XAI tax training program faces delayed payments

Elon Musk’s AI startup xAI asked employees in March to share tax returns and supporting documents to train Grok on tax tasks before the April 15 US filing deadline.

Internal chats show xAI offered US$420 and early access to X Money to participants.

The company later expanded the request to family members and friends who had used accountants.

However, some participants still had not been paid two months later.

The issue comes amid layoffs and management changes at xAI.

This follows a separate project last year called Macrohard that also offered bonuses for employee training data and paid after delays.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

xAI’s data push clashes with its public privacy rules and court claims

xAI asked employees, then family and friends, to share tax returns and related files so Grok could learn tax tasks before the April 15 US filing deadline. That sits awkwardly with xAI’s privacy policy, which tells users not to put personal information into prompts sent to its services 1.
The effort also jars with xAI’s lawsuit against California. In that case, xAI says its training datasets are “valuable trade secrets” 2.
xAI has also sued OpenAI, the company behind ChatGPT, over alleged trade secret theft. It says it follows “industry standard and industry leading” practices to guard confidential data, which can seem hard to square with requests for sensitive financial records from staff 3.

The hunt for narrow data exposes strain behind AI’s glossy image

Needing to gather tax documents this way suggests a wider AI problem. Good specialized data is hard to get, even for a company reported to be valued at about US$250 billion in the combined SpaceX-xAI valuation. SpaceX is Musk’s rocket and satellite company 4.
That method differs from xAI’s usual training approach, which the company says draws on public X posts and other internet data 1.
The reported payment delays also hint at a mismatch between xAI’s valuation and its day-to-day operations. They fit a broader tech risk where rush jobs can leave internal projects messy 5.