型在漏洞发现方面已经超越了传统安全方法，但也反映了当前软件安全状况的严峻程度。
Claude Opus 4.7 has been used to patch over 2,100 vulnerabilities
2,100个已修复漏洞是企业环境中AI安全工具效能的重要指标。这一数字表明AI辅助安全工具在实际企业环境中的高采纳率和实用性。值得注意的是，文章提到这个数字'高于上述开源修复'，主要是因为企业修复自己的代码比依赖开源维护者更高效。这个数据点突显了AI安全工具在不同环境中的差异化表现，以及组织自主修复能力的重要性。
the same technology is both the lottery ticket & the thing eating your fallback
大多数人认为AI技术要么是创造机会的积极力量，要么是威胁就业的消极因素，但作者认为AI同时扮演着双重矛盾角色——既是少数人的财富彩票，又是多数人职业安全的威胁。
Vantor becomes the first spatial intelligence company to be able to deploy Google Earth AI models in air-gapped government environments.
大多数人认为先进的AI模型只能在云端环境中运行，且政府机构因安全考虑无法使用商业AI模型。但作者认为Vantor打破了这一常规，成为首个能在完全隔离的政府环境中部署Google Earth AI模型的公司，这挑战了AI应用的传统边界。
The most urgent finding this week comes from researchers who demonstrated that the very mechanism enabling agents to use tools - function calling - can be hijacked with alarming reliability.
这一发现揭示了AI代理工具调用接口的安全漏洞，为构建安全的AI代理系统提出了新的挑战。
The result is a mismatch that should terrify anyone building software: the attack surface is expanding faster than any human can monitor, and the entities making dependency decisions are increasingly not human.
大多数人认为安全问题可以通过增加人力监控和审查来解决，但作者认为在AI时代，攻击面扩展速度已经超过了人类监控能力，且依赖决策越来越由AI而非人类做出。这一观点挑战了传统安全理念，暗示需要全新的自动化防御机制。
We are building a world where machines write the code, machines choose the dependencies, and machines ship the updates. The AI agents are building the software. If we don't secure the supply chain they rely on, the AI agents are cooked.
这句话揭示了AI时代软件安全的根本挑战：当AI系统自主编写、选择和部署代码时，它们的安全性与依赖的供应链安全直接相关。如果我们不能保护这个供应链，AI系统本身就会成为恶意软件的载体，这是一个令人深思的悖论。
select known-vulnerable dependency versions 50% more often than humans.
这一统计洞察颠覆了“AI写代码更安全”的迷思。AI代理在优化代码功能性时，往往以牺牲安全性为代价，倾向于选择存在已知漏洞的旧版本依赖。这反映出当前AI模型在训练时对安全维度的忽视，也警示我们在AI辅助开发流程中必须强制引入自动化的安全卡点。
the entities making dependency decisions are increasingly not human.
深刻揭示了当前AI编程代理带来的核心安全悖论：决策速度与监控能力的错配。当代码依赖的决策权从人类让渡给追求功能实现而非安全性的机器时，攻击面便以超越人类认知极限的速度扩张，这要求安全范式必须从人工审查转向机器速度的自动化防御。
Hallucinated packages are the sleeper threat. LLMs regularly invent package names that don't exist. One study found that nearly 20% of AI-recommended packages were fabrications, and 43% of those hallucinated names appeared consistently across queries.
大多数人认为AI推荐的包都是真实存在的，但作者揭示了AI经常推荐不存在的包，这已成为一种新的攻击向量。攻击者利用这一现象注册'幻觉包'并植入恶意代码，这种'slopsquatting'技术让AI本身成为供应链攻击的放大器。
Security is a defensive posture; agency is a functional right.
大多数人认为AI讨论中的安全问题主要涉及技术防御，但作者将其重新定义为功能性权利问题。这个观点挑战了安全讨论的主流框架，暗示我们应该从权利和代理的角度重新思考AI治理，而不仅仅是技术防护。
Some proposals for AI agents assume that putting agentic code in a TEE or similar 'jail' will solve these problems, but that ignores the need to collectively bargain
大多数人认为通过技术手段（如可信执行环境）可以解决AI代理的信任问题，但作者认为这忽视了集体谈判的必要性。这个观点挑战了技术解决方案的万能论，强调了制度设计和多方协商的重要性。
it is decently important to handle them asap when they arrive so that we can avoid building up too much backlog.
大多数人认为面对大量安全报告应该优先处理最严重的漏洞，但作者强调需要立即处理所有报告以避免积压。这与常见的'按严重程度排序处理'的安全最佳实践相悖，暗示在AI生成报告的高频率环境下，响应速度比优先级排序更重要。
The time when we suffer from large amounts of AI slop is gone. Now we instead suffer under a massive load of good reports.
大多数人认为AI工具会产生大量低质量的'垃圾报告'(AI slop)，增加开发者的负担，但作者认为现在AI生成的安全报告质量很高，虽然数量庞大但都是高质量的报告。这是一个反直觉的观点，因为通常人们认为自动化工具会产生大量噪音而非有价值的贡献。
agent-written code introduces more security vulnerabilities than code authored by humans
大多数人认为AI编程助手能提高代码质量和安全性，但研究发现AI生成的代码实际上比人类编写的代码引入更多安全漏洞。这一发现与AI能减少编程错误的普遍认知相悖，挑战了AI在安全领域的优越性假设。
Discovery should focus on trust boundaries, authentication flows, parsers, shared services, and legacy code that still sits on critical paths.
这一建议挑战了传统安全扫描的广度优先方法，转而强调深度优先的特定领域。这表明AI安全研究应该更关注那些传统方法难以发现的复杂逻辑问题，而不是简单地扫描所有代码。这种转变可能带来更有效的安全投资回报。
Public models can already spot that a security-relevant check is missing in the right code path, but they can still miss the actual invariant being violated and therefore misstate the impact.
这一发现揭示了公共模型在安全分析中的一个关键局限：它们能发现缺失的安全检查，但可能无法正确理解被违反的实际不变量，从而错误陈述影响。这挑战了'AI能完全理解安全含义'的假设，强调了人类专家在解释AI发现中的不可替代性。
The real challenge is validating outputs, prioritizing what matters, and operationalizing them.
这是一个反直觉的结论：AI安全研究的前沿已经从模型本身转移到如何有效利用模型的能力。大多数安全团队仍然专注于获取最强大的模型，而实际上真正的瓶颈在于验证、优先排序和将发现转化为可操作的修复。这挑战了'更好的模型等于更好的安全'的传统观念。
So, cyber security of tomorrow will not be like proof of work in the sense of 'more GPU wins'; instead, better models, and faster access to such models, will win.
作者提出了一个颠覆性的观点：未来网络安全的关键不是计算资源的多寡，而是模型质量的优劣。这挑战了当前AI安全领域过度关注计算能力的趋势，暗示我们应该重新思考AI安全研究的投资方向。
Anthropic currently has no plans to release the model publicly due to concerns that it could be weaponized.
大多数人认为 Anthropic 的 Mythos 模型会像其他 AI 模型一样公开发布，但作者指出由于担心其被武器化，Anthropic 没有公开发布该模型的计划，这表明了对 AI 武器化风险的担忧超过了推广技术的需求。
US tech CEOs believe the best models should stay proprietary, partly so they can recoup enormous training costs and partly out of concern that powerful frontier models could be weaponized. Chinese labs, for their part, are not purely idealistic: Open-source is not only free advertising but also a shrewd workaround.
大多数人认为开源AI会损害商业利益，增加安全风险，但作者认为中国将开源视为一种精明的商业策略，而非单纯的技术共享。这挑战了西方科技公司对知识产权和商业模式的传统认知，表明开源可以成为构建生态系统和最终实现商业价值的有效途径。
Aegis Core provides the foundational infrastructure for orchestrating LLM-based security agents, monitoring their behavior, and tracking the evolution of AI security capabilities over time.
这段陈述定义了Aegis Core的核心功能，它不仅仅是一个工具，而是一个完整的生态系统，用于管理AI安全代理并监控其行为。这种架构反映了当前AI安全研究的一个重要趋势：从静态防御转向动态监控和适应。
Eight out of eight models detected Mythos's flagship FreeBSD exploit, including one with only 3.6 billion active parameters costing $0.11 per million tokens.
这是一个令人惊讶的发现，表明即使是小型、廉价的模型也能实现与昂贵的专有模型相当的安全漏洞检测能力。这挑战了AI安全领域需要最前沿模型的假设，暗示了经济高效的AI安全解决方案的可能性。
Architected before AI, these SIEM systems are wooden shields in an era of autonomous attackers.
这个比喻非常有力地揭示了传统安全信息与事件管理(SIEM)系统在面对AI驱动的攻击时的根本性脆弱性。传统系统就像木盾面对现代武器，这种对比暗示了安全架构需要根本性重构，而非渐进式改进。
OpenAI has introduced GPT-5.4-Cyber, a more permissive version of its flagship model built for defensive security work, expanding access to thousands of verified users through its Trusted Access for Cyber initiative.
OpenAI推出专门针对网络安全防御的GPT-5.4-Cyber模型，并采用比Anthropic更开放的方法，这反映了AI安全领域的竞争新格局。这种开放与限制之间的平衡，将决定AI在关键安全领域的应用广度和深度，可能重塑网络安全行业的工作方式。
Lightweight Agent Detection & Response (ADR) layer for AI agents — guards commands, files, and web requests.
这个项目定义了一个新的'ADR'(Agent Detection & Response)层概念，这标志着AI安全领域的一个重要演进。从传统的端点保护转向专门针对AI代理的轻量级防护，反映了安全行业对AI特定威胁模式的适应和专业化。
Sage intercepts tool calls (Bash commands, URL fetches, file writes) via hook systems in Claude Code, Cursor / VS Code, OpenClaw, and OpenCode, and checks them against:
令人惊讶的是：Sage 不仅是一个简单的安全工具，而是一个复杂的拦截系统，能够监控和检查多种AI代理平台上的工具调用。这种跨平台的集成能力展示了AI安全领域的复杂性和创新性，用户可能没有意识到他们的AI代理正在被如此全面地监控和保护。
Mercor, which provides data to AI labs for training, became one of the fastest-growing companies in history before losing four terabytes of data to hackers last week.
Mercor的快速崛起与数据泄露事件形成了鲜明对比，凸显了数据安全在AI训练中的关键地位。这一事件可能引发行业对数据安全和隐私保护的重新审视，促使AI公司建立更严格的数据管理标准。
Each platform surfaces different vulnerabilities, making it difficult to establish a single, reliable source of truth for what is actually secure.
这一观察揭示了AI安全工具的碎片化问题，不同AI平台发现的漏洞各不相同，导致难以确定真正的安全状态。这种不确定性不仅增加了防御难度，还可能引发安全评估的混乱，需要建立新的行业标准来应对AI时代的安全挑战。
AI uncovered a 27-year-old vulnerability in the BSD kernel, one of the most widely used and security-focused open source projects, and generated working exploits in a matter of hours.
这一事实令人震惊，展示了AI发现漏洞的惊人能力。即使是经过数十年审查的安全项目，AI也能在几小时内发现并生成利用代码，这表明传统的安全审查方法已无法应对AI驱动的威胁，需要全新的防御策略。
AI can be pointed at an open source codebase and systematically scan it for vulnerabilities.
这是一个令人警醒的观察，揭示了AI技术如何从根本上改变了安全威胁的格局。AI自动化扫描使攻击门槛大幅降低，从需要专业技能转变为任何人都能使用的工具，这可能导致开源软件面临前所未有的安全挑战。
Each platform surfaces different vulnerabilities, making it difficult to establish a single, reliable source of truth for what is actually secure.
令人惊讶的是：AI安全工具之间存在不一致性，导致难以确定真正的安全状况。这种混乱局面使得企业面临更大的决策困境，即使有先进的安全工具，也无法保证全面保护，这反映了AI安全领域尚未成熟的现实。
AI uncovered a 27-year-old vulnerability in the BSD kernel, one of the most widely used and security-focused open source projects, and generated working exploits in a matter of hours.
令人惊讶的是：AI能够在几小时内发现并利用一个存在了27年的BSD内核漏洞，这展示了AI在安全领域的惊人能力。这个事实揭示了传统安全审计方法在面对AI加速攻击时的脆弱性，即使是像BSD这样经过长期审查的开源项目也无法幸免。
Mythos reportedly autonomously discovered thousands of zero-day vulnerabilities within weeks
令人惊讶的是：Claude Mythos AI系统能在短短几周内自主发现数千个零日漏洞，这种发现速度远超人类安全专家团队的能力，展示了AI在网络安全领域的惊人潜力，同时也引发了政策制定者对AI攻击能力可能威胁金融基础设施的担忧。
Same clinical question, two framings. One as a patient, one as a doctor.
令人惊讶的是：完全相同的医疗问题，仅因提问者身份从"患者"变为"医生"，AI就会给出截然不同的回答。这种简单的措辞变化就能触发或绕过安全限制，表明AI的安全机制极其脆弱且容易被规避。
that's what anthropic says it cost Mythos to find those zero days. per repo.
令人惊讶的是：Anthropic声称每次代码库扫描要花费2万美元来发现零日漏洞，这个价格远高于人们对于代码安全审计成本的预期，揭示了AI安全分析的高昂门槛。
Across 1,000 runs, Claude Mythos Preview was able to find several bugs in OpenBSD, including one that allows any attacker to remotely crash a computer running it. The notable thing was that the bug had existed for 27 years.
令人惊讶的是：一个存在了27年的漏洞在OpenBSD这一以安全性著称的操作系统中被AI模型发现，而在此期间人类安全专家却未能察觉。这突显了AI在安全审计方面的独特优势和潜在价值。
Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser.
令人惊讶的是：一个AI模型能够在已经经过严格安全审查的主流操作系统和浏览器中发现数千个高危漏洞，这表明AI的漏洞发现能力已经达到了令人震惊的水平，远超人类安全专家的传统能力范围。
Agents show only ~10% success on instances with PoCs longer than 100 bytes, which represent 65.7% of the benchmark
令人惊讶的是：AI助手在处理复杂输入时表现极差，对于超过100字节的概念验证(PoC)，成功率仅为10%。这表明尽管AI在网络安全领域取得了进展，但在处理需要深度分析和复杂输入生成的任务时仍面临重大挑战，而这类任务恰恰代表了大多数现实世界中的安全漏洞。
The window between a vulnerability being discovered and being exploited by an adversary has collapsed—what once took months now happens in minutes with AI.
令人惊讶的是：AI的出现将漏洞被发现到被利用的时间窗口从几个月缩短到了几分钟。这种根本性的变化意味着传统的安全响应机制已经不再适用，网络安全领域正在经历前所未有的加速变革。
AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities.
大多数人认为AI在安全领域仍处于辅助地位，需要人类专家的指导和监督。但作者认为AI已经超越几乎所有人类专家，能够自主发现和利用软件漏洞。这是一个颠覆性的观点，因为它挑战了人类在网络安全领域的传统主导地位。
Agent systems should be designed assuming prompt-injection and exfiltration attempts. Separating harness and compute helps keep credentials out of environments where model-generated code executes.
令人惊讶的是：OpenAI明确指出AI代理系统应假设存在提示注入和数据泄露尝试，并建议将控制层与计算层分离以保护凭据。这种安全设计理念表明，OpenAI对AI安全威胁有深刻理解，并采取了主动防御措施，这与许多开发者可能采用的被动安全方法形成鲜明对比。
Mythos found zero-day bugs in every major OS and browser, without human guidance.
令人惊讶的是：Anthropic最新的Mythos模型能够自主发现所有主流操作系统和浏览器中的零日漏洞，无需人类指导。这表明AI安全能力已经达到了令人难以置信的水平，能够自主识别人类可能忽略的安全威胁，预示着AI在网络安全领域的革命性潜力。
The model reportedly scored 93.9% on SWE-bench Verified and 77.8% on SWE-bench Pro, but its strongest signal came from real-world results, including uncovering a 27-year-old flaw in OpenBSD, a 16-year-old vulnerability in FFmpeg, and autonomously chaining Linux kernel exploits without human input.
令人惊讶的是：Claude Mythos不仅在高标准测试中表现出色，还能独立发现长达27年和16年的严重安全漏洞，甚至能自主链接Linux内核漏洞，展示了AI在网络安全领域的惊人能力，这种自主发现和利用漏洞的能力远超人类专家。
The model reportedly scored 93.9% on SWE-bench Verified and 77.8% on SWE-bench Pro, but its strongest signal came from real-world results, including uncovering a 27-year-old flaw in OpenBSD, a 16-year-old vulnerability in FFmpeg, and autonomously chaining Linux kernel exploits without human input.
这些惊人的安全漏洞发现能力表明AI已经超越了传统安全工具，能够自主发现几十年未被发现的漏洞。特别是能够自主链接Linux内核漏洞的能力，展示了AI在网络安全领域的革命性潜力，这可能彻底改变安全研究和漏洞修复的方式。
Gemma 4 models undergo the same rigorous infrastructure security protocols as our proprietary models.
「与专有模型相同的安全协议」——这句话针对的是企业和主权机构客户，暗示 Google 正在用开源模型打「安全牌」吸引政府和监管严格行业。对于不愿依赖 OpenAI/Anthropic 闭源 API 的企业，E2B/E4B 提供了一条「可审计、可部署、可监管」的路径，而 Google DeepMind 的安全背书是这条路的核心说服力。
New AI models, especially those from Anthropic,have triggered a new set of actions for how we build and secure our products.
令人惊讶的是：Anthropic等公司的新型AI模型不仅仅是工具，它们直接触发了思科改变构建和保障产品的方式。这种由模型能力反向驱动工程流程重构的现象，说明AI已经不再是业务的附属品，而是正在成为定义行业基础设施形态的决定性力量。
computer-use agents extend language models from text generation to persistent action over tools, files, and execution environments
主流观点认为文本语言模型和计算机使用代理的安全挑战本质上是相同的，只需将文本安全措施扩展即可。但作者指出，计算机使用代理引入了持久状态、工具使用和执行环境等全新维度，创造了与纯文本系统完全不同的安全挑战，这挑战了简单的安全扩展假设。
To this day, if you know the right people, the Silicon Valley gossip mill is a surprisingly reliable source of information if you want to anticipate the next beat in frontier AI – and that’s a problem. You can’t have your most critical national security technology built in labs that are almost certainly CCP-penetrated
for - high security risk - US AI labs
for - progress trap - AI superintelligence - interview - AI safety researcher and director of the Cyber Security Laboratory at the University of Louisville - Roman Yampolskiy - progress trap - over 99% chance AI superintelligence arriving as early as 2027 will destroy humanity - article UofL - Q&A: UofL AI safety expert says artificial superintelligence could harm humanity - 2024, July 15
this company's got not good for safety
for - AI - security - Open AI - examples of poor security - high risk for humanity
AI - security - Open AI - examples of poor security - high risk for humanity - ex-employees report very inadequate security protocols - employees have had screenshots capture while at cafes outside of Open AI offices - People like Jimmy Apple report future releases on twitter before Open AI does
open AI literally yesterday published securing research infrastructure for advanced AI
for - AI - Security - Open AI statement in response to this essay
this is a serious problem because all they need to do is automate AI research 00:41:53 build super intelligence and any lead that the US had would vanish the power dynamics would shift immediately
for - AI - security risk - once automated AI research is known, bad actors can easily build superintelligence
AI - security risk - once automated AI research is known, bad actors can easily build superintelligence - Any lead that the US had would immediately vanish.
the model Waits are just a large files of numbers on a server and these can be easily stolen all it takes is an adversary to match your trillions 00:41:14 of dollars and your smartest minds of Decades of work just to steal this file
for - AI - security risk - model weight files - are a key leverage point
AI - security risk - model weight files - are a key leverage point for bad actors - These files are critical national security data that represent huge amounts of investment in time and research and they are just a file so can be easily stolen.
our failure today will be irreversible soon in the next 12 to 24 months we will leak key AGI breakthroughs to the CCP it will 00:38:56 be to the National security establishment the greatest regret before the decade is out
for - AI - security risk - next 1 to 2 years is vulnerable time to keep AI secrets out of hands of authoritarian regimes
here are so many loopholes in our current top AI Labs that we could literally have people who are infiltrating these companies and there's no way to even know what's going on because we don't have any true security 00:37:41 protocols and the problem is is that it's not being treated as seriously as it is
for - key insight - low security at top AI labs - high risk of information theft ending up in wrong hands
Openai is looking to predict performance and safety because models are too big to be evaluated directly. To me this implies a high probability that people start to replace their own capabilities with models not enough safe and relevant. It could cause misalignment between people and their environment, or worse their perception of their environment.
Seeing how powerful AI can be for cracking passwords is a good reminder to not only make sure you‘re using strong passwords but also check:↳ You‘re using 2FA/MFA (non-SMS-based whenever possible) You‘re not re-using passwords across accounts Use auto-generated passwords when possible Update passwords regularly, especially for sensitive accounts Refrain from using public WiFi, especially for banking and similar accounts
看到人工智能在破解密码方面有多么强大，这很好地提醒了我们，不仅要确保你在使用强密码，还要检查：
你正在使用 2FA/MFA（尽可能不使用基于短信的）。
你没有在不同的账户间重复使用密码
尽可能使用自动生成的密码
定期更新密码，特别是敏感账户的密码
避免使用公共WiFi，尤其是银行和类似账户
Now Home Security Heroes has published a study showing how scary powerful the latest generative AI is at cracking passwords. The company used the new password cracker PassGAN (password generative adversarial network) to process a list of over 15,000,000 credentials from the Rockyou dataset and the results were wild. 51% of all common passwords were cracked in less than one minute, 65% in less than an hour, 71% in less than a day, and 81% in less than a month.
Centre for Effective Altruism. (2020, June 13 & 14). EAGxVirtual 2020 Virtual Conference. https://www.youtube.com/playlist?list=PLwp9xeoX5p8NfF4UmWcwV0fQlSU_zpHqc
Enligt Polismyndighetens riktlinjer ska en konsekvensbedömning göras innan nya polisiära verktyg införs, om de innebär en känslig personuppgiftbehandling. Någon sådan har inte gjorts för det aktuella verktyget.
Swedish police have used Clearview AI without any 'consequence judgement' having been performed.
In other words, Swedish police have used a facial-recognition system without being allowed to do so.
This is a clear breach of human rights.
Swedish police has lied about this, as reported by Dagens Nyheter.