When AI builds itself当人工智能开始自我构建

For most of AI’s history, humans drove every step in its development cycle. But at Anthropic, we are delegating a growing share of AI development to AI systems themselves, which is speeding up our work.在人工智能发展史的大部分时间里，人类主导了其开发周期的每一步。但在 Anthropic，我们正将越来越多的 AI 开发工作委托给 AI 系统本身，这正在加快我们的工作进度。

Taken far enough, and given enough compute, that trend points to an AI system capable of fully autonomously designing and developing its own successor. This is called recursive self-improvement. We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.如果这一趋势发展得足够远，并拥有足够的算力，它将指向一个能够完全自主设计和开发其后继者的 AI 系统。这就是所谓的递归自我改进。我们目前尚未达到这一阶段，递归自我改进也并非必然发生。但它到来的速度可能会超出大多数机构的预期。

Using public benchmarks and previously unreported data from within Anthropic, The Anthropic Institute is showing that AI is already accelerating the development of AI systems. To take just one example: today, Anthropic engineers on average ship 8x as much code per quarter as they did from 2021-2025.通过公开基准测试以及 Anthropic 内部此前未披露的数据，Anthropic 研究所表明，AI 已经在加速 AI 系统的开发。仅举一例：如今，Anthropic 工程师每季度提交的代码量平均是 2021 年至 2025 年期间的 8 倍。

The technical trends discussed in this piece suggest that AI systems are going to become much more capable in coming years. These trends have huge implications. AI that can build itself would be a major development in the history of technology—one that could bring enormous good for the world in science, healthcare, and beyond. But full recursive self-improvement also might increase the risks of humans losing control over AI systems. If systems are capable of fully building their own successors, the ways we secure them, monitor them, and shape their behavior all grow much more important.本文讨论的技术趋势表明，未来几年 AI 系统的能力将大幅提升。这些趋势具有深远的影响。能够自我构建的 AI 将成为技术史上的一项重大发展——它可能在科学、医疗保健及其他领域为世界带来巨大的福祉。但完全的递归自我改进也可能增加人类失去对 AI 系统控制的风险。如果系统有能力完全构建自己的后继者，那么我们确保其安全、对其进行监控以及引导其行为的方式将变得愈发重要。

Evidence from the outside world来自外部世界的证据

The rate at which AI models improve is accelerating. The length of tasks that they can reliably complete on their own has been doubling roughly every four months, up from an earlier trend of doubling every seven months. In March 2024, Claude Opus 3 could complete software tasks that take humans about four minutes to complete. A year later, Claude Sonnet 3.7 managed tasks that took about an hour and a half. A year after that, Claude Opus 4.6 managed 12-hour tasks.¹ If this trend holds, tasks that take a skilled person days could come into range this year. In 2027, AI systems could be capable of tasks that take a person weeks.AI 模型的改进速度正在加快。它们能够独立完成的任务时长大约每四个月翻一番，而之前的趋势是每七个月翻一番。2024 年 3 月，Claude Opus 3 可以完成人类需要约四分钟才能完成的软件任务。一年后，Claude Sonnet 3.7 处理的任务耗时约一个半小时。又过了一年，Claude Opus 4.6 处理了耗时 12 小时的任务。如果这一趋势持续下去，熟练人员需要数天才能完成的任务可能在今年内进入其能力范围。到 2027 年，AI 系统可能具备处理人类需要数周时间才能完成的任务的能力。

The same pattern appears on coding and research benchmarks. Benchmarks measure the performance of models in a given domain, and they’re “saturated” when models achieve close to 100% performance.² SWE-bench is a standard test of real-world software engineering: it hands a model an actual open-source codebase and a real bug report, and asks it to write a code change that fixes the issue and passes the project’s own tests. Models have gone from scoring in the low single digits to saturating the benchmark in two years.同样的模式也出现在编码和研究基准测试中。基准测试衡量模型在特定领域的表现，当模型达到接近 100% 的性能时，即被称为“饱和”。SWE-bench 是衡量真实世界软件工程水平的标准测试：它向模型提供一个实际的开源代码库和真实的错误报告，要求其编写代码更改以修复问题并通该项目的测试。模型在两年内从仅获得个位数的低分，达到了基准测试饱和。

CORE-Bench tests whether a model can reproduce existing research, a prerequisite for them to conduct original research. It gives an AI model the code and data behind a published paper, and asks it to rerun everything and confirm it can replicate the paper’s results. AI systems went from succeeding at reproducing the results roughly 20% of the time in 2024 to saturating the benchmark fifteen months later. METR, which runs the benchmark measuring how well models can complete long-duration tasks, found that Claude Mythos Preview could work for “at least” 16 hours and was “at the upper end of what [METR] can measure without new tasks.”CORE-Bench 测试模型是否能够复现现有研究，这是它们进行原创研究的前提。它向 AI 模型提供已发表论文背后的代码和数据，并要求其重新运行所有内容以确认能够复现论文结果。AI 系统在 2024 年复现结果的成功率约为 20%，十五个月后便达到了基准测试饱和。METR 运行的基准测试用于衡量模型完成长期任务的能力，结果发现 Claude Mythos Preview 能够工作“至少”16 小时，并且“处于 [METR] 在无需新任务的情况下所能测量的上限”。

Public benchmarks say a lot about the capabilities of these systems. But they can’t reveal the impact AI systems are having on speeding up AI development itself. For that, we need direct evidence from within AI companies like Anthropic.公开的基准测试在很大程度上反映了这些系统的能力。但它们无法揭示 AI 系统在加速 AI 开发本身方面所产生的影响。为此，我们需要来自 Anthropic 等 AI 公司内部的直接证据。

Evidence from within Anthropic来自 Anthropic 内部的证据

Building a frontier model takes two broad categories of work. There is engineering: writing the code, standing up the infrastructure, and overseeing the model training. And there is research: deciding what experiments to run, interpreting what comes back, and figuring out which ideas to try next.构建前沿模型需要两大类工作。一是工程：编写代码、搭建基础设施并监督模型训练。二是研究：决定进行哪些实验、解读返回的结果，以及找出接下来要尝试的想法。

Across both engineering and research, the picture is consistent. In engineering, Claude can be handed an underspecified problem and figure out how to solve it; humans supply the goal, but they no longer need to supply the method. In research, Claude can already match or outperform skilled humans at executing a well-specified experiment. However, large performance gaps persist when it comes to Claude exercising judgement in choosing goals in both engineering and research. That’s the gap between AI today and a future system that could autonomously design its own successor.在工程和研究两个方面，情况是一致的。在工程方面，Claude 可以接收一个描述不详的问题并找出解决方法；人类提供目标，但不再需要提供具体方法。在研究方面，Claude 在执行定义明确的实验时，已经能够匹配或超越熟练的人类。然而，在工程和研究中选择目标时，Claude 在运用判断力方面仍存在巨大的性能差距。这就是当今 AI 与未来能够自主设计其后继者的系统之间的鸿沟。

It’s common for employees at Anthropic to receive more open-ended and important tasks as they gain more experience. Early on, they execute a task someone else specified, like, “The export button isn’t working, please fix it.” With experience, they’re handed a goal and design the approach themselves, such as, “Investigate why the network slows down under heavy load.” At the most senior levels, they are deciding which problems are worth working on at all: “What should the team build next quarter?” We can use internal Anthropic data to see how far Claude has come in being able to handle these different kinds of tasks.Anthropic 的员工随着经验积累，通常会接到更开放、更重要的任务。起初，他们执行别人指定的任务，比如“导出按钮无法工作，请修复它”。随着经验增加，他们会接到一个目标并自行设计方案，例如“调查为什么网络在高负载下变慢”。在最高级别，他们决定哪些问题值得投入精力：“团队下个季度应该构建什么？”我们可以利用 Anthropic 的内部数据，看看 Claude 在处理这些不同类型任务方面的进步程度。

Claude writes a significant proportion of Anthropic’s code. As of May 2026, more than 80% of the code we merge into Anthropic’s codebase was authored by Claude.³ Before Claude Code launched in research preview in February 2025, this number was in the low single digits. That shift also shows up in the amount of output per engineer. Lines of code merged per engineer per day stayed constant through Anthropic’s first four years (2021-2024), then began to climb upward in 2025 when Claude began to run code rather than just suggesting it for an engineer to copy and paste. The slope steepened again in 2026 when models began to work autonomously over longer time horizons. These two inflection points are shown in the chart below. In the second quarter of 2026, the typical engineer was merging 8× as much code per day as they were in 2024.⁴ This is because much of the code is written by Claude, with the engineer directing and reviewing, rather than typing it themselves.Claude 编写了 Anthropic 代码库中很大比例的代码。截至 2026 年 5 月，我们合并到 Anthropic 代码库中超过 80% 的代码是由 Claude 编写的。在 Claude Code 于 2025 年 2 月发布研究预览版之前，这一数字仅为个位数。这种转变也反映在每位工程师的输出量上。在 Anthropic 的前四年（2021-2024 年），每位工程师每天合并的代码行数保持不变，但在 2025 年 Claude 开始运行代码（而不仅仅是建议工程师复制粘贴）时，这一数字开始攀升。2026 年，当模型开始在更长的时间跨度内自主工作时，增长斜率再次变陡。这两个拐点如下图所示。2026 年第二季度，典型工程师每天合并的代码量是 2024 年的 8 倍。这是因为大部分代码是由 Claude 编写的，工程师负责指导和审查，而不是亲自输入。

A caveat: Lines of code is an imperfect measure, as it measures quantity over quality. So 8× lines of code/engineer/day in the second quarter of 2026 is almost certainly an overstatement of the true productivity gain. Nonetheless, it indicates an acceleration. At Anthropic, we don’t reward people for how many lines of code they write; rather, team members are producing more code simply because they’re using AI systems to write more code.需要注意的一点是：代码行数是一个不完美的衡量指标，因为它衡量的是数量而非质量。因此，2026 年第二季度每位工程师每天 8 倍的代码行数几乎肯定夸大了真实的生产力增益。尽管如此，它确实表明了加速的趋势。在 Anthropic，我们不会因为员工编写的代码行数多而奖励他们；团队成员产出更多代码，仅仅是因为他们正在使用 AI 系统来编写更多代码。

The increase in lines of code written lines up with subjective impressions of large productivity increases. In a March 2026 poll of 130 employees from across Anthropic research teams, the median respondent estimated that they produced around 4x as much output with Mythos Preview as they would have without access to any AI models, on the kinds of projects they would have been working on regardless.⁵ We expect that the true degree of uplift in March was somewhat lower.⁶ Nevertheless, we find the overall claim plausible, and in line with our other observations: a significant fraction of Anthropic technical staff is accomplishing their core work multiple times faster than they could without AI assistance.代码行数的增加与人们对生产力大幅提升的主观印象一致。在 2026 年 3 月对 Anthropic 各研究团队 130 名员工的调查中，中位数受访者估计，在使用 Mythos Preview 的情况下，他们在无论如何都会从事的项目上产生的输出量，大约是如果不使用任何 AI 模型时的 4 倍。我们预计 3 月份真实的提升幅度会稍低一些。尽管如此，我们认为这一总体主张是合理的，并与其他观察结果一致：很大一部分 Anthropic 技术人员完成核心工作的速度比没有 AI 辅助时快了好几倍。

We also see evidence that people at Anthropic are using Claude to do work that simply wouldn’t have happened otherwise, like building exploratory tooling and addressing long-deferred cleanup. For example, in April 2026, Claude shipped over 800 fixes that reduced a class of API errors by a factor of one thousand. The engineer overseeing Claude estimated that a human would have taken four years to complete this work; solving other people’s bugs is slow and painstaking, and humans struggle to hold that much unfamiliar context in their head at once.
我们还看到证据表明，Anthropic 的员工正在使用 Claude 完成原本根本不会进行的工作，例如构建探索性工具和处理长期搁置的清理工作。例如，2026 年 4 月，Claude 发布了 800 多个修复程序，将一类 API 错误的发生率降低了一千倍。负责监督 Claude 的工程师估计，人类需要四年时间才能完成这项工作；解决别人的错误是缓慢且费力的，人类很难同时在脑海中保持如此多不熟悉的上下文。

The code that Claude writes is “good” and improving. “Good code” means two things: it works, and it is written in a manner that allows another engineer to understand it and build upon it. On the first criterion, the evidence is clear. The rate at which Anthropic staff correct, redirect, or take over mid-task from Claude has been falling steadily for a year, including on the most complex and open-ended tasks. This means problems with no clear specification, where the engineer isn’t sure what the answer looks like. This is evident in Claude’s success rate over time on tasks of different difficulties, as shown in the graph below. Claude writes code that works.Claude 编写的代码是“好的”并且在不断改进。“好的代码”意味着两点：它能运行，并且其编写方式使另一位工程师能够理解并在此基础上进行构建。在第一个标准上，证据很明确。一年来，Anthropic 员工在任务中途纠正、重定向或接管 Claude 工作的频率一直在稳步下降，包括在最复杂和最开放的任务中。这意味着在没有明确规范、工程师不确定答案是什么样的问题上，Claude 表现出色。这一点在下图中 Claude 在不同难度任务上的成功率随时间的变化中显而易见。Claude 编写的代码是可以运行的。

On the most open-ended tasks, Claude’s success rate reached 76% in May 2026, up 50 percentage points in six months. To give an example of tasks in this difficulty tier, a routine upgrade began crashing tens of thousands of training jobs. An engineer pointed Claude at the live incident with little more than some text content and cluster access. Working through the running jobs and testing one environment setting at a time, Claude isolated the single obscure debugging flag that was triggering the crash, reproduced it reliably, and confirmed a fix. In about two hours, Claude delivered what would normally be two to three days of work.在最开放的任务中，Claude 的成功率在 2026 年 5 月达到 76%，在六个月内提高了 50 个百分点。举一个此类难度任务的例子，一次例行升级导致数万个训练作业崩溃。一名工程师只提供了少量文本内容和集群访问权限，就让 Claude 处理这一实时事件。通过逐个测试运行中的作业和环境设置，Claude 隔离了触发崩溃的那个晦涩的调试标志，可靠地复现了该问题，并确认了修复方案。在大约两个小时内，Claude 完成了通常需要两到三天的工作量。

The second criterion is writing code that another engineer can understand and build on. Here the gap between humans and AI persists, but is closing fast. There isn’t full consensus among staff at Anthropic, but many believe that the Claude-written code was still worse in quality than human-written code at Anthropic in late 2025, and is roughly at parity today. We expect it to be better within the year.第二个标准是编写其他工程师能够理解并在此基础上进行构建的代码。在这方面，人类与 AI 之间的差距依然存在，但正在迅速缩小。Anthropic 员工之间尚未达成完全共识，但许多人认为，在 2025 年底，Claude 编写的代码质量仍不如 Anthropic 的人类编写代码，而今天已大致持平。我们预计一年内它将表现得更好。

This has changed the way that Anthropic now reviews its own code. Proposed changes to our codebase are now read by an automated Claude reviewer that looks for bugs, security flaws, and other defects before it can merge. Using this tool, we ran a retrospective analysis, and found that an automated Claude review of every change to our codebase would have caught roughly a third of the bugs behind past incidents on claude.ai before they ever reached production. The engineers who wrote that code are among the best in the world at building these systems. Claude is now catching the mistakes that they missed.这改变了 Anthropic 目前审查自身代码的方式。我们代码库的提议更改现在由一个自动化的 Claude 审查员阅读，它会在合并前查找错误、安全漏洞和其他缺陷。使用该工具，我们进行了回顾性分析，发现如果对我们代码库的每次更改都进行自动化的 Claude 审查，将能在 claude.ai 过去的事故发生前捕获约三分之一的漏洞。编写这些代码的工程师是世界上构建此类系统最优秀的人才。Claude 现在正在捕获他们错过的错误。

Claude is good at running experiments to hit a goal that someone else has set. Every time Anthropic releases a model, we run the same test: we give Claude some code that trains a small AI model, and ask it to make that code run as fast as possible while still passing the same correctness checks. The goal and the success metrics are fixed in advance, so Claude’s job is to find speedups by rewriting the code, running it, timing it, and repeating. It’s a miniature version of an experimental research loop. In May 2025, Claude Opus 4 averaged a ~3x speedup over the starting code. By April 2026, Claude Mythos Preview was achieving ~52x. For calibration, a skilled human researcher would need four to eight hours to reach 4x.⁷ In this part of the research workflow—optimizing steps within a clearly defined experiment—Claude has gone from super helpful to superhuman in under a year.Claude 擅长运行实验以实现他人设定的目标。每次 Anthropic 发布模型时，我们都会进行相同的测试：我们给 Claude 一些训练小型 AI 模型的代码，并要求它在保持相同正确性检查的同时，尽可能快地运行该代码。目标和成功指标是预先设定的，因此 Claude 的任务是通过重写代码、运行、计时和重复来寻找加速方法。这是实验性研究循环的微缩版本。2025 年 5 月，Claude Opus 4 的平均速度比初始代码快约 3 倍。到 2026 年 4 月，Claude Mythos Preview 达到了约 52 倍的提升。作为校准，一名熟练的人类研究员需要四到八个小时才能达到 4 倍的提升。在研究工作流的这一部分——即在定义明确的实验中优化步骤——Claude 在不到一年的时间里从超级助手变成了超人。

Claude is getting better at proposing its own experiments. In April 2026, Anthropic published the first demonstration of Claude running an open-ended research project end to end. Claude-powered agents were given an open problem in AI safety—roughly, can a weaker model reliably supervise a stronger one?—and were left to solve it. This involved proposing hypotheses, testing them, sharing findings with parallel agents, and iterating. The task has a clear performance “floor” and “ceiling”: the floor is how well the weak supervisor would do on its own; the ceiling is how the strong model does when trained on correct answers. Two human researchers, over about a week, recovered roughly 23% of that gap; the agents recovered 97% over 800 cumulative hours and used roughly $18,000 in compute. There are some caveats to this work; the result didn’t transfer cleanly to production-scale models, and humans still chose the problem and created the scoring rubric. But within those bounds, the agents designed every experiment themselves. Direction-setting was the only meaningful role a human played.Claude 越来越擅长提出自己的实验建议。2026 年 4 月，Anthropic 发布了 Claude 端到端运行开放式研究项目的首次演示。Claude 驱动的智能体被赋予了 AI 安全领域的一个开放性问题——大致来说，较弱的模型能否可靠地监督较强的模型？——并让它们自行解决。这涉及提出假设、测试假设、与并行智能体共享发现并进行迭代。该任务有一个明确的性能“下限”和“上限”：下限是弱监督者独立完成的效果；上限是强模型在接受正确答案训练时的表现。两名人类研究员在大约一周的时间里恢复了大约 23% 的差距；智能体在累计 800 小时内恢复了 97% 的差距，并使用了价值约 18,000 美元的算力。这项工作存在一些局限性；结果并未完全迁移到生产规模的模型中，且人类仍然选择了问题并创建了评分标准。但在这些范围内，智能体自行设计了每一个实验。设定方向是人类参与的唯一有意义的角色。

Claude is getting better at steering research sessions towards research findings. We examined real Claude Code sessions (between January and March 2026) where Anthropic researchers were working with Claude on an open-ended investigative problem, like figuring out why a training run kept crashing, or why a model scored poorly on a benchmark. In each case, we found a moment where the researcher took a detour: they pursued a direction that sent the session sideways before it eventually got back on track. We then showed various Claude models only the work from before the session went off-course and asked what it would do next. A separate Claude that was able to see how the session eventually turned out then judged whether the AI or the human suggested the better next step.⁸Claude 越来越擅长引导研究会话以获得研究发现。我们检查了真实的 Claude Code 会话（2026 年 1 月至 3 月期间），当时 Anthropic 的研究人员正在与 Claude 一起处理一个开放式的调查性问题，例如弄清楚为什么训练运行不断崩溃，或者为什么模型在基准测试中得分很低。在每种情况下，我们都发现研究人员走了一段弯路：他们追求的方向使会话偏离了轨道，直到最终回到正轨。然后，我们只向各种 Claude 模型展示会话偏离轨道之前的工作，并询问它接下来会做什么。另一个能够看到会话最终结果的 Claude 随后判断 AI 或人类谁提出了更好的下一步。

Because we deliberately picked moments (n=129) where we know the human’s choice had room for improvement, this isn’t a like-for-like comparison between model and human judgement. What these moments give us is a set of realistic, challenging situations where the right next step is not obvious, and where the human’s choice serves as a useful yardstick to compare model performance over time. On this measure, our best model in November 2025 (Opus 4.5) beat the human choice 51% of the time; in April 2026 (Mythos Preview), this grew to 64%. The day-to-day work of research is largely a chain of these next-step decisions, making this a relevant measure of the model’s ability to eventually run an investigation of its own. We view this result as an early signal that AI systems are getting better at making the kinds of judgement calls that AI research depends on.因为我们特意挑选了（n=129）我们知道人类的选择有改进空间的时刻，所以这不是模型判断与人类判断之间的同等比较。这些时刻为我们提供了一组现实且具有挑战性的情境，其中正确的下一步并不明显，且人类的选择可以作为衡量模型随时间推移性能的有用标尺。在此衡量标准下，我们 2025 年 11 月的最佳模型（Opus 4.5）在 51% 的情况下击败了人类的选择；到 2026 年 4 月（Mythos Preview），这一比例增长到 64%。研究的日常工作在很大程度上是一连串此类下一步决策，这使其成为衡量模型最终自主运行调查能力的相关指标。我们将此结果视为 AI 系统正变得越来越擅长做出 AI 研究依赖的各种判断的早期信号。

What might the future of work at Anthropic look like?Anthropic 未来的工作会是什么样子？

The evidence suggests that the human role is narrowing at each step in the AI development process. Once human- and AI-authored code quality reach parity, humans will stop writing code entirely, and shift to only reviewing it. But if they can’t review code as quickly as Claude can generate it, human review will become the bottleneck to AI development. Similarly, once Claude can run experiments, the question shifts towards “Which of these experiments is worth running?” Put simply: the doing (i.e., writing the code, running the experiment, producing the result) now costs almost nothing in human time, even if it still has costs in compute.证据表明，人类在 AI 开发过程的每一步中所扮演的角色都在缩小。一旦人类编写的代码质量与 AI 编写的代码质量持平，人类将完全停止编写代码，转而只进行审查。但如果他们审查代码的速度赶不上 Claude 生成代码的速度，人类审查将成为 AI 开发的瓶颈。同样，一旦 Claude 能够运行实验，问题就会转向“这些实验中哪些值得运行？”简而言之：执行工作（即编写代码、运行实验、产生结果）现在几乎不消耗人类时间，即使它仍然消耗算力。

An area of human comparative advantage, for now, is research taste and judgment, including choosing which problems matter, which results to trust, and when an approach is a dead end.目前，人类的比较优势在于研究品味和判断力，包括选择哪些问题重要、信任哪些结果，以及判断何时一种方法已进入死胡同。

What if we’re wrong?如果我们错了怎么办？

A natural objection to the evidence presented above is that the work that is still in human hands—choosing which problems to work on—is what matters most. Without that judgment, Claude is a capable assistant, but not a system that could drive AI progress on its own.对上述证据的一个自然反驳是，仍然掌握在人类手中的工作——即选择处理哪些问题——才是最重要的。没有那种判断力，Claude 只是一个能干的助手，而不是一个能独立推动 AI 进步的系统。

It is genuinely unclear whether today’s training methods and architectures could unlock that capacity. But AI is rarely advanced by “eureka!” moments. There have been a few of these in AI’s recent history, like the Transformer architecture, or mixture-of-experts models, but paradigm-shifting ideas arrive years apart. In between, most progress is incremental: we scale something up, see what breaks, fix it, and try again. That is exactly the kind of workflow Claude now excels at. Edison said that genius is 1% inspiration and 99% perspiration. But we see perspiration becoming increasingly automated. It’s becoming clear that much of what advances the frontier is automatable; large-scale research progress is mostly a function of tools and resources, which dictate how fast you can run experiments, how many you can run at once, and how quickly you can get results.目前尚不清楚当今的训练方法和架构是否能解锁这种能力。但 AI 很少通过“尤里卡！”时刻取得进步。在 AI 的近期历史上，确实有过几次这样的时刻，比如 Transformer 架构或混合专家模型，但范式转移的想法相隔多年才会出现。在此期间，大多数进步是渐进的：我们扩大规模，看看哪里出了问题，修复它，然后再试一次。这正是 Claude 现在擅长的工作流。爱迪生曾说，天才就是 1% 的灵感加上 99% 的汗水。但我们看到汗水正变得日益自动化。越来越明显的是，推动前沿发展的很大一部分是可以自动化的；大规模的研究进展主要取决于工具和资源，它们决定了你能以多快的速度运行实验、你能同时运行多少个实验，以及你能多快获得结果。

Even if we suppose that Claude never achieves good research taste, a conservative reading of our evidence still implies compounding acceleration. If humans spend most of their time on the single-digit fraction of work that is direction-setting, while Claude handles the rest, that means each engineer or researcher is steering far more work than before. The evidence we see suggests that people at Anthropic are both moving faster and covering a broader surface. In practice, this means that AI already makes Anthropic move much faster than it did before the advent of effective AI tools.即使我们假设 Claude 永远无法获得良好的研究品味，对我们证据的保守解读仍然意味着复合加速。如果人类将大部分时间花在仅占一小部分的方向设定工作上，而由 Claude 处理其余部分，这意味着每位工程师或研究员都在引导比以前多得多的工作。我们看到的证据表明，Anthropic 的员工不仅行动更快，而且覆盖的领域也更广。在实践中，这意味着 AI 已经使 Anthropic 的行动速度比有效 AI 工具出现之前快得多。

The less conservative reading is that the early evidence on Claude’s improving research judgment—narrow as it is today—is an indicator that this capability is improving as well. “Research taste” might be just another AI capability that AI systems fail at for a time, then get good at. We’ve seen a similar pattern with other qualitative skills, like AI systems being able to explain why a joke is funny, demonstrate theory of mind, and solve linguistic riddles.不那么保守的解读是，关于 Claude 研究判断力提升的早期证据——尽管今天还很有限——是一个指标，表明这种能力也在提升。“研究品味”可能只是 AI 系统在一段时间内表现不佳，随后变得精通的另一种 AI 能力。我们在其他定性技能上也看到了类似的模式，例如 AI 系统能够解释笑话为什么好笑、展示心智理论以及解决语言谜题。

Possible futures可能的未来

What happens next depends on two things: whether the trend continues, and what we choose to do if it does. We can imagine at least three future scenarios:接下来会发生什么取决于两件事：趋势是否会持续，以及如果趋势持续，我们选择做什么。我们可以设想至少三种未来情景：

1. The trend stalls, but today’s AI capabilities are widely diffused. This article features many exponential trajectories. But these trajectories may actually turn out to be S-curves. We may be approaching the bend in the curve, where returns to scale diminish and the line straightens, then flattens. The judgment that separates a competent researcher from a great one might be a capability that cannot come from scaling up training inputs like compute and data. If so, getting past this bottleneck would require a new idea, like an architectural approach that supplants the Transformer architecture that all current frontier models use.
   
   Alternately, the binding constraint to AI progress could be in the supply chain, not the model: advancing and diffusing the frontier may require more energy and compute than presently exists. The pace of chip fabrication, grid expansion, or interconnect bandwidth may be the constraint, rather than intelligence itself. We also cannot rule out an exogenous shock to the AI ecosystem that dramatically slows things, like a sudden diminishment in the supply of compute or electricity, either of which would slow progress and make forward investment by labs more expensive. Or we may not be anticipating some other barrier to progress.
   
   Even if model capabilities were frozen at today’s level, we would expect major changes to occur in the world. Project Glasswing is one early sign: in its first weeks, Mythos Preview found more than ten thousand high- and critical-severity software vulnerabilities across the world’s most important systems—enough that the bottleneck in cyber defense has already shifted from finding vulnerabilities to patching them fast enough. And we are still early in the diffusion of today’s models into the wider economy, where a 100-person company can increasingly do the work of a 1,000-person one, because each employee will sit atop a pyramid of agents.
   
   We include this scenario for completeness, but we don’t believe it’s likely. Every capability we can measure, including those that feel “squishier,” like quality of code and success on open-ended tasks, has so far followed the same curve. We have not yet seen that curve bend. Of the three futures we consider, this one would give governments and societies the most time to adapt. We are more worried about the next two, which would move faster and leave far less room for preparation.趋势停滞，但当今的 AI 能力得到广泛普及。本文呈现了许多指数级轨迹。但这些轨迹最终可能被证明是 S 曲线。我们可能正在接近曲线的弯曲处，规模收益递减，线条变直，然后趋于平缓。区分称职研究员和伟大研究员的判断力可能是一种无法通过扩大算力和数据等训练输入规模而获得的能力。如果是这样，克服这一瓶颈将需要新的想法，例如一种取代所有当前前沿模型所使用的 Transformer 架构的架构方法。或者，AI 进步的束缚约束可能在于供应链，而非模型：推动和普及前沿技术可能需要比目前现有的更多的能源和算力。芯片制造、电网扩展或互连带宽的速度可能是制约因素，而不是智能本身。我们也不能排除 AI 生态系统受到外源性冲击导致发展显著放缓的可能性，例如算力或电力供应突然减少，这两者都会减缓进步并使实验室的前瞻性投资变得更加昂贵。或者我们可能没有预见到其他阻碍进步的障碍。即使模型能力冻结在当今水平，我们也预计世界会发生重大变化。Project Glasswing 是一个早期迹象：在最初的几周内，Mythos Preview 在全球最重要的系统中发现了超过一万个高危和严重软件漏洞——以至于网络防御的瓶颈已经从寻找漏洞转移到了以足够快的速度修补漏洞上。而且，当今模型在更广泛经济中的普及仍处于早期阶段，一家 100 人的公司越来越能够完成 1,000 人公司的工作，因为每位员工都将坐拥一个智能体金字塔。我们为了完整性包含此情景，但我们认为它不太可能发生。我们能测量的每一项能力，包括那些感觉更“模糊”的能力，如代码质量和开放式任务的成功率，到目前为止都遵循相同的曲线。我们尚未看到该曲线弯曲。在我们考虑的三种未来中，这种未来将给予政府和社会最多的适应时间。我们更担心接下来的两种，它们的发展速度会更快，留给准备的时间会少得多。
2. AI labs continue to see compounding efficiency gains. In this scenario, AI development becomes substantially automated, but humans continue to set research directions and judge results. Organizations that use AI systems would become much more efficient as time goes on, so we could expect to see significant productivity multipliers on each person in this organization. 100-person companies could do the work of 10,000- or 100,000-person organizations. This would revolutionize knowledge work and government services, but could also be turned to harmful ends, from authoritarian surveillance of whole populations to influence operations that tailor manipulation to each individual and run at a scale no human team could match. The role of humans at companies like Anthropic would shift. People would partner with AI systems to scale up research and generate new insights, and together they would build the systems needed to verify that AI outputs can be trusted.
   
   The evidence we’ve laid out here suggests that we’re likely heading into this scenario. But speeding up one part of a process often just shifts the bottleneck elsewhere: overall pace is capped by the parts that haven’t sped up. In computing, this is known as Amdahl’s law, and the same logic can apply to organizations. Anthropic has already encountered one signature of Amdahl’s law: as we’ve begun to push more code around the organization, human code review has become a new bottleneck.
   
   We’ve also encountered this friction outside engineering. There has been an explosion of new ideas, initiatives, tools, and simulations, as a result of Anthropic employees working with highly capable models—far more than we have the capacity to pursue. The rate at which organizations can spot and fix these bottlenecks may be a skill that improves over time, and it may become the most important skill for any organization.AI 实验室继续获得复合效率收益。在这种情景下，AI 开发变得基本自动化，但人类继续设定研究方向并评判结果。随着时间的推移，使用 AI 系统的组织将变得更加高效，因此我们可以预期该组织中的每个人都会产生显著的生产力乘数。100 人的公司可以完成 10,000 人或 100,000 人组织的工作。这将彻底改变知识工作和政府服务，但也可能被用于有害目的，从对整个人口的威权监控到针对每个人量身定制操纵并以人类团队无法比拟的规模运行的影响力行动。像 Anthropic 这样的公司中，人类的角色将发生转变。人们将与 AI 系统合作以扩大研究规模并产生新的见解，他们将共同构建必要的系统，以验证 AI 的输出是否值得信任。我们在此处列出的证据表明，我们很可能正走向这种情景。但加速流程的一部分往往只是将瓶颈转移到了其他地方：整体进度受限于尚未加速的部分。在计算领域，这被称为阿姆达尔定律，同样的逻辑也适用于组织。Anthropic 已经遇到了阿姆达尔定律的一个特征：随着我们开始在组织内推动更多代码，人类代码审查已成为一个新的瓶颈。我们在工程之外也遇到了这种摩擦。由于 Anthropic 员工使用高性能模型工作，出现了大量新的想法、倡议、工具和模拟——远远超过了我们有能力追求的范围。组织发现并修复这些瓶颈的速度可能是一种随时间推移而提高的技能，它可能成为任何组织最重要的技能。
3. AI systems themselves become capable of full recursive self-improvement, and begin building their successors. If technical trends in advancing capabilities continue, and AI systems are able to develop the capabilities inherent to transformative human ingenuity, then it is plausible that AI systems could design and refine themselves.
   
   In this world, the pace of progress in AI development becomes determined entirely by the availability of compute (or the speed of discovering various efficiencies in algorithmic training or inference) for AI systems. Humans play a substantially diminished role in their development, likely moving most of our effort towards oversight, validation, and verification of an expanding “virtual lab” run by AI systems. We expect that systems capable of automated AI research and development would have skills that would transfer to the rest of science, allowing them to begin to revolutionize other fields.
   
   How the alignment problem gets solved—or not—in this future is something we are least certain about. Models could prove to be sufficiently aligned and capable enough of research taste that they discover and implement novel solutions that we have not yet reached. They could also be sufficiently wise to halt development if not. Alternatively, the rare occurrences of misalignment present in today’s models could compound as the models build their successors, growing more frequent but less understood until we lose control of them. It’s possible that we can’t build, integrate, and verify the tools that we’d need to understand which trendline we are actually on.
   
   We do not have good intuitions for what this world would look like, because our economy is currently driven by humans and human-built tools. By its nature, a world driven by fast recursive self-improvement could become dominated by the self-improving model as its capabilities fully eclipse those of humans and the model proliferates across the broader economy. It is difficult to predict what the economy looks like if human labor stops being competitive.
   
   Even if model development became fully automated and recursive, we can’t predict what that would mean for most humans’ daily lives. Amdahl’s law applies here as well. Recursive intelligence could lead to achieving many of the benefits outlined in Machines of Loving Grace, quickly in some domains. We expect that embodied intelligence (i.e., robotics) might quickly follow recursive intelligence, and follow a similar path of increasing returns at decreasing cost. More powerful intelligence might help us build things in the physical world more quickly, run more productive clinical trials of lifesaving drugs, and develop novel forms of coordination.
   
   But achieving recursive improvement alone does not suggest an immediate change in how industrial production occurs, societies organize, or markets function. More intelligence can’t learn what a drug does over decades of use, can’t hold elections sooner than a constitution dictates, and can’t turn a stranger into an old friend in a weekend. For most people, the felt pace of this future will still be set by the bottlenecks, even if the laboratory upstream runs at the speed of compute. That collision, where recursive intelligence building itself ever faster meets the world of humans, relationships, and governance, is another part of this future we can’t predict.

What should we do?我们应该做什么？

If it were possible to effectively slow the development of this technology to give ourselves more time to deal with its immense implications, we think that would likely be a good thing. But if a slowdown simply lets the least cautious actors catch up technologically, it could leave everyone less safe. Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures.如果能够有效减缓这项技术的发展，以便给我们更多时间来应对其巨大的影响，我们认为这可能是一件好事。但如果减速只是让最不谨慎的参与者在技术上赶上来，那可能会让所有人都变得不那么安全。在没有全球协调机制的情况下，公司和政府将不得不在竞争和地缘政治压力下做出关于安全的艰难决策。

We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology. The Anthropic Institute will conduct research—in collaboration with many others—and take actions to help build the systems that a credible slowdown or pause would require. These systems would enable frontier AI developers to verify that others globally have actually stopped or slowed, and that a bad actor could not use the auspices of a coordinated slowdown to jump ahead in secret. If such systems existed, we expect that we would slow down or temporarily pause, if other developers at or near the frontier also did so in a verifiable manner.我们认为，世界能够选择减缓或暂时暂停前沿 AI 开发，以使社会结构和对齐研究能够跟上技术进步的步伐，这对世界是有利的。Anthropic 研究所将与许多其他机构合作开展研究，并采取行动帮助构建可信的减速或暂停所需的系统。这些系统将使前沿 AI 开发人员能够验证全球其他人确实已经停止或减速，并且坏行为者无法利用协调减速的幌子在秘密中实现跨越式发展。如果存在这样的系统，我们预计如果其他处于或接近前沿的开发人员也以可验证的方式这样做，我们也会减速或暂时暂停。

A meaningful slowdown or pause would require multiple well-resourced labs at or near the frontier, in multiple countries, agreeing to stop under the same conditions. It would also require that each can verify that the others have actually stopped. Due to the unique characteristics of AI systems, the detectability (a lower standard than verifiability) element of this arms control problem is much more challenging than with other technologies. Training runs are far easier to conceal than missile silos, their inputs are general-purpose, and the incentive to defect quietly is enormous, because whoever continues while others pause could inherit the lead. A credible pause also has to specify what triggers it, what lifts it, and who adjudicates.有意义的减速或暂停将需要多个处于或接近前沿、资源充足的实验室在多个国家达成一致，在相同条件下停止开发。它还需要每个实验室都能验证其他实验室确实已经停止。由于 AI 系统的独特特征，这一军备控制问题的可检测性（比可验证性标准更低）要素比其他技术要困难得多。训练运行比导弹发射井更容易隐藏，它们的输入是通用的，而且悄悄背叛的动机巨大，因为任何在别人暂停时继续开发的人都可能继承领先地位。可信的暂停还必须明确触发条件、解除条件以及裁决者。

None of this is necessarily impossible in principle—the world has built verification regimes for other complex technologies (e.g., the Intermediate-Range Nuclear Forces Treaty)—but those regimes took decades to build both the infrastructure and the trust. We don’t have that long. A unilateral pause by one lab, by contrast, is achievable immediately, but accomplishes much less: it would change who the front-runner is, but it would not create the wider deliberative process that is currently missing.原则上，这些都不一定是不可能的——世界已经为其他复杂技术建立了验证制度（例如《中导条约》）——但这些制度花费了数十年时间来建立基础设施和信任。我们没有那么长的时间。相比之下，一个实验室的单方面暂停可以立即实现，但成就甚微：它会改变谁是领跑者，但它不会创建目前缺失的更广泛的审议过程。

In the coming months, we will organize conversations where policymakers, researchers, civil society, and other AI companies can help answer some of the questions this piece raises, especially around full recursive self-improvement and how to create better options for coordination and deliberation. We’ll publish what comes out of it. The window to investigate the questions together is here, and people outside AI companies should be involved in this deliberation.在接下来的几个月里，我们将组织对话，让政策制定者、研究人员、民间社会和其他 AI 公司帮助回答本文提出的一些问题，特别是关于完全递归自我改进的问题，以及如何为协调和审议创造更好的选择。我们将发布对话成果。共同调查这些问题的窗口期就在这里，AI 公司以外的人士应该参与到这项审议中来。

Marina Favaro and Jack Clark co-authored this piece, with editorial support from Santi Ruiz. Shan Carter, Romello Goodman, and Nikki Makagiansar created the visuals from data collected by Brian Calvert and Jun Shern Chan. Daniel Freeman, Jim Baker, Max Young, Sarah Pollack, Francesco Mosconi, Holden Karnofsky, Andy Jones, Kevin Troy, Chloe Lubinski, Anton Korinek, Meg Tong, Andrew Ho, Dan Altman, Drake Thomas, Jack Shen, Sasha de Marigny, and Avital Balwit provided feedback.Marina Favaro 和 Jack Clark 共同撰写了本文，Santi Ruiz 提供了编辑支持。Shan Carter、Romello Goodman 和 Nikki Makagiansar 根据 Brian Calvert 和 Jun Shern Chan 收集的数据制作了视觉效果。Daniel Freeman、Jim Baker、Max Young、Sarah Pollack、Francesco Mosconi、Holden Karnofsky、Andy Jones、Kevin Troy、Chloe Lubinski、Anton Korinek、Meg Tong、Andrew Ho、Dan Altman、Drake Thomas、Jack Shen、Sasha de Marigny 和 Avital Balwit 提供了反馈。

Update 9/18/20262026 年 9 月 18 日更新