比特币橙子Trader|11月 20, 2026 00:28
This might be the biggest hidden danger of the AI bubble!
A post on V2EX is quite interesting:
A company experienced online malfunctions and accidental deletion of user data due to AI-generated code. The boss immediately ordered a complete ban on AI coding, demanding programmers return to hand-writing code.
In 2026, the AI industry is facing a dangerous misalignment:
The rate of productivity growth has outpaced the rate of organizational management, testing, and security verification.
In a 2026 survey of 1,528 developers and technology decision-makers by GitLab, 85% believed that AI had shifted the bottleneck of software development from writing code to reviewing and verification, and 92% encountered governance issues with AI-generated code.
Veracode's 2026 security testing also found that approximately 44% of AI code generation tasks resulted in detectable OWASP Top 10 security vulnerabilities.
Meanwhile, hackers are also enjoying the productivity brought by AI.
Google's threat intelligence team disclosed that in the second quarter of 2026, they observed attackers, after compromising cloud resources, using AI agents to plan, build, and execute large-scale credential theft operations in less than six hours.
AI is simultaneously improving the efficiency of both attackers and defenders, but enterprise defense systems may not be keeping pace.
The most likely next step for enterprises is not to completely abandon AI, but rather to revoke its permissions:
Agents will not be allowed unrestricted access to production databases, customer data, and core business systems. Code must undergo automated testing, security scanning, and human approval.
Individuals will also reduce the permissions they grant AI with email, files, and accounts.
AI can still assist in tasks, but its ability to directly execute high-risk tasks will be limited.
For model companies, the danger lies in the fact that users may still exist, but the token consumption per user, API calls, and agent workload may all decrease.
This precisely hits the most sensitive business logic in the AI industry.
An economic study by Federal Reserve staff points out that the long-term profitability of AI model companies depends on the breadth and intensity of user adoption. The recouping of data center investments also depends on the continued penetration of AI throughout the economy and the revenue growth of model companies.
If enterprises begin to restrict high-consumption agent workflows, the revenue expectations, computing power utilization, and financing capabilities of model companies could all be affected.
Hundreds of billions of dollars in computing power investment cannot be supported by just occasional chats on ChatGPT.
I increasingly feel that the biggest risk of the AI bubble may not be that technological development falls short of expectations, but rather that technology is developing too rapidly, while enterprises are hesitant to truly integrate it into their core businesses.
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink