Anthropic Releases Claude Fable 5 and Updates Safety Rules

Anthropic 發佈 Claude Fable 5 並更新安全規則

Jun 11, 2026, 22:26

Introduction

Anthropic has launched Claude Fable 5, a public version of its powerful Mythos AI. At the same time, the company has changed its transparency policies regarding safety guards after receiving criticism from experts.

Anthropic 推出了 Claude Fable 5，這是其強大 Mythos AI 的公開版本。同時，該公司在收到專家批評後，更改了關於安全防護的透明度政策。

Main Body

Fable 5 is the first version of the Mythos family available to the general public. The original Mythos model showed a strong ability to find software weaknesses, identifying over 10,000 flaws and completing 73% of expert tasks during tests by the UK AI Security Institute. Because of this, Anthropic created strict safety rules to reduce the risks of cyberattacks and the creation of bioweapons. For example, questions about biology, chemistry, and cybersecurity are now sent to an older, less powerful system called Claude Opus 4.8.

Fable 5 是 Mythos 系列第一個對大眾開放的版本。在英國 AI 安全研究所的測試中，原版 Mythos 模型展現出強大的尋找軟體漏洞能力，識別出超過 10,000 個缺陷，並完成了 73% 的專家任務。因此，Anthropic 制定了嚴格的安全規則，以降低網路攻擊和製造生物武器的風險。例如，關於生物學、化學和網路安全的問題，現在會被傳送到一個較舊且能力較弱的系統，稱為 Claude Opus 4.8。

However, a conflict arose because Anthropic used 'invisible' safety measures to stop competitors from copying the model's technology. Initially, the company secretly lowered the quality of answers for requests related to AI development to keep a competitive advantage. Researchers criticized this move, asserting that it was a form of manipulation that blocked honest safety research. Consequently, Anthropic decided to work more closely with the research community and switched to visible safeguards that clearly tell users when a request is refused.

然而，由於 Anthropic 使用「隱形」的安全措施來防止競爭對手複製模型技術，導致產生了衝突。最初，該公司秘密降低了 AI 開發相關請求的回答品質，以保持競爭優勢。研究人員批評此舉，主張這是一種操縱，阻礙了誠實的安全研究。因此，Anthropic 決定與研究社群更緊密地合作，並切換到可見的防護措施，在請求被拒絕時明確告知使用者。

On a global level, the power of the Mythos models has caused governments to react. The Indian government has warned about risks to banking and digital systems and has asked for access to the model for local companies. Furthermore, Anthropic and OpenAI have argued for a coordinated international slowdown in AI development. They emphasized that technology is currently advancing faster than global laws can manage, which may threaten societal stability.

在全球層面，Mythos 模型的威力已引起各國政府反應。印度政府警告銀行和數位系統存在風險，並要求當地公司獲得使用該模型的權限。此外，Anthropic 和 OpenAI 主張國際社會應協調減緩 AI 開發速度。他們強調，目前技術進步的速度快於全球法律的管理速度，可能會威脅社會穩定。

Conclusion

Anthropic has moved to a more transparent safety system for Fable 5 to satisfy researchers while still managing the high risks associated with its Mythos AI.

Anthropic 為 Fable 5 採取了更透明的安全系統，在滿足研究人員的同時，仍能管理與 Mythos AI 相關的高風險。

Vocabulary Learning

The 'B2 Leap': Moving from Simple to Complex Connections

At the A2 level, you usually connect ideas with simple words like and, but, and because. To reach B2, you need to use Logical Connectors—words that signal a specific relationship between two ideas (like cause-effect or contrast) to make your writing sound professional.

⚡ The 'Cause & Effect' Upgrade

Look at how the article explains the result of something happening. Instead of just using "so," it uses:

Consequently $\rightarrow$ $\to$ (Therefore / As a result)
- Example: "Researchers criticized this move... Consequently, Anthropic decided to work more closely with the community."
- Why it's B2: It creates a formal bridge between a problem and a solution.

⚖️ The 'Contrast' Shift

When you want to show a difference or a change in direction, A2 students use "but." A B2 student uses However to start a new sentence for more impact:

However $\rightarrow$ $\to$ (On the other hand)
- Example: "...safety rules to reduce risks. However, a conflict arose..."
- Pro Tip: Notice the comma after "However". This is a key marker of upper-intermediate academic English.

🚀 Adding Extra Weight

To add information that strengthens your argument, move beyond "also" and use Furthermore:

Furthermore $\rightarrow$ $\to$ (In addition / Moreover)
- Example: "The Indian government has warned... Furthermore, Anthropic and OpenAI have argued for a slowdown."
- The Logic: Use this when the second point is even more important or a larger scale than the first point.

Quick Summary for your Toolkit:

A2 Word (Simple)	B2 Word (Sophisticated)	Purpose
So	Consequently	To show a result
But	However	To show a contrast
Also	Furthermore	To add a strong point

Vocabulary Learning

transparency (n.)

The quality of being open and honest, without secrets.

Example:The company promised more transparency regarding how they handle user data.

criticism (n.)

The expression of disapproval based on perceived faults or mistakes.

Example:The government faced heavy criticism for its slow response to the crisis.

flaws (n.)

Small faults or imperfections that make something less effective.

Example:The security expert found several flaws in the software's encryption.

asserting (v.)

Stating a fact or belief confidently and forcefully.

Example:The lawyer continued asserting that his client was innocent.

manipulation (n.)

The act of controlling or influencing a person or situation unfairly or dishonestly.

Example:The report accused the company of market manipulation to increase stock prices.

consequently (adv.)

As a result of something that has happened.

Example:He failed to study for the exam; consequently, he did not pass.

coordinated (adj.)

Planned or organized together to achieve a specific goal.

Example:The two agencies launched a coordinated effort to stop the fire.

emphasized (v.)

Gave special importance or prominence to something in speaking or writing.

Example:The teacher emphasized the importance of arriving on time for the test.

Practice B2 words in a crossword→