If a Company Deploys AI Tools but Doesn't Provide Tokens, Does That Count as 'Failing to Provide Necessary Working Conditions'?

Has everyone started “raising shrimp” yet?

Since LLMs became widely known to ordinary people, we’ve never seen anything quite like this push—from individuals to companies to regulators—all enthusiastically promoting the same thing.

Many companies have started embracing “lobster” (DeepSeek, a homophone in Chinese) or other agents, connecting various processes and devices, and pushing employees to use “lobster” extensively.

Until the boss saw the “lobster” bill.

So I have a thought experiment worth exploring:

If a company provides a full suite of AI tools but doesn’t allocate token quotas, can an employee terminate the contract and claim compensation (severance pay “N”) under Article 38 of the Labor Contract Law (failure to provide working conditions)?

If an employee refuses to use AI at their own expense (and fails to meet the company’s efficiency standards), can the company fire them for “incomplete performance”?

• This article represents only the author’s personal views and should not be considered legal advice.

I. If You Want the Horse to Run, the Horse Must Buy Its Own Hay

Installing shrimp is easy; raising shrimp is hard.

And once you see the bill, playing blind is also hard.

Big companies—especially leading model companies—might give their employees tokens for their own models.

Slightly smaller companies might agree to reimburse employees for AI service purchases.

And of course, some companies have bosses who subtly hint at these AI tools, even having IT help everyone install them, but never mention the cost.

Paying to work—we all know the deal.

But not using them isn’t an option either.

Driven by AI marketing influencers, bosses have “clearly understood” AI’s power.

Things like “one person completed what a team of three would do in two days”

Or “one person set up a bunch of shrimp that replaced several departments”

Or “a planner built an AI workflow that eliminated all outsourcing” (don’t screenshot that!)

Without AI, there’s no way to meet the boss’s ever-increasing efficiency expectations.

The new corporate contradiction of our era has become:

The contradiction between the boss’s growing need for efficiency and the equipment’s unbalanced, insufficient token preparation.

II. If the Company Doesn’t Provide Tokens, Can the Employee Claim Compensation?

This is outrageous!

They install Claude Code and “lobster” for me, but won’t buy any plan, and tell me to “figure it out myself.”

I quit!

But to terminate the contract under Article 38 of the Labor Contract Law—“failure to provide labor protection or working conditions as stipulated in the labor contract”—and claim severance pay “N,” we need to clarify a legal concept first:

What exactly are legally defined “working conditions”?

In judicial practice, courts are very strict about what constitutes “failure to provide working conditions.”

For internet and service-sector companies, this typically means: the company doesn’t provide you with a computer, cuts off internet and power, doesn’t assign you an office desk, or—if you’re a driver—doesn’t even give you a car to drive, making it objectively impossible for you to perform your job.

So the question is: without AI tokens, can an employee still work?

Legal departments would probably argue in court like this:

“Although the plaintiff didn’t have AI API quotas, IT confirmed that their computer was working fine—Word could type, Excel could make tables, the company network was accessible, and the plaintiff even browsed video sites during lunch breaks. We merely encouraged AI use; we never said work couldn’t be done without AI! The plaintiff could complete their work using their own skills.”

“Unless the plaintiff’s abilities are no longer sufficient to meet the position’s requirements.”

Yes, as long as employees can still use traditional “primitive” methods (e.g., manual typing, hand-drawing, manual coding) to complete normal work tasks, courts will struggle to find that the company “failed to provide working conditions.”

Therefore, unless your job title is “AI Prompt Engineer,” “LLM Fine-tuning Specialist,” “Agent Specialist,” or some other AI-related position, and your labor contract explicitly states that the company must provide unlimited or sufficient quotas for a specific model, then relying solely on “the company won’t give me tokens” to claim constructive dismissal has a very high risk of losing—you likely won’t get “N” and might even lose your monthly unemployment benefits due to impulsive resignation.

III. Can an Employee Be Fired for Not Using AI (at Their Own Expense)?

“Improve efficiency and reduce costs”—if you don’t hustle, you’ll be covered.

Bosses, after reading marketing content, believe employee efficiency should increase tenfold with AI assistance.

Originally writing 1 press release per day, after a brainstorm, the KPI is now set at 10 per day.

Employees see this KPI—impossible without AI.

Then they look at the bill—just saying “hello” to “lobster” costs the price of a budget meal.

They grit their teeth thinking about buying a Code Plan, but then reconsider—wait, why should I pay to work?

So they choose to slack off: if the company won’t pay, I’ll do it manually.

The result is obvious: the employee only produces 2 pieces per day and fails the monthly assessment.

Can the company fire the employee for “incompetence”?

The company would likely have to pay compensation (2N).

Although Article 40 of the Labor Contract Law states that if an employee is incompetent and remains so after training or job adjustment, the employer can terminate the contract with 30 days’ notice or one month’s extra pay, there’s a crucial hidden prerequisite in practice:

The company’s performance standards must be “reasonable.”

What is reasonable?

It means the standard that most employees in the same position, within a normal 8-hour workday, using the tools and conditions provided by the company, can reach with some stretch (a reachable KR).

If the company neither reimburses tokens nor provides Code Plan accounts, but evaluates output by the standard of “fully AI-armed premium workhorses,” this typically constitutes “setting objectively unattainable unreasonable KPIs” in legal terms.

Think about it: if your boss sends you to Shanghai on a business trip but won’t buy you a train or plane ticket—expecting you to walk there—can they fire you for “low efficiency” when you don’t arrive on time?

So at the arbitration hearing, as long as the employee can prove:

Their output meets the industry’s normal manual operation standards;

The company’s ultra-high KPIs are absolutely impossible to achieve without paid AI tools (but are feasible with them);

The company refuses to provide or reimburse for those tools.

Then, the company firing the employee for “performance failure” is easily deemed illegal termination. The employee can then rightfully demand 2N (illegal termination compensation).

IV. If an Employee Self-Funds Token Purchases, Can the Company Fire Them for Leaking Trade Secrets?

Some companies neither pay nor mandate AI use, but clever employees might think:

“To clock out earlier or outperform my teammates, why not buy tokens myself, connect to a commercial API, and secretly use AI to help me work?”

Great idea!

Few ideas manage to simultaneously involve paying to work and dangerously toeing legal lines.

If you do this, the company can not only fire you without compensation (no “N”), but might even countersue you for damages.

Why?

When you use an external commercial LLM API (even domestically compliant ones), your prompts and uploaded files inevitably leave the company’s internal network and travel to third-party model providers’ servers.

If you’re an admin just asking AI to polish or generate a Mid-Autumn Festival holiday notice, no problem.

But—

If you’re a programmer feeding core company code to AI to find bugs;

If you’re an HR feeding the entire employee salary sheet to AI for data analysis;

If you’re a salesperson feeding core client lists and quotes to AI for follow-up emails…

Congratulations—legally speaking, you’ve completed the act of “unauthorized transmission of company confidential information to a third party.”

In labor practice, when companies discover this, here’s what usually happens:

Step 1: The company pulls out the employee handbook or confidentiality agreement, pointing to the clause: “It is strictly prohibited to upload internal company documents, data, or code to unauthorized external networks or third-party software.”

Step 2: The company classifies your conduct as “serious violation of the employer’s rules and regulations” (Article 39, Item 2 of the Labor Contract Law) or “serious dereliction of duty or engaging in malpractice for personal gain, causing substantial damage to the employer” (Article 39, Item 3).

Step 3: Legal termination of the labor contract, zero compensation, out the door that same day.

In the AI wave, bosses already want to “optimize” employees but lack excuses. Employees buying their own APIs are essentially handing the boss a “royal sword” for legal layoffs.

If the uploaded data truly constitutes legally defined “trade secrets,” the company could even file a criminal complaint with police, charging the employee with trade secret infringement.

Conclusion

Self-funding AI for efficiency is admirable in spirit but extremely risky.

If the company hasn’t purchased an enterprise AI plan (enterprise plans generally include agreements that data won’t be used for training or retained) and hasn’t explicitly authorized you to input company business data into external AI, never feed real core company data to LLMs!

If you absolutely must, make sure to “data-mask” (replace all key names, amounts, core parameters, and project codes with placeholders like Zhang San, Li Si, Product A, Company B). Take an extra five minutes to fill in manually or use local models—never gamble your career by throwing raw data at AI.

V. Final Thoughts

Technology keeps advancing, but the fundamental logic of labor-management dynamics never changes.

“Lobster” or any other magical creature of the future—AI is ultimately just a productivity tool.

Tools can be upgraded, but labor law’s baseline cannot be lowered; workers’ risk awareness must be upgraded along with the tools.

Bosses who want to enjoy AI’s cost reduction and efficiency gains must bear the infrastructure costs and corresponding compliance management costs. Trying to shift the cost of production tools onto workers through a “pay to work” model is not only morally “getting something for nothing” but also legally untenable.

As workers, we should embrace AI—learning to use various AI tools can indeed help us clock out earlier. But while protecting our wallets, we must also protect company trade secrets—and that means protecting our own freedom.

When the boss talks big about having you churn out AI content while staying silent on reimbursement, just smile and ask:

“Boss, I’ve checked with my lawyer—don’t we need to buy an enterprise AI plan? And maybe do an AI compliance review?”