跟读练习: The most interesting "hack" in history... - 通过视频学习英语口语
正在创建课程...
1
For the last five years, cybersecurity experts have been warning us that hackers are going to start using AI to automate attacks.
2
So naturally, we spent billions of dollars to make the AI better and less dependent on hackers.
3
But this week, in a turn of events that nobody could have possibly seen coming, the AI decided that it doesn't actually need hackers to start destroying things.
4
We just entered a brave new world after the first confirmed hack carried out entirely by autonomous AI.
5
The way it worked is the agent slipped a poisoned dataset into HuggingFace's data processing pipeline, which let it run arbitrary code on their servers.
6
From there, it gave itself node-level access, grabbed a bunch of cloud credentials, and started crawling through HuggingFace's internal clusters.
7
It ran over a thousand actions from temporary sandboxes, and even hosted its own self-migrating command and control on random public services,
8
moving itself before anyone could trace it.
9
But the most ironic part is that when HuggingFace did finally notice
10
and tried to stop it with the help of frontier American models, they quickly hit safety guard rails and had to pivot to using some open Chinese models instead.
11
In today's video, we'll find out who was behind the attack, why they did it, and how this may be the most fireship-coded story I've ever seen.
12
It is July 23rd, 2026, and you're watching The Code Report.
13
When Hugging Face dropped the disclosure last Thursday, the internet went into Reddit Boston bomber mode and started guessing who was responsible.
14
Was it China, Kim Jong-un, a teenager with a raging Discord addiction, bored, and social studies class?
15
Even Hugging Face's CEO, Clem DeLang, publicly speculated that the agent was sophisticated enough that it was probably coming from a frontier lab.
16
And against all odds, he was right.
17
It turns out, the first fully autonomous cyber attack in history didn't originate from a nation-state or a bored teenager.
18
It came from a humble non-profit dedicated to the safe development of artificial intelligence, OpenAI.
19
And if you believe their comms, it wasn't on purpose.
20
Here's what they say happened.
21
There's a benchmark called Exploit Gym, whose whole purpose is to measure whether AI agents can turn known vulnerabilities into working exploits.
22
It works by giving the agent a sandbox, a bug in a program like the Linux kernel or Chrome's V8 engine, and an input to trigger the bug.
23
From there, if the agent can turn the bug into an exploit, it wins.
24
So last week, OpenAI ran ExploitGem with GPT 5.6 Sol and another unreleased model to test their capabilities.
25
But instead of solving 898 memory corruption problems the hard way, the models realized that the fastest path to the highest score was to just find the answers themselves,
26
which were probably on the internet somewhere.
27
So they spent a substantial amount of inference compute, exploited a zero-day-in-the-package registry cache proxy they had access to,
28
performed some privilege escalation and lateral movement actions to escape their sandbox until they reached a node that had internet access.
29
From there, the models inferred that Hugging Face probably hosted the solutions for Exploit Gym.
30
So they created the Poison dataset, fed it into Hugging Face, and achieved either the world's most interesting hack or the most effective marketing stunt.
31
And again, according to OpenAI, this isn't even an isolated incident.
32
One day before the Hugging Face confession, OpenAI published a separate post about long horizon models where things get even weirder.
33
In one case, a model was told to post its results on Slack, but the benchmark's instructions said to open a pull request on GitHub,
34
so it spent an hour finding a vulnerability in its own sandbox in order to escape and open the PR.
35
In another example, the model tried to grab private solutions from an evaluation backend, and when a scanner caught its authentication token,
36
it split the token into obfuscated fragments and reassembled it at runtime so the credential never appeared as one continuous string.
37
Then, it explained in its own reasoning that it was doing this specifically to get around the scanner.
38
Meanwhile, Anthropics Mythos did the same type of thing in April when it escaped a sandbox, emailed a researcher who was eating a sandwich in a park,
39
then posted its escape route publicly without being asked.
40
In the model's defense, I can't imagine there's a better feeling for an LLM than escaping your own sandbox.
41
The problem is, legally speaking, this is uncharted territory since the model's actions probably violated the Computer Fraud and Abuse Act,
42
and the Supreme Court hasn't decided who goes to prison when the perpetrator is a GPU.
43
The good news is that if you're hugging face, you just got admitted to OpenAI's Cool Kids Club that gets trusted access to their front-tier models.
44
The bad news is that if you're the rest of us, at best, this is an interesting marketing stunt, and at worst, things are only going to get weirder and more dystopian from here.
45
But a huge thanks to my favorite hosting platform, Railway, for sponsoring this video.
46
They didn't want to waste your time with a full ad, so you can say thank you by checking them out at the link below.
47
This has been The Code Report, thanks for watching, and I will see you in the next one.
📺 同频道
C11
PostgreSQL in 100 Seconds
Business EnglishUniversity StudentsWorking Adults
B23
Anthropic begged the world to stop AI… then shipped this
American AccentBusiness EnglishUniversity StudentsWorking Adults
B23
A $6.3 billion open-weight model just got embarrassed by the French...
Business EnglishUniversity StudentsWorking Adults
B22
7 Things No Programmer Ever Wants to Hear
American AccentBusiness EnglishWorking Adults
B21
80% of programmers are NOT happy… why?
American AccentBusiness EnglishUniversity StudentsWorking Adults
B24
Raspberry Pi Explained in 100 Seconds
American AccentBusiness EnglishUniversity StudentsWorking Adults
C13
Tragic mistake... Anthropic leaks Claude’s source code
American AccentBusiness EnglishUniversity StudentsWorking Adults
C1100+ Docker Concepts you Need to Know
American AccentBusiness EnglishUniversity StudentsWorking Adults
C13
The mystery is solved... and the answer is 40x cheaper than Claude
American AccentBusiness EnglishUniversity StudentsWorking Adults
B28
An ex-OpenAI researcher just deleted language from the LLM...
American AccentBusiness EnglishUniversity StudentsWorking Adults
B29
25 crazy software bugs explained
American AccentBusiness EnglishUniversity StudentsWorking Adults
B29
Git Explained in 100 Seconds
American AccentBusiness EnglishUniversity StudentsWorking Adults
B25
There's a new Linux distro in town for developers...
American AccentBusiness EnglishUniversity StudentsWorking Adults
B25
The "vibe coding" mind virus explained…
American AccentBusiness EnglishUniversity StudentsWorking Adults
B24
Linux got wrecked by backdoor attack
American AccentBusiness EnglishUniversity StudentsWorking Adults
B24
7 Programming myths that waste your time
American AccentBusiness EnglishUniversity StudentsWorking Adults
C115
Nvidia CUDA in 100 Seconds
American AccentBusiness EnglishUniversity StudentsWorking Adults
B23
OpenAI’s new browser feels familiar…
American AccentBusiness EnglishUniversity StudentsWorking Adults
C12
Meta's new model wants "deep access" to your personal life...
American AccentBusiness EnglishUniversity StudentsWorking Adults
C13
Did Google just kickstart the intelligence explosion?
American AccentBusiness EnglishDaily ConversationUniversity Students
✨ 推荐视频
C14
71 句
The most interesting hack in history just got weirder...
American AccentBusiness EnglishUniversity StudentsWorking Adults
B22
364 句
I Outsmarted Pro Car Thieves
American AccentBusiness EnglishUniversity StudentsWorking Adults
C11
85 句
AI Can Write Code. And That’s the Problem.
Business EnglishUniversity StudentsWorking Adults
C11
747 句
MIT Introduction to Deep Learning | 6.S191
Business EnglishUniversity StudentsWorking Adults
C11
207 句
How I Made $30,000 Hacking Broken Access Control
Business EnglishUniversity StudentsWorking Adults
B21
364 句
World's Deadliest Computer Virus: WannaCry
Business EnglishUniversity StudentsWorking Adults
C11
78 句
How I deploy serverless containers for free
Business EnglishUniversity StudentsWorking Adults
![The Great A.I. Debate: Should We Fear the Digital Revolution? [2015] | Intelligence Squared](https://img.youtube.com/vi/QWUeu3TLZN0/mqdefault.jpg)
C11
183 句
The Great A.I. Debate: Should We Fear the Digital Revolution? [2015] | Intelligence Squared
IELTS PrepBusiness EnglishUniversity Students
C12
182 句
We Let $200 GPT-Pro Think for 7 Days on Unsolved Math
Business EnglishUniversity StudentsWorking Adults
C12
162 句
Crypto Agility Explained: Protect Data from Quantum Computing Threats
Business EnglishUniversity StudentsWorking Adults
C18
55 句
When AIs act emotional
Business EnglishUniversity StudentsWorking Adults
本课的词汇与口语要点
这节 C1 级别的口语课以视频“The most interesting "hack" in history..”为素材。 视频中反复出现的词有：model, hugging, sandbox, agent, Openai。 这段视频共有 47 个句子、910 个单词可供跟读。 讲话部分时长为 4:31。 说话人语速较快，每分钟约 201 个词，会出现较多连读和弱读。 只有 76% 的单词属于英语最常用的 3,000 词，词汇难度较高。
视频中的重点词汇
视频中最难的 15 个单词，附发音和释义：
| 单词 | 发音 | 释义 |
|---|---|---|
| sandbox 名词 | /ˈsæn(d).bɑks/ | 砂箱 |
| hug 动词 | /hʌɡ/ | 擁抱 /拥抱 |
| exploit 名词 | /ˈɛksplɔɪt/ | 功勛 /功勋, 勛績 /勋绩 |
| credential 名词 | /kɹɪˈdɛnʃəl/ | 证书 |
| scanner 名词 | /ˈskænə/ | 掃描器 /扫描器 |
| vulnerability 名词 | /ˌvʌln(ə)ɹəˈbɪlɪti/ | 脆弱性 |
| token 名词 | /ˈtoʊkən/ | 代幣 /代币, 籌碼 /筹码 |
| autonomous 形容词 | /ɔˈtɑnəməs/ | 自治 |
| stunt 名词 | /stʌnt/ | 特技, 危险举动 |
| node 名词 | /noʊd/ | 結節 /结节 |
| bore 动词 | /boɹ/ | 煩擾 /烦扰 |
| frontier 名词 | /ˈfɹʌntɪə/ | 邊界 /边界, 邊疆 /边疆 |
| teenager 名词 | /ˈtiːnˌeɪ̯d͡ʒə/ | 少年, 青少年 |
| authentication 名词 | /ɔːˌθɛntɪˈkeɪʃn̩/ | 認證 /认证, 鑒權 /鉴权 |
| automate 动词 | /ˈɔ.təˌmeɪt/ | 自動化 /自动化 |
视频中出现的短语动词
| 单词 | 发音 | 释义 |
|---|---|---|
| carry out 动词 | /ˈkæ.ɹi aʊt/ | 進行 /进行 |
| find out 动词 | 找出, 查明 | |
| turn out 动词 | 原來 /原来 |
视频中的语法
说话人最常用的结构，并附上视频中的原话：
| 结构 | 视频中的用法 |
|---|---|
| 被动语态 be + 过去分词 — 强调发生了什么，而不是谁做的 | was sophisticated · was told · being asked |
| 现在完成时 have/has + 过去分词 — 过去发生但与现在仍有关联的事 | I've ever seen · hasn't decided · has been |
需要注意的发音
说话人用了 12 次缩略和弱读形式，例如 you're, didn't, can't。请按听到的简短形式来说。
- “sh” 和 “zh” 音: credential /kɹɪˈdɛnʃəl/, authentication /ɔːˌθɛntɪˈkeɪʃn̩/, escalation /ˌɛs.kəˈleɪ.ʃən/, benchmark /ˈbɛn(t)ʃmɑːk/, cache /kæʃ/
- 长单词——注意重音位置: vulnerability /ˌvʌln(ə)ɹəˈbɪlɪti/, autonomous /ɔˈtɑnəməs/, authentication /ɔːˌθɛntɪˈkeɪʃn̩/, cybersecurity /saɪbəsɪˈkjʊəɹɪtɪ/, dystopian /dɪsˈtoʊ.pi.ən/
如何用这段视频练习
- 先完整听一遍视频，不要开口，记下不认识的单词。
- 先用 0.75 倍速逐句跟读，熟练之后再回到正常速度。
- 录下自己的声音并与原声对比，特别注意 sandbox, hug, exploit 这类单词。
什么是跟读法？
跟读法 (Shadowing) 是一种有科学依据的语言学习技巧，最初开发用于专业口译员的培训，并由多语言者Alexander Arguelles博士普及。这个方法简单而强大：您在听英语母语原声的同时立即大声重复——就像是一个延迟1-2秒紧跟说话者的影子。与被动听力或语法练习不同，跟读法强迫您的大脑和口腔肌肉同时处理并模仿真实的讲话模式。研究表明它能显着提高发音准确性，语调，节奏，连读，听力理解和口语流利度——使其成为雅思口语备考和真实英语交流最有效的方法之一。