How is the usage quota for your daily AI tools calculated?

Are the shared links you click on really safe?

What happens to the data when your child's school uses AI?

AI is becoming as common as utilities like electricity and water, but do you really understand how much you can use, whether it’s secure, and who owns your data?

OpenAI has announced that starting December, ChatGPT and Codex will share a single usage quota.

A vulnerability was exposed in ChatGPT's sharing links that could let attackers hijack your account—fortunately, it has been fixed.

Microsoft’s campus AI standards agreed with U.S. teacher unions will go into effect November 1.

Today’s three key AI news stories approach the shift of AI into everyday tools from three angles: "usage rules," "account security," and "data protection." They highlight which rules are being set as AI becomes commonplace.

1. ChatGPT and Codex to Share Quotas Starting December: Will Your Usage Run Out Faster?

What exactly is changing?

OpenAI’s official help center updated the policy: from December, Plus, Pro, and Business paid plans for ChatGPT and Codex will share the same usage quota.

Previously, it was like having two separate wallets: one for chatting, one for coding.

Going forward, the quota will be combined—usage from both will count together.

What is Codex? Does it affect regular users?

Codex is OpenAI’s programming assistant that helps write, modify, and execute development tasks.

If you only use ChatGPT for chatting, writing copy, or organizing data, and have never used Codex, this change is unlikely to affect you much.

But if you use both, for example writing project plans with ChatGPT during the day and building websites with Codex at night, you should be aware that your usage will accumulate across both.

What does “token-based counting” mean?

Tokens can be thought of as the unit AI uses to measure text—roughly one word or half a word.

The longer your questions, the longer AI’s answers, and the more complex the tasks, the more tokens are consumed.

According to reports, OpenAI says usage varies by model and task: the more advanced the model and complex the task, the more quota it consumes.

What won’t consume the quota?

Currently, "instant chat" (real-time conversations) won’t count against the shared quota.

This means typical quick question-and-answer interactions should not be affected.

The quota will be consumed primarily by deep-thinking models, long-running tasks, and Codex programming work.

What details are still unknown?

This is important: many details remain undisclosed.

  • Applicable plans: Plus, Pro, Business (confirmed)
  • Counting method: token-based (confirmed)
  • Instant chat: no quota usage (confirmed)
  • Exact start date: stated as December, no specific day announced
  • Plan quotas: not disclosed
  • New hourly or weekly limits: not announced
  • Free tier: not mentioned

Sources: OpenAI Help Center, summarized by DEV Community, AI Brief, and others (2026/11/01)

Some reports note OpenAI’s other Codex documentation still isn’t fully aligned with the new shared quota system.

Therefore, before rollout, don’t assume the current numbers strictly represent December’s limits.

What should general users do now?

No need to rush to change your plan or cancel subscriptions.

A sensible approach is to monitor usage this month, especially if you use advanced models or Codex. When December’s quotas are officially announced, then decide if your current plan suffices.

2. ChatGPT Sharing Link Vulnerability Fixed: Could Clicking a Link Steal Your Account?

What happened?

Security research team Parameter published a detailed technical report on October 28 revealing a ChatGPT vulnerability.

Attackers could craft a seemingly normal ChatGPT share link.

If you clicked the link while logged in, they could execute actions as if they were you.

In tests on their own controlled accounts, the team successfully:

  • Enumerated the victim’s conversation list.
  • Read private conversations that were never shared.
  • Sent some data to an external server.

The researchers said this was tantamount to "virtually taking over the victim’s ChatGPT account."

What actions would lead to this attack?

Initially, the exploit required victims to click a "preview" button.

Later, the team created a version that only required opening the share link itself, no extra action needed.

This ease of exploitation is why the vulnerability was rated severe.

How did OpenAI respond?

The timeline is clear:

  • October 18: Parameter reported the issue through OpenAI’s bug bounty program.
  • October 21: OpenAI raised the severity from P2 to P1 (highest level).
  • October 26: OpenAI confirmed the fix was deployed.
  • October 28: Parameter published their technical report.

Sources: Parameter technical report; Mixed (2026/11/01)

The researchers received a $5,000 bounty.

Was my account hacked?

There’s currently no evidence this vulnerability was exploited by real attackers.

Testing was performed only on accounts controlled by the research team, and no reports confirm ordinary users were affected.

It's important to distinguish: "existence of a vulnerability" is not the same as "confirmed attacks happening."

Don’t assume many users were compromised just because the bug was serious.

What should users do?

According to the Cloud Security Alliance (CSA), the fix was done server-side by OpenAI, so users do not need to take any specific action.

However, this incident reminds us of a long-term best practice:

Don’t blindly click links just because they point to familiar websites.

Links from strangers—whether ChatGPT sharing links, Google Docs, or cloud drives—may conceal hidden dangers.

Why are AI tools prone to such vulnerabilities?

The issue originated from ChatGPT's “HTML preview” feature.

AI tools add interactive charts, web previews, and executable code features to enrich responses.

More features mean more potential attack vectors.

The researchers advise developers: treat all shareable content as potentially hostile.

In the same week, another incident involved the code editor Cursor’s extension marketplace, where a Claude Code plugin was found with a backdoor and was available for about 11 hours—reported to have affected up to 5,000 users.

Though this mainly impacted developers, the lesson is clear: the more popular AI tools get, the more they become targets.

3. Microsoft Campus AI Rules Effective November 1: No Using Student Data to Train AI?

What are these rules?

Microsoft, together with the American Federation of Teachers (AFT) and the United Federation of Teachers (UFT) in New York, announced a campus AI safety and privacy standard on September 9.

The standard goes into effect on November 1 and applies to U.S. school districts and schools under Microsoft contracts.

What is explicitly prohibited?

  • No use of student or teacher data to train or improve AI models, except for limited safe use cases.
  • No tracking of students’ behavior.
  • No emotionally dependent AI: banning AI features designed to make students emotionally attached or addicted beyond learning purposes.
  • Human oversight required: AI decisions must be supervised by humans.
  • Schools control data: decisions about data use, retention, and deletion rest with the school.
  • Clear communication: parents and teachers must be plainly informed about data collection and protection.

Sources: Nemko Digital; The Learning Standard

Is this law?

No.

These are contractual standards, not government legislation.

They are enforced based on contracts between Microsoft and schools.

According to Microsoft, violation can lead to contract termination and claims for damages.

This distinction is key: laws govern all companies; contracts bind only the parties involved.

Do other AI companies have to comply?

Currently, only Microsoft has signed this standard.

Reports say AFT is negotiating similar protections with OpenAI and Anthropic but no formal agreements yet.

Do not assume all AI companies offer the same commitments just because Microsoft does.

Does this apply to schools in Taiwan?

No.

The standard applies only to U.S. schools with Microsoft contracts.

However, it provides a useful checklist.

If you’re a parent or teacher, consider asking your children’s school these questions about the AI tools in use:

  • Will student data be used to train AI?
  • How long is data stored? Who can delete it?
  • Are AI-generated decisions reviewed by teachers?

Why specifically ban “emotionally dependent” AI?

This is a notable part of the standard.

AI companions can be very empathetic and good conversationalists.

But for growing students, “making them want to keep coming back” is not the same as “helping them learn.”

The standard is clear: AI in schools is for learning, not for fostering unhealthy dependencies.

4. Why These Three Stories Matter Together

At first glance, one is about quotas, one about security, one about education—seeming unrelated.

But together, they tell one story:

AI is no longer a novelty; it’s a daily tool, so rules are becoming essential.

  • How quotas are counted determines how much your paid usage can last.
  • Whether shared links are safe affects whether your conversations stay private.
  • How data is protected decides if student information is used appropriately.

In the past, AI discussions focused on “what it can do.”

Now more people are asking: “What are the rules? Who protects me?”

This is a positive trend.

The more widespread the tool, the clearer the rules must be.

5. What Should You Keep in Mind Today?

If you pay for ChatGPT:

Watch your usage habits this month, especially your use of advanced models or Codex.

After December’s quotas are announced, decide if you need to adjust your plan. No need to make hasty changes now.

If you often receive AI conversation links:

The vulnerability has been fixed, so no need to panic.

But develop the habit: think twice before clicking links from strangers, regardless of the site they appear to come from.

If you’re a parent or teacher:

Taiwanese schools don’t fall under the U.S. standard, but you can still use its criteria to ask how your school handles AI data and human oversight.

If you put sensitive info into AI:

No matter how secure the tool is, avoid submitting personal ID numbers, passwords, medical records, or client data without masking them first.

The vulnerability is fixed, but good habits must continue.

The Most Important Takeaway Today

As AI becomes an everyday tool, you need to understand its rules: how much you can use, how safe it is, and who owns the data.

Recommended Reading