Anthropic's S-1 Shows 1% Attrition. Pitch the 940 Before Day 27
Anthropic's S-1 shows 940 pre-2025 staff, 1% attrition and 20+ lockup windows. Here is who to pitch, when to reach them, and how to find them.
If you're trying to hire Anthropic engineers, the S-1 works as a target list with a calendar attached. Headcount went from about 940 to about 5,300 in under two years, some insiders can sell on day one, and more than 20 lockup windows open over the following year. The same filing also shows regrettable attrition of about 1%, so very few people are actually looking to leave. The real questions are who to pitch, when, and with what offer.
What does Anthropic's S-1 say about its employees?
The S-1 reports about 5,300 employees as of Sept 30, 2026, up from about 940 on Dec 31, 2024. That means roughly 82% of current staff joined in 2025 or later. The filing also warns that early employees hold equity that "may represent a significant portion" of their wealth, which could make it hard to "retain and motivate" them.
Anthropic calls its staff "Ants." Trailing-12-month revenue per employee is over $6.6M, up from $1.4M a year earlier. Headcount grew about 170% while revenue grew about 1,200%. For most of that growth, the team was far smaller than it is today.
| Segment | Figure | Source |
|---|---|---|
| Headcount, Dec 31, 2024 | ~940 | S-1 |
| Headcount, Sept 30, 2026 | ~5,300 | S-1 |
| Joined since 2025 (derived) | ~4,360 (~82%) | 5,300 minus 940 |
| US profiles with "Anthropic" in headline | 4,384 | Refolk's index |
| UK profiles with "Anthropic" in headline | 568 | Refolk's index |
| Technical-titled Anthropic profiles, worldwide | 1,059 | Refolk's index |
Treat the index counts as upper bounds. A headline keyword match also picks up alumni founders, people at "Ode with Anthropic," and unrelated companies with "Anthropos" in the name. In a sample of 25 US profiles, only 18 listed Anthropic as their current employer.
Why pitch the 940 instead of the 4,360 newer hires?
The pre-2025 cohort is mostly or fully vested and about to be liquid, so leaving costs them little and the lockup calendar tells you when they'll be thinking about what comes next. No other group at Anthropic comes with a published timetable like that.
The cliff story is backwards for 2026 hires
A vesting cliff is the waiting period before any equity vests. Leave earlier and you get nothing. It's easy to assume a standard 12-month cliff. The S-1's financial notes say otherwise. Options granted from January 2026 vest 12.5% after six months, then monthly. Before that, 25% vested after one year. (RSU terms weren't in the excerpts I reviewed.)
So a 2026 hire clears the cliff in six months and vests monthly after that. Not much cliff is holding them in place, which means the usual advice to wait until someone passes their one-year mark doesn't apply. The newer hires don't have a cliff date to time your message around. The early ones do have a date to work with, and it's the lockup schedule.
The 940 have the opposite profile
Here's why the early cohort is the one worth pitching:
- Little left to forfeit. Most of their equity is vested, so leaving costs them relatively little.
- About to be liquid. The lockup windows turn paper wealth into money they can use, and that money lets them take a risk.
- Money alone won't move them. If their stakes are anywhere near as large as estimates suggest, a bigger offer changes little, so the conversation has to be about the work.
- Some have already left. Behnam Neyshabur and Harsh Mehta joined in late 2024, so they belong to this cohort. They left in December 2025 to found Mirendil, about 13 months later.
How do you find the pre-2025 cohort on LinkedIn, GitHub and the open web?
Filter on tenure, not title. On LinkedIn that means a start date in 2024 or earlier. On GitHub it means contributions to anthropics/* repositories before 2025. Titles won't help, because nearly everyone has the same one.
In Refolk's index, 1,059 profiles worldwide combine "Anthropic" with a Member of Technical Staff, Research Engineer or Research Scientist title. In the sample, 24 of 25 were titled "Member of Technical Staff." Member of Technical Staff (MTS) is the flat, catch-all title Anthropic gives engineers and researchers whatever their seniority. A search for "Senior" or "Staff" in the title finds almost nobody. A 2023 hire and someone who started in March look identical on paper.
Here's how to separate them:
- LinkedIn start dates. Current employer Anthropic, start date 2024 or earlier. Drop anyone whose "Anthropic" match comes from a different company name or an alumni headline.
- GitHub history. Look for commits, reviews or issues on anthropics/* repositories dated before January 2025.
- Open-web footprint. Look for paper author lists, model and system documentation, and conference talks from before 2025. These show who did the work, which LinkedIn often doesn't.
- Cross-reference all three. A person with a 2024 start date, pre-2025 repo activity and an author credit is almost certainly in the 940.
The fourth step is the slow part. Matching one LinkedIn profile to one GitHub handle to one author name, one person at a time, is where the hours go. Refolk does that matching for you. You describe the cohort in plain English and get back profiles already linked across GitHub, LinkedIn and the open web.
San Francisco is by far the top region in the index for Anthropic profiles, so start there. Then run the same search for the UK, where the count of 568 is about 13% of the US figure.
What pitch actually moves an early Anthropic employee?
A specific research problem they would own. The S-1 lists "shifts in individual beliefs or alignment with other organizations" among the factors that could hurt retention, so the filing itself names belief and alignment as a reason people leave.
Run the money math and the comp pitch falls apart. A third-party estimate (not from the filing) says that at a $2T valuation the roughly 940 pre-2025 hires would hold nine figures each on average. Even if that's off by an order of magnitude, a bigger base salary or a richer grant barely registers for this group. What does register is a chance to own something they can't own inside a 5,300-person company.
The ex-Anthropic founders show what that looks like:
- Mirendil: Behnam Neyshabur and Harsh Mehta raised a $200M seed at about a $1B valuation. Some outlets reported $10B, which is wrong.
- Humans&: co-founded by Andi Peng, who worked on Claude 3.5 to 4.5 post-training. It raised a $480M seed.
- Ricursive Intelligence: Anna Goldie and Azalia Mirhoseini (ex-Anthropic and ex-DeepMind) raised $335M and rehired former collaborators.
Someone sitting on a vested stake that large doesn't leave for a raise. They leave for a problem with their name on it.
So put three things in the first message: the exact problem, the ownership (what they decide and what they ship), and room for the two or three people they'd want to bring with them. Leave out the comp band. If the problem is good enough, comp comes up on the second call.
You're also competing for the same title across the market, and seeing who else is hiring for it tells you what candidates are comparing your offer against.
How many Anthropic employees will realistically leave?
Very few. At about 1% regrettable attrition in the 12 months to Sept 30, you'd expect roughly 53 regretted departures a year across 5,300 people, and 19% of regretted leavers have come back.
The retention numbers at the top are even tighter. All seven co-founders are still there. More than three-quarters of the first 100 employees are still there. Anthropic is also still pulling people in: Andrej Karpathy and Ross Nordeen joined in 2026. That pull works against you.
People do leave, though. John Schulman joined Anthropic, stayed briefly, and moved to Thinking Machines Lab. That's the honest reading of the base rate: departures happen, but they're rare, specific, and usually about the work.
Three things follow:
- Cold outreach to the whole company is spam math. If 1% leave regretfully, messaging 5,300 people mostly annoys 5,247 of them.
- A hire can boomerang. With a 19% return rate, plan your onboarding to make the problem stick, not just the offer.
- The realistic win is a small team. Mirendil and Humans& both formed when a few people left together. Pitch the pair or trio who already work together.
That leaves founders with an awkward gap. With roughly 53 regretted exits a year across the whole company, the 940 are a long shot that can take months to land, and the research or infra seat you need filled is open today. You need a second pipeline running while the long pitch plays out.
When should you reach out around the lockup windows?
Reach out before a window opens, not after. Some insiders can trade on day one, more windows open at days 27, 48, 62 and 76, there are more than 20 over roughly a year, and a second wave follows Anthropic's full FY2026 results in early 2027.
Why before? In the weeks ahead of an unlock, people are adding up what the stake is worth and asking what they'd do if money stopped being a constraint. A message that arrives while they're thinking that through becomes part of the plan. A message that arrives after the sale lands in the same inbox as every other recruiter who read the calendar.
The listing is reportedly targeted for the week of Nov 16. For illustration only, assume first trade on Nov 17:
| Lockup window | Illustrative date | Outreach move |
|---|---|---|
| Day 1 | Nov 17 | Map the 940, no outreach yet |
| Day 27 | ~Dec 14 | First messages go out the week before |
| Day 48 | ~Jan 4 | Follow up after the holidays |
| Day 62 | ~Jan 18 | Pitch teams, not individuals |
| Day 76 | ~Feb 1 | Close conversations already open |
| FY2026 results | Early 2027 | Second wave, re-run the search |
The windows are staggered, so not everyone unlocks at once, and from the outside you can't tell who is in which group. Watch for signals instead. These include headline changes to "ex-Anthropic" or "building something new," GitHub activity moving to new organizations, and co-authors suddenly showing up at the same new company. Re-running your cohort search before each window catches those changes without checking 1,059 profiles by hand.
FAQ
When can Anthropic IPO lockup employees sell their shares?
Some insiders can trade on day one of public trading. More windows open at days 27, 48, 62 and 76, and there are more than 20 windows over about a year. Another set of lockups starts expiring after Anthropic publishes its full FY2026 results, which puts a second wave in early 2027. You can't tell from the outside which employees fall into which window, so treat each date as a reason to check in, not a guarantee that a specific person can sell.
Is Anthropic's vesting cliff still 12 months?
Not for recent options. The S-1's financial notes say options granted from January 2026 vest 12.5% after six months and then monthly. Earlier grants vested 25% after one year. So 2026 hires clear their cliff quickly and have little cliff holding them in place. The pre-2025 cohort is mostly vested already, which is why the lockup calendar, not the cliff, is the date that matters for them.
Should I cold-message every Anthropic engineer after the IPO?
No. Regrettable attrition was about 1% in the 12 months to Sept 30, which works out to roughly 53 regretted departures a year across 5,300 people. A blanket campaign mostly annoys people who aren't leaving and burns your name with the ones who might. Target the pre-2025 cohort, prefer small teams that already work together, and lead with a specific problem.
How do I tell an early Anthropic employee from a new one when the titles match?
Use tenure signals. In Refolk's index, 24 of 25 sampled technical profiles carry the same "Member of Technical Staff" title, so titles tell you nothing. Check for a LinkedIn start date of 2024 or earlier, contributions to anthropics/* GitHub repositories before 2025, and author credits on pre-2025 papers or documentation. Someone who shows all three is almost certainly in the original 940.
The data behind this
Try it on the search you came here for
Stop building boolean strings. Just describe the person.
Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.
01Describe them
One plain sentence. Role, city, stack, stage, whatever matters to you.
02I read the web live
GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.
03You read the shortlist
Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.
- Staff backend engineers in NYC who shipped Rust in production
- Series A fintechs in SF under 50 people, growing headcount this year
- Maintainers of fast-growing Rust web frameworks on GitHub
- No boolean, no filters, no seat to buy. One box.
- Read at search time, so a profile updated yesterday counts today.
- Every step visible as it runs, every name with its reason.
500 free credits on sign-up. No card, no demo call. See real searches.