AI Side Hustles Misleading — the headline number is real but unrepresentativ

By the Income Reality Check Editorial Team · Updated 2026-05-22 · 7 min read

Anthropic’s fellows program: where the $15,000 a month really comes from

Verdict: Misleading — the headline number is real but unrepresentative. Anthropic does pay roughly $15,000 a month to its fellows, but the program is a competitive AI safety research job, not a side hustle for people who happen to like Claude.

A YouTube video from the channel All Thingz Real, titled “Anthropic Will Pay You $15,000 A Month To Use Claude Ai,” has drawn over 53,000 views by pitching what it calls a barely-known “fellows program.” The headline number checks out. The premise around it does not. This is a four-month empirical research fellowship that historically takes roughly one or two applicants out of every hundred, and the work has almost nothing to do with chatting with a chatbot.

What the video actually claims

The narrator opens the official Anthropic Fellows Program page on screen and reads through its top-line facts. He flags that it’s remote-friendly. He repeats the line about “$3,850 USD” base compensation alongside a 40-hour week and a four-month term, then quickly moves on. He says the five available streams — AI safety, AI security, ML systems and performance, reinforcement learning, and economics — can be looked up “in Chat GPT or Claude if you don’t know what that means.” The tone is encouraging: it’s worth giving a shot, the application is short, you can do it in a few minutes with a resume.

Three things get treated as throwaways. The video glances past the “fluent in Python” line in the general qualifications. It glances past the fact that each cohort lists its own deeper minimums. And it glances past the fact that the figure in the title — $15,000 a month — is never quite the figure on the screen. The stipend the creator reads aloud is $3,850, but he never explicitly says whether that’s weekly or monthly.

It’s weekly.

That distinction is the entire pitch. Multiplied across four weeks, $3,850 USD per week works out to roughly $15,400 a month, which is where the title’s number really comes from — not a casual access pass to Claude.

What the program actually is

According to the official program announcement on Anthropic’s alignment research blog, the Fellows Program funds engineers and researchers to investigate the company’s “highest priority AI safety research questions” alongside an Anthropic mentor (Anthropic Alignment, 2026). Fellows commit 40 hours a week for four months — May and July are the 2026 cohorts — and produce public outputs, usually a paper. The streams cover scalable oversight, adversarial robustness and AI control, model organisms, mechanistic interpretability, AI security, and model welfare.

The compensation breakdown is where the “$15,000” framing falls apart in two different directions. Anthropic publishes both figures on the same page:

So there are two $15,000-ish numbers floating around — the monthly compute budget Anthropic provides for running experiments, and the de-facto monthly value of the weekly stipend. The video conflates them into a single income claim. The video also doesn’t mention the official Greenhouse job posting, which states fellows must have work authorization in the U.S., U.K., or Canada, and that visa sponsorship is not offered (Anthropic Greenhouse, 2026). For viewers in India, Nigeria, the Philippines, or most of the EU, the program isn’t open to them at all unless they already hold the right paperwork.

Who actually gets in?

This is the part the video skips entirely. AI safety researchers tracking the field publicly estimate the Anthropic Fellows Program’s acceptance rate at well under 2% — one widely-shared community write-up puts a recent cohort at 32 fellows selected from more than 2,000 applicants, working out to roughly 1.3% (LessWrong, 2025). That’s tighter than admission to most Ivy League graduate programs.

The median applicant, per the same write-up, is “near the end of their Masters degree or doing a PhD, having completed a few projects and internships with shown AI safety interest.” Anthropic’s own page softens that — “we care much more about your ability to execute on research than your credentials” — but the company is itself a generational hiring magnet, having closed a $30 billion funding round at a roughly $380 billion valuation in early 2026 (CNBC, 2026). The applicant pool reflects that. A separate U.S. labor data note from the Bureau of Labor Statistics estimates AI-related occupations are among the fastest-growing in the projections, which is what’s drawing thousands of candidates to programs like this in the first place (BLS, 2025).

What does “fluent in Python” really mean here? Read between the lines of the Greenhouse posting and it means you can implement empirical ML research ideas — load a transformer, train a probe, run an evaluation, iterate on a hypothesis — without supervision. It is not the kind of Python you pick up in a weekend bootcamp. The cohort you’d be sitting in is competing against people who’ve already published papers, contributed to open-source ML libraries, or worked at frontier labs.

What you’d realistically earn

Let’s split this honestly. If you are in the small slice of the audience that could plausibly get in — say, a CS or physics graduate with research experience, located in the U.S., U.K., or Canada — then yes, the gross stipend works out to roughly $15,400 a month for four months, totaling around $61,600 USD before tax. That is real money, and it’s competitive with entry-level industry research engineer compensation, though well below the $200K-plus total comp that senior ML engineers can earn at frontier labs once you fold in equity.

If you are not in that slice — and the video clearly addresses a general audience, not active AI researchers — the realistic number is zero. Not zero because the program is a scam (it isn’t), but zero because applying without the underlying research aptitude is functionally a lottery ticket with single-digit-percent odds. The video implies the application is fast, which is true; what it does not say is that “fast application” plus “1-2% acceptance” plus “no visa sponsorship” plus “four-month commitment” is a very different proposition from a side hustle. Spending an evening on the form is reasonable. Budgeting around the income is not.

And the income, if you do get in, is wages, not passive earnings. You owe income tax on it. Fellows in the U.S. would receive a 1099 or W-2 and pay regular IRS rates; U.K. and Canadian fellows would settle with HMRC and the CRA respectively. None of that gets mentioned on screen.

Who this is (and isn’t) for

This is for someone with a strong technical background — ideally a CS, math, or physics degree, with self-directed Python research projects already in hand — who already wants to work at a frontier AI lab full-time, holds work authorization in the U.S., U.K., or Canada, and can commit four months at 40 hours a week to a single research stream. Roughly a quarter to a half of past fellows have converted to full-time Anthropic roles, which is genuinely the prize. This isn’t for someone discovering Claude this week and looking for a way to monetize a chatbot subscription.

A note on income framing

The U.S. Federal Trade Commission has been openly drafting a new Earnings Claim Rule aimed at money-making opportunity sellers and multi-level marketers who imply specific income figures the typical participant won’t actually earn (FTC, 2025). The Anthropic program is none of those things — it’s a real research fellowship — but the YouTube framing around it is the exact pattern regulators have been flagging in adjacent contexts: a real number, pulled out of context, presented as accessible to a much wider audience than will ever realistically capture it.

This isn’t unique to this channel. We’ve covered similar reframings in the $20,000-a-month Claude Code YouTube pitch, and we walked through how the different Claude tiers actually work in our breakdown of every level of Claude.

What to remember

The Anthropic Fellows Program is a real, well-funded research fellowship that pays roughly $15,400 a month in stipend for four months, runs in the U.S., U.K., or Canada, and admits roughly one in 75 applicants. The video gets the dollar number right. It gets almost everything else — the role, the audience, the workload, the geographic reach — quietly wrong.