Refolk
October 19, 2026·9 min read

Character.AI's Chapter 11 Docket: Up to $920K a Head, and Who's Left

Character.AI's Chapter 11 docket prices Disney's 66 hires at up to $920K a head. How to use it as a comp benchmark and a map of who is still hireable.

Character.AI bankruptcy employeesDisney Character.AI hiringAI engineer compensation 2026hiring from bankrupt startupsChapter 11 filing talent sourcing
Character.AI's Chapter 11 Docket: Up to $920K a Head, and Who's Left

If you're pricing an AI req or trying to source from Character.AI, the most useful document available right now is a Delaware bankruptcy filing. In it, Disney estimates about $60M a year in run-rate compensation for roughly 65 offered employees, which is about $920K a head, and adds a $30M retention pool on top. Of roughly 86 full-time staff, 66 accepted. For recruiters, the filing works like a public offer letter, and it tells you more about who's still hireable than any profile search will.

What did Disney actually pay for Character.AI's team?

Disney estimates about $60M a year in run-rate compensation for roughly 65 offered employees. That works out to about $923K per head, plus a $30M retention pool and $20M for a technology license. Disney calls its review preliminary, so read the per-head figure as a ceiling.

Here's what the term sheet lays out:

  • People: offers to about 65 staff (66 accepted) at a ~$60M annual run-rate.
  • Retention: a $30M pool, paid out in the first and second years after close.
  • Technology: $20M for the post-training stack, training recipes, reward model, benchmarks, evaluation framework, and nine patents.
  • Excluded: the consumer app, user data, and finished models.

The reason it got this far is simple arithmetic. Character.AI was losing roughly $8M to $9M a month while holding about 12 million users. Its sale banker, Houlihan Lokey, contacted 28 potential buyers and got back one offer it couldn't use. Disney's bid is a talent bid with a license attached. Karandeep Anand, Character.AI's former CEO, is now Disney's first chief technology officer, and Disney is making these hires right after laying off hundreds of its own people in late September.

$923K
Estimated annual comp per Character.AI hire in Disney's term sheet
Derived from Disney's preliminary $60M run-rate estimate across about 65 offered employees.

How does $920K compare with AI engineer compensation in 2026?

Disney's per-head figure comes in about 16% above the OpenAI median software engineer and about 6x the US AI engineer median on Levels.fyi data. That makes it a frontier-lab price. It doesn't describe the broader market.

Benchmark (US total comp)Annual TCDisney per-head as multiple
Disney / Character.AI run-rate per head~$923K1.0x
OpenAI median SWE~$795K1.16x
Meta AI Engineer median$448K2.1x
US ML Engineer median$279K3.3x
US AI Engineer median$154K6.0x

Disney figure derived from the term sheet. Others from Levels.fyi or write-ups of Levels.fyi data. Multiples derived.

Disney is paying above OpenAI because it's buying a post-training and serving team that already ran a product with about 12 million users, intact, rather than assembling one hire at a time. That's a team price. It isn't an individual offer.

The practical rule: use $920K to anchor comp for post-training and RL specialists with production scale behind them. Don't use it for a generic "AI engineer" req. Hiring managers who quote this number for a fine-tuning generalist will price themselves out of the $154K to $279K band where most of the market actually sits.

Does the $180M three-year figure hold up?

Not cleanly. Several outlets put Disney's three-year cost at about $180M and say that figure includes the $30M retention pool. But $60M times three is already $180M before any retention money.

One of two things has to be true:

  1. Annual comp is closer to $50M, or about $760K per head, and the $30M pool fills the gap to $180M.
  2. The $60M is a rough preliminary estimate, and the "includes retention" wording is loose.

Either way, the honest range is $760K to $923K per head. If you use this filing in a comp conversation, quote the range and say where it came from. A hiring manager who later finds out the $923K was a ceiling will stop trusting your other numbers too.

Who from Character.AI didn't go to Disney?

About 20 of the roughly 86 full-time employees didn't accept Disney offers. They aren't one group, and they aren't equally reachable.

Headcount from the filings and reporting:

  • ~86 full-time staff on September 30
  • 66 accepted Disney offers
  • 56 of those had formally resigned by the petition date
  • 10 accepted but are waiting on US work visa updates
  • 18 remain on Character.AI's payroll
  • 8 are expected to stay through the bankruptcy case
  • ~10 on payroll aren't expected to stay through the case (derived)
  • ~2 left entirely (derived)

The 8 who stay are most likely keeping the app running. Disney staff support it under a transition agreement until February 2027. The roughly 10 leaving payroll are the most important group for recruiters. Disney's $30M retention pool doesn't cover them, so nothing financial ties them in place.

The legal scope matters too. Character.AI agreed not to enforce non-compete, non-solicit, confidentiality, or invention-assignment covenants against offered employees, but only where those covenants would stop someone from working for Disney. It also released Disney hires (including the former CEO) from claims like breach of fiduciary duty, and agreed not to hire them back for three years. None of that applies to the people who didn't go. Assume their normal confidentiality and IP obligations still hold, and don't ask them about anything those obligations cover.

Why are public profiles behind the docket?

Profiles lag because people don't update their employer while a deal is still open. In Refolk's index of professional profiles, 71 US profiles mention Character.AI in their headline. In a 25-profile sample, 16 still list Character.ai as their current employer and only 2 list Disney, even though the court papers say 56 people had resigned. The sample is small, so treat it as directional.

There are three reasons for the gap:

  • Visas. Ten accepted hires are waiting on US work visa updates. Changing your employer publicly before that's sorted is a risk with no upside.
  • The sale hasn't closed. Disney's outside closing date is December 17, 2026.
  • The auction. A rival bidder could still win, so a profile edit made today might be wrong next month.

This means a "current company = Disney" filter will miss most of the team. A "current company = Character.AI" filter mixes the 66 who left with the 20 who didn't. The docket tells you who is in which group before any profile does.

The fix is to search on what people did, not on the employer field. That's the gap Refolk is built for. You describe the person in plain English, for example "worked on post-training or serving at Character.AI", and it matches on the work rather than the job title, so a stale employer field doesn't hide anyone.

How do you read a Chapter 11 docket for talent sourcing?

Read six parts of the docket: the term sheet, the employee and retention exhibits, the covenant and release language, the bid procedures, the creditor matrix, and the asset description. Between them you get headcount, comp, retention timing, how far the covenant waivers reach, and every deadline. This one is Case No. 26-11603 (BLS) in the District of Delaware.

What each part tells you:

  1. Term sheet. The buyer's estimated comp run-rate and the number of offers. Here that's ~$60M and ~65 people. Divide one by the other and you have a per-head benchmark.
  2. Employee and retention exhibits. Pool size and vesting. A $30M pool across 66 people averages about $455K each, paid in years one and two. Those payout dates are when people become movable again.
  3. Covenant and release language. Who is released from what, and in whose favor. Here the waivers only cover joining Disney, which tells you which group carries fewer legal strings.
  4. Bid procedures. Bid deadline, auction date, sale hearing, and any clauses that let a rival bidder take over staff.
  5. Creditor matrix. Who's owed money and how much. Google LLC is the largest listed unsecured creditor at about $4.1M.
  6. Asset description. What the buyer pays for and what it leaves behind. Code, models, and data that stay behind need people to run them.

Use the docket alongside profile data, not instead of it. The court papers tell you who moved and on what terms. The profiles tell you what each person actually built.

When can a Disney hire still consider another offer?

An accepted offer isn't final until the sale hearing on November 24. People become movable again when the retention payouts land in the first and second years after close. Those are the two windows that matter.

The bid procedures set the near-term calendar:

  • Nov 20, 5:00 p.m. ET: bids due
  • Nov 23: auction if needed, in $500K increments
  • Nov 24: sale hearing
  • Dec 17, 2026: Disney's outside closing date

The clause that changes things: if Disney loses the auction, the winning bidder can interview up to 80% of the staff Disney took on and make them offers. So until the sale hearing, 66 accepted offers are effectively open again. Don't try to disrupt a court-supervised sale. But if a candidate brings up a competing offer during this period, it's a legitimate conversation.

The longer window is the retention schedule. With about $455K per head on average paid out over two years, expect the first wave of movement once the year-one payout lands. If the deal closes by the outside date, that's late 2027. Put it in your calendar now.

Why do buyers pay for the people, not the code?

In distressed AI deals, the team is the asset. Disney's staff cost (about $180M over three years) is roughly 9x the $20M license, and Google's 2024 deal priced Character.AI the same way.

In 2024, Google brought back Noam Shazeer, Daniel De Freitas, and roughly 20 to 30 core researchers in a deal worth about $2.7B. Shazeer has since announced he's leaving Google for OpenAI. Acqui-hired talent doesn't necessarily stay put once the deal that brought it in has played out.

When the employee-cost exhibit dwarfs the asset bid, the filing is telling you exactly who the buyer wants.

That ratio is worth checking in every distressed AI filing. If the people cost several times what the IP does, the buyer is hiring a team, and that team will sit behind retention cliffs you can read in the exhibits.

FAQ

How many Character.AI employees joined Disney?

66 of roughly 86 full-time employees accepted Disney offers. 56 had formally resigned by the petition date, and the other 10 are waiting on US work visa updates. Character.AI kept 18 on payroll, but only 8 are expected to stay through the bankruptcy case. About 20 people did not go to Disney.

Is $920K a realistic AI engineer compensation benchmark for 2026?

Only for a narrow slice of the market. It's a preliminary team average for buying a proven post-training team intact, and if the $180M three-year figure includes the retention pool, the real number may be closer to $760K. The US AI engineer median on Levels.fyi data is about $154K, and the ML engineer median is about $279K. Use the Disney figure for frontier post-training specialists only.

Can I recruit Character.AI employees who accepted Disney offers?

They're free to consider other offers, and the auction clause means a rival winning bidder could interview up to 80% of them before the November 24 sale hearing. After that, retention payouts in years one and two give them a reason to stay at Disney until the money lands. The people with nothing holding them are the roughly 12 non-Disney staff who are leaving or have already left.

Where can I read the Character.AI bankruptcy filing?

The case is No. 26-11603 (BLS) in the US Bankruptcy Court for the District of Delaware. Start with the term sheet, the bid procedures order, and the creditor list. Most of the useful sourcing information is in the employee and retention exhibits.

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.

  1. 01Describe them

    One plain sentence. Role, city, stack, stage, whatever matters to you.

  2. 02I read the web live

    GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.

  3. 03You read the shortlist

    Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.

  • 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.

Read next