Core Summary: Why “Deletion” Is a Lie?
The central argument of this article is quite profound and even counterintuitive: in an open online and digital world, once information is released, it is permanent. The so-called “deletion,” “withdrawal,” or “forgetting” is almost physically impossible.
The author illustrates this point through three seemingly unrelated fields: military secrecy, brain-computer interfaces, and large AI models, revealing a common underlying principle: information is irreversible.
- Military Secrecy: Security measures do not focus on “erasing” leaked documents because, once they are disclosed, the enemy has already seen them; you can only arrest the person, not make them “forget.”
- Brain Science: Once your brainwave data is uploaded, even if the company deletes the database, your “cognitive fingerprint” has been recorded and cannot be truly erased.
- AI Models: The so-called “machine forgetting” by large models is merely an act; with a slight adjustment, the data can be recovered.
In one sentence: Don’t rely on “withdrawals” or “deletions” to protect you. True security lies in restraint before you send the information, not in remedies after it’s sent.
---
In-Depth Analysis: Understanding the Irreversibility of Information from Five Perspectives
To help you grasp this complex concept more easily, we break down the article into five specific scenarios and logical layers.
1. The Truth About Military Secrecy: Not About Preventing Theft, but About Preventing Access
Many think military secrecy is about stopping documents from being stolen. However, the author emphasizes that the first lesson in secrecy is to accept that information cannot be deleted.
- Logical Breakdown:
- Access Control: Why are top-secret documents assigned to specific individuals? Because once someone reads them, the information is stored in their memory, which cannot be erased.
- Physical Isolation: Why are secure computers not connected to the internet? Because once connected, the information enters the “black box” of the internet, making it impossible to track copies or their locations.
- Core Conclusion: All military secrecy measures (such as decommissioning periods and export approvals) are designed to prevent information from being released in the first place, as it becomes absolutely irreversible once it’s out.
- Plain Language: It’s like posting a photo on social media. You can delete it, but if someone takes a screenshot, shares it, or saves it to the cloud, you can’t control that. Military secrecy aims to prevent the release of information from the start.
2. The Disaster Caused by “Face Saving”: You Can’t Control Secondary Dissemination
The article mentions a real case where a temporary worker at a military company, Zhang, posted photos of new equipment to a family group to save face, resulting in a five-year sentence.
- Logical Breakdown:
- Misconception: People think family groups are secure, but in reality, military secrecy considers every copy as potentially dangerous.
- Chain Reaction: Once Zhang posted the photos, relatives shared them, and so on, spreading the information across the internet.
- Core Conclusion: You can’t control how members of a group share or distribute the information; you can’t locate all copies.
- Plain Language: You think you’re only sharing with your family, but they might share it further, creating an uncontrollable spread. Military secrecy is meant to prevent well-intentioned but careless individuals.
3. Legal “Innocence” vs. Physical “Leakage”: The Hegseth Case
US Secretary of Defense Hegseth leaked military strike plans via the encrypted messaging app Signal. The report concluded he didn’t “leak” the information legally, but he acknowledged a risk.
- Logical Breakdown:
- Legal Perspective: He might not be held responsible legally because no direct damage was caused or certain exemptions applied.
- Physical Perspective: The information was still transmitted. Although Signal uses end-to-end encryption, copies could exist on servers or by hackers.
- Core Conclusion: Legal innocence doesn’t mean the information didn’t get out. The law may exempt you, but the damage has already been done.
- Plain Language: It’s like posting an offensive message online and then deleting it. The law might say you’re not guilty, but the message has been shared and recorded. The risk and harm have occurred; the law just didn’t punish you.
4. Why Recover the Drowned F-35C: Because “Not Knowing” Is Valuable
In 2022, a US F-35C crashed in the South China Sea. Why was the wreckage recovered from 3,780 meters deep?
- Logical Breakdown:
- Conventional Thinking: The plane was destroyed, so why recover it?
- Military Thinking: The wreckage contains valuable materials and technical details that could be used by enemies.
- Core Conclusion: The value of military technology lies in its secrecy; exposure destroys that value.
- Plain Language: Your company’s secrets are like a recipe; even if the factory burns down, the recipe remains in the ashes. If competitors retrieve the ashes, they can reproduce the formula. The Navy’s effort was to prevent the enemy from accessing this information.
5. The False Deletion of AI and Brain Data: You and the Models Are Both “Forgetting”
The article then discusses brain-computer interfaces and AI models:
- Brain Data (Cognitive Fingerprint): A senator sued a company for its brainwave-tracking device, requesting the data to be deleted. The court complied, but the data was actually irretrievable.
- Truth: Neural data is unique; it includes attention, emotions, stress responses, and unconscious intentions. Once collected, it forms a “cognitive fingerprint.”
- Core Conclusion: You can delete database records, but you can’t erase the processed signals. Even after anonymization, the data can be re-identified.
- AI Models (Machine Forgetting): Companies claim they can “forget” training data, but a 2026 ICML paper showed this is merely an illusion. Small adjustments can restore the original information.
- Core Conclusion: Irreversible and non-catastrophic “forgetting” is extremely difficult in AI. The data is embedded in the model’s structure, making it impossible to completely erase.
- Plain Language: Your brain’s activities (emotions, intentions) are more private than written text. Companies claim to delete data, but it’s already scanned and archived. AI models may “forget” your messages, but they’re just stored in its memory, accessible with a slight reminder.
---
Lessons for Ordinary People: Think Before You Click “Send”
Although this article focuses on military, AI, and brain science, it has significant implications for us:
1. Give Up the Illusion of Withdrawal: On platforms like WeChat, Weibo, or email, once information is sent, it’s permanent. Don’t rely on “withdrawals” to protect your privacy or security.
2. Be Cautionous with Sharing: Many leaks happen due to casual sharing, not hacking. Be cautious, especially with work-related, private, or sensitive information.
3. Understand the Value of Confidentiality: Whether it’s business secrets, personal data, or technical details, their value lies in their secrecy. Exposure gives your opponent the upper hand.
4. Reevaluate Data Rights: When companies claim they can delete your data, realize this is often a legal promise, not a physical reality.
In conclusion, the author’s words are worth remembering:
> Before you press the send button, open an API, or upload training data, have you considered that the information you’re releasing will never be completely gone in an open environment?
This is not alarmist rhetoric; it’s a fundamental law of the digital age.