Connects technical AI-risk work with wider public influence

Center for AI Safety

The Center for AI Safety is a US nonprofit combining technical work on advanced-AI risks with public communication. Its connections include philanthropic grants, scientific and organizational leadership, and creator programmes. Its associated Action Fund is a separate organization with its own funding records.

CAIS works on reducing risks from advanced AI and combines research with public advocacy. Dan Hendrycks is a prominent figure in that work, and FLI’s biography of Jaan Tallinn identifies Tallinn’s CAIS board role. These relationships connect a technical research institution with donors and organizations elsewhere in the AI-risk field.

The historical resources include four Good Ventures awards totaling about $12.49 million between November 2022 and October 2023. The final award was listed as an exit grant. Silicon Valley Community Foundation’s 2023 return also records a $1,455,200 cash grant. These records explain a significant resource base during that period and give the organization a financial history extending beyond its most visible public statements.

CAIS also appears in Frame’s creator-training network. Frame names it as a participating organization and describes contracts involving creators Mikey Posada and Veronica Hylak. This is a practical route by which an organization working on technical risk can gain communication capacity: creators receive access, experience and opportunities while an institution can reach people beyond its specialist audience.

The Center for AI Safety Action Fund should be followed through its separate record. Tallinn’s donor ledger reports $3.221 million associated with that fund across five 2023–2024 entries. Keeping the two organizations legible helps readers understand which body is carrying out which work. The broader significance of CAIS remains the way technical research, philanthropic backing and public communication reinforce one another.

Assessment updated 2026-09-12

Detailed records and research

What the records show

Center for AI SafetyCAIS potential-refund reserve

$5,259,802 potential-refund reserve

Year ended 2023-12-31

2023 Schedule D Part X uses conditional reserve language; this entry is not a payment receipt. Financial details: value: 5,259,802; currency: USD; kind: potential refund reserve; period: Year ended 2023-12-31 Amount basis: carrying value

Center for AI SafetyCAIS 2024 liability close

Refund reserve no longer listed

Year ended 2024-12-31

2024 return carries $5,259,802 at opening. Closing other liabilities are $483,825, solely an operating lease liability in Schedule D. Cash settlement versus release/reclassification is unspecified.

Jaan TallinnCenter for AI Safety

Board member

Observed 2026-09-05

FLI official biography supplies CAIS governance bridge.

Dan HendrycksCenter for AI Safety

Executive / research director

Observed 2026-09-05

Current organization biography at cutoff.

Center for AI SafetyFTX debtors / bankruptcy estate

$5.2m resolution reported by debtors

2023-12-16 filing

Court-filed disclosure lists CAIS among completed donation-recovery resolutions. This is a party's filed account; exact agreement, payment mechanics and reserve reconciliation remain missing. Financial details: value: 5,200,000; currency: USD; kind: party-reported settlement resolution; period: 2023-12-16 filing Amount basis: reported settlement

Good VenturesCenter for AI Safety

$1,866,559 · Exit grant award

2023-10

Indexed first-party historical award record; not verified cash received or current sponsorship. Source identifies this as an exit grant. Financial details: value: 1,866,559; currency: USD; kind: exit grant award; period: 2023-10 Amount basis: indexed award

Further reading

Baseline evidence custody

Creator commission acquisition — 8 September 2026

Read the original sources 8

What the connections say

16 relationships

Read the wider story

AI & money

Who funds the AI debate?

Large grants, creator programmes and policy campaigns are visible. Their arrangements differ in how much control they give a sponsor.

Read the story